Predicting Sustainable Crop Yields: Deep Learning and Explainable AI Tools

[Find the meaning and references behind the names: Vadim, Jamila, Moscow, Doi, Vladimir, June, Rice, Bauman, India, Northern, Basel, Local, Unique, Mustafina, Manoj, Gas, Hot, Ivan, Dhiya, October, Andrei, September, Lime, Tools, Root, Obe, Data, Deep, Under, Heart, Tuning, Winter, High, Far, Year, Era, State, Jayabalan, Open, Aleksei, November, Rot, Fall, Evolution, Role, Russia, Season, Rabi, Strong, Fine, Tel, March, Beans, Boost]

Citation: Malashin, I.; Tynchenko, V.; Gantimurov, A.; Nelyub, V.; Borodulin, A.; Tynchenko, Y Predicting Sustainable Crop Yields: Deep Learning and Explainable AI Tools Sustainability 2024 , 16 , 9437 https://doi.org/10.3390/su 16219437 Academic Editors: Dhiya Al-Jumeily OBE, Jamila Mustafina and Manoj Jayabalan Received: 27 September 2024 Revised: 26 October 2024 Accepted: 29 October 2024 Published: 30 October 2024 Copyright: © 2024 by the authors Licensee MDPI, Basel, Switzerland This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/) sustainability Article Predicting Sustainable Crop Yields: Deep Learning and Explainable AI Tools Ivan Malashin 1, * , Vadim Tynchenko 1, * , Andrei Gantimurov 1 , Vladimir Nelyub 1,2 , Aleksei Borodulin 1 and Yadviga Tynchenko 1,3 1 Artificial Intelligence Technology Scientific and Education Center, Bauman Moscow State Technical University, 105005 Moscow, Russia 2 Scientific Department, Far Eastern Federal University, 690922 Vladivostok, Russia 3 Laboratory of Biofuel Compositions, Siberian Federal University, 660041 Krasnoyarsk, Russia * Correspondence: ivan.p.malashin@gmail.com (I.M.); vadimond@mail.ru (V.T.); Tel.: +7-926-875-7128 (I.M.) Abstract: Optimizing agricultural productivity and promoting sustainability necessitates accurate predictions of crop yields to ensure food security. Various agricultural and climatic variables are included in the analysis, encompassing crop type, year, season, and the specific climatic conditions of the Indian state during the crop’s growing season. Features such as crop and season were one-hot encoded. The primary objective was to predict yield using a deep neural network (DNN), with hyperparameters optimized through genetic algorithms (GAs) to maximize the R 2 score. The bestperforming model, achieved by fine-tuning its hyperparameters, achieved an R 2 of 0.92, meaning it explains 92% of the variation in crop yields, indicating high predictive accuracy. The optimized DNN models were further analyzed using explainable AI (XAI) techniques, specifically local interpretable model-agnostic explanations (LIME), to elucidate feature importance and enhance model interpretability. The analysis underscored the significant role of features such as crops, leading to the incorporation of an additional dataset to classify the most optimal crops based on more detailed soil and climate data. This classification task was also executed using a GA-optimized DNN, aiming to maximize accuracy. The results demonstrate the effectiveness of this approach in predicting crop yields and classifying optimal crops Keywords: sustainable agriculture; yield optimization; machine learning; explainable AI 1. Introduction In recent years, agriculture has experienced a transformative evolution driven by technological advancements, heralding the era of precision agriculture [ 1 ]. This paradigm shift focuses on integrating technologies such as sensor systems [ 2 ], artificial intelligence (AI) [ 3 ], and machine learning (ML) [ 4 ] to optimize farming practices and boost agricultural productivity while minimizing environmental impact [ 5 ]. At the heart of this evolution is the capacity to leverage extensive datasets that encompass soil properties, climate variables, and agronomic factors, and their effects on crop growth, yield, and overall agricultural sustainability 1.1. Seasonal Crop Cultivation in India In India, a wide variety of crops are cultivated across different seasons, each with unique requirements and challenges. Rice is a staple crop grown during the Kharif (monsoon) season, typically from June to November [ 6 ]. It requires abundant water [ 7 ], making it highly vulnerable to droughts or erratic rainfall. Maize is adaptable and grown in both Kharif and Rabi (winter) seasons [ 8 ], yet faces threats from pests like fall armyworm and temperature fluctuations. The chickpea, a winter crop [ 9 ], is sown from October to November and harvested in March. It is drought-resistant but sensitive to diseases like root rot [ 10 ], especially in drier regions. Kidney beans, cultivated mainly in northern India [ 11 ], thrive Sustainability 2024 , 16 , 9437. https://doi.org/10.3390/su 16219437 https://www.mdpi.com/journal/sustainability

[Find the meaning and references behind the names: Liu, Forest, Map, Change, Khanal, Cool, January, Molly, Punjab, Storm, Farm, Mung, Round, Low, Caren, Rust, Set, Borer, Fruit, Show, Field, Matter, Corn, Pod, February, Vector, Areas, Dry, Gram, Papaya, Coffee, Pigeon, July, Moth, Summer, Pose, Frost, Combat, Due, Mlr, Free, Black, Cotton, Quality, Rising, Seven, Short]

Sustainability 2024 , 16 , 9437 2 of 29 during the Kharif season in cool, frost-free conditions, yet they face challenges (such as root diseases) and depend on well-distributed rainfall Pigeon peas are drought-tolerant legumes grown in the Kharif season but are vulnerable to pod borer infestations and uneven rainfall [ 12 ]. Similarly, moth beans and mung beans are short-duration legumes grown in dry areas during the Kharif season [ 13 ]; they are resilient but suffer from pod-shattering [ 14 ] and disease risks, particularly in low-water conditions. Black gram and lentils are primarily Rabi crops [ 15 ], requiring minimal water, but are sensitive to moisture stress and diseases like rust and wilt Pomegranates, cultivated year-round, are drought-tolerant yet require careful irrigation and disease management to combat fungal blight [ 16 ]. Bananas, grown in tropical regions throughout the year, face fungal threats, high water demand, and storm damage [ 17 ]. Mangoes, a summer fruit [ 18 ], are harvested from February to June but can be impacted by extreme temperatures and erratic rainfall, which affect flowering [ 19 ]. Grapes, grown from January to May, are highly sensitive to fungal infections, requiring precise irrigation and temperature control Watermelon [ 20 ] and muskmelon [ 21 ] are summer crops, often grown under high temperatures and sensitive to soil salinity and pests like aphids. Apples are temperate crops grown in the Himalayas [ 22 ] from June to September, but changing climate conditions, such as rising temperatures and irregular snowfall, challenge their yields. Oranges, grown mainly from October to March, require dry conditions but are sensitive to frost and water stress [ 23 ]. Papayas, cultivated year-round in tropical areas [ 24 ], require well-drained soil and consistent irrigation but are threatened by the papaya ringspot virus [ 25 ]. Coconuts are grown in coastal areas throughout the year and depend on regular rainfall [ 26 ], although drought and pests like the rhinoceros beetle pose risks. Cotton is a Kharif crop sown from June to September [ 27 ]; it requires warm, dry conditions but faces pest challenges from bollworms and is highly rainfall-dependent. Jute is a monsoon crop [ 28 ], grown from March to July, thriving in waterlogged soil but vulnerable to erratic rainfall affecting fiber quality [ 29 ]. Lastly, coffee [ 30 ], grown in high-altitude regions [ 31 ] and harvested from November to March, is sensitive to unpredictable rainfall, pests, and temperature variations driven by climate change Each of these crops represents a distinct set of agricultural challenges in India, including water dependence, vulnerability to pests and diseases, and sensitivity to fluctuating climate conditions, all of which are critical factors in achieving sustainable yields 1.2. Literature Outlook The integration of ML and remote sensing in soil and crop yield prediction is actively highlighted in the scientific literature. For instance, Khanal [ 32 ] uses aerial imagery [ 33 ] and ML to predict soil properties and corn yield at Molly Caren Farm in Ohio [ 34 ]. By analyzing multispectral images [ 35 ] and field data from seven plots in 2013, the research compares ML models like random forest and neural network models. Findings show these models outperform traditional methods, with neural networks excelling in predicting soil organic matter and cation exchange capacity, and Random Forest performing best for corn yield prediction. This approach demonstrates the potential of remote sensing and ML for accurate mapping, enhancing agricultural management tailored to local conditions Liu et al. [ 36 ] utilized Landsat and MODIS-NDVI data along with climatic, topographic data, and soil samples to map seven soil properties (texture, electrical conductivity, pH, nitrogen, phosphorus, potassium, and organic matter) in Punjab, Pakistan from 2000 to 2020. Comparing three statistical models—support vector machine (SVM), random forest regression [ 37 ] (RFR), and multiple linear regression (MLR)—RFR generally provided the highest accuracy. This highlights the effectiveness of ML over MLR in handling nonlinear relationships, revealing a decline in cultivated areas and high soil electrical conductivity due to salinity, with organic matter and nitrogen levels generally low.

[Find the meaning and references behind the names: Sri Lanka, Trees, Sri, Real, Lanka, Diaz, Bayes, Burdett, Yadav, Key, Max, Apple, Fields, Ann, Loss, Power, Leaf, Land, Six, Speed, Multi, Ability, Mlp, Agro, Architecture, Spike, Cross, Wheat, Sidhu, Tang, Wickramasinghe, Study, Wind, Min, End, Canada, Rahman, Yala, Scales, Focus, Sunshine]

Sustainability 2024 , 16 , 9437 3 of 29 Different types of soil support various crop growth based on their unique characteristics. Understanding these soil features is essential for optimal crop selection [ 38 ]. Rahman et al. [ 39 ] introduced a model that predicts soil series based on land type and recommends suitable crops accordingly, employing ML algorithms including weighted k-nearest neighbor (kNN) [ 40 ], bagged trees [ 41 ], and Gaussian kernel-based support vector machines (SVMs) [ 42 ] for soil classification. Experimental results demonstrate that the SVM approach outperforms existing methods, highlighting its effectiveness in agricultural decision-making Yadav et al. [ 43 ] developed a model for assessing soil fertility, recommending suitable crops, and predicting crop yield based on soil features. Various ML algorithms including SVM, random forest, Naive Bayes [ 44 ], Linear Regression, Multilayer Perceptron [ 45 ] (MLP), and ANN [ 46 , 47 ] were employed for soil classification and yield prediction Results indicate that the ANN approach, utilizing deep learning (DL) architectures for enhanced accuracy, outperforms traditional methods in predicting crop yields based on soil characteristics Wickramasinghe et al. [ 48 ] explored the relationship between rice yield and climate variables in a key region of Sri Lanka [ 49 ] using various statistical and ML methods. They analyzed factors such as rainfall, temperature (min/max), evaporation, wind speed, and sunshine hours, leveraging three decades of rice yield and monthly climate data. Models developed included ANN, SVM regression [ 50 ] (SVMR), multiple linear regression [ 51 ] (MLR), Gaussian process regression [ 52 ] (GPR), power regression [ 53 ] (PR), and robust regression (RR). GPR outperformed others in yield prediction accuracy, validated with data from the 2019 Yala season Enhancing agricultural management through AI and ML was addressed by Diaz et al. [ 54 ], which highlighted the challenge of declining soil quality in intensive agriculture by exploring how AI and ML can estimate soil quality indicators [ 55 ] (SQIs) from agro-industrial data. The focus was on recent studies using remote sensing to predict crop yields at regional and local scales, evaluating different spectral bands, data preprocessing methods, and ML algorithms. The review proposed a model integrating SQI, environmental factors, and crop management data to enhance agricultural practices and optimize crop yield predictions through ML insights Burdett et al. [ 56 ] investigated the relationship between crop yield, soil properties, and topographic characteristics using high-resolution data in Southwestern Ontario, Canada Analyzing a dataset of 145,500 observations on corn and soybean yields [ 57 ], soil nutrients, and topographic features, this study compared multiple analytical methods including multiple linear regression, ANN, decision trees, and random forests. Random forest emerged as the most effective method, achieving R 2 values of 0.85 for corn and 0.94 for soybeans, outperforming other techniques such as MLR. Cross-validation experiments demonstrated the ability of random forest models to predict yield variations in fields not included in the training dataset, indicating their potential in precision agriculture for identifying high-yield areas based on soil and topographic attributes. Another solution for precision agriculture is presented in [ 58 ], which introduces WH-DETR, a high-precision, end-to-end wheat spike detection network built on an enhanced RT-DETR architecture. This network achieves an impressive 95.7% Average Precision by utilizing multi-scale feature extraction, optimized convolution techniques, and the EIoU loss metric. WH-DETR outperforms existing methods. It is also interesting to consider the approach by Tang et al. [ 59 ], who propose a multi-scale inverse bottleneck residual network model based on ResNet-50 for accurate diagnosis of apple leaf diseases, achieving 98.73% accuracy in recognizing seven types of leaves, including six diseases. This model enhances computational efficiency and feature representation, outperforming classical methods by 1.82% Sidhu et al. [ 60 ] compared traditional linear regression (LR) with boosted regression trees (BRTs) for predicting crop yield responses to climate change. BRTs showed superior accuracy by identifying breakpoints in climate-yield relationships and handling complex interactions among variables affecting crop yields. In simulations and real data from India, BRTs predicted

[Find the meaning and references behind the names: Mae, Rapid, Less, Square, Pearl, Motia, Prior, Development, Aimed, China, Part, Days, Severe, Mean, Smart, Table, Rate, Shm, Nine, Dlm, Millet, Good, Property, Rajasthan]

Sustainability 2024 , 16 , 9437 4 of 29 a less severe negative impact on rice, wheat, and pearl millet compared to LR, highlighting the importance of robust modeling techniques in diverse agricultural settings Li et al. [ 61 ] focused on enhancing crop yield projections under future climates using a combination of dynamic linear models [ 62 ] (DLM), RF, and nine global gridded crop models [ 63 ] (GGCM). By integrating RF with GGCM, significant improvements were achieved in predicting maize and soybean yields across China. Key factors influencing yields included chilling days, crop pests and diseases, and drought for maize, and crop pests and diseases, tropical days, and drought for soybean. The approach reduced uncertainties for maize and soybean, offering a robust framework to enhance future climate impact assessments on crop yields Understanding soil nutrients and properties helps in managing soil health effectively Advances in sensing and computational technologies have made vast amounts of farmland data accessible, enabling the rapid adoption of ML techniques to analyze soil conditions. Motia et al. [ 64 ] discussed the use of ML in predicting and assessing soil properties to enhance agricultural soil health management [ 65 ] (SHM), highlighting key ML algorithms, tools, performance metrics, and identifying challenges and future research directions. ML shows promising potential for sustainable agricultural development through improved soil property prediction and management practices India, the world’s second-largest exporter of agricultural products, relies heavily on accurate crop yield predictions for food security. Jhajharia et al. [ 66 ] used ML techniques to predict crop yield in the Rajasthan region of India, utilizing multi-source data including vegetation indices and weather data. Random forest proved to be the most accurate method, achieving a coefficient of determination ( R 2 ) of 0.77, a root mean square error (RMSE) of 0.39 t/ha, and a mean absolute error (MAE) of 0.28 t/ha. The results provide a good estimation of crop yield prior to harvest, enabling farmers to prepare for environmental changes that may impact yield Agriculture is a key part of India’s economy and a primary source of employment, exhibiting resilience during the COVID-19 pandemic with a 3.4% growth rate in 2020–2021 To sustain the growing population and ensure food security, researchers [ 67 ] leveraged technologies like remote sensing (RS) and ML to create smart, sustainable, and lucrative farming systems. This paper presents a comprehensive review of studies on the application of RS and ML in addressing agriculture-related challenges in India, covering crop management, soil management, and water management. The review, conducted from 2015 to 2022, highlights the potential of intelligent geospatial data analytics in Indian agriculture, with a focus on crop management, where RS sensors and ML techniques have yielded substantial improvements in agricultural monitoring, enabling effective management and valuable recommendations Table 1 summarizes various studies on modeling crop yield based on soil and climate properties using statistical and ML techniques. Each study focuses on different aspects such as predicting soil properties, assessing crop yield under climate variables, and exploring the application of ML in agriculture This study aimed to enhance the sustainability of agricultural practices by improving crop yield predictions through the integration of transformed data with state-specific climatic averages and seasonal adjustments. By incorporating aggregated climatic data—such as temperature, humidity, pressure, and precipitation—specific to each season and state, and omitting the year of cultivation from production data, this study sought to increase the accuracy of yield predictions. One-hot encoding (OHE) was applied to categorical features like crop type and season. A Deep Neural Network (DNN) was utilized for yield prediction, with hyperparameters optimized using genetic algorithms (GAs) to maximize the R 2 metric Furthermore, this study aimed to interpret the DNN model through explainable AI (XAI) techniques like LIME, highlighting the importance of features such as crop type. Following this analysis, a classification model was developed to identify the most sustainable crops for cultivation based on soil and climatic conditions, again employing a GA-optimized DNN to enhance classification accuracy. Ultimately, this study aimed to demonstrate the

[Find the meaning and references behind the names: Malik, Chili, Aim, Frame, Time, Tomato, Area, Non, Early]

Sustainability 2024 , 16 , 9437 5 of 29 effectiveness of this integrated approach in promoting sustainable agricultural practices by improving both crop yield predictions and crop classification Table 1. Summary of studies on modeling crop yield and soil properties using statistical and ML techniques Reference Focus Applied Model Limitations Khanal [ 32 ] Predicting soil properties and corn yield using aerial imagery and ML Random forest, neural network Limited to Molly Caren Farm, Ohio; empirical data-driven approach Rahman et al. [ 39 ] Predicting soil series and recommending crops based on land type Weighted k-NN, Bagged Trees, SVM Assumes homogeneous application of agro-inputs; potential variability in field conditions Wickramasinghe et al. [ 48 ] Modeling rice yield based on climate variables in Sri Lanka ANN, SVMR, MLR, GPR, PR, RR Relies on historical data up to early 2019; limited to specific region and time frame Diaz et al. [ 54 ] Estimating soil quality indicators and predicting crop yields using ML and remote sensing Various ML algorithms Challenges in data integration; variable performance across different agricultural contexts Malik et al. [ 68 ] Comparative analysis of soil properties and crop yield prediction using ML KNN, Naïve Bayes, decision trees Limited to tomato, potato, and chili crops; assumes uniform environmental conditions Liu et al. [ 36 ] Mapping soil properties in Punjab, Pakistan using remote sensing and ML SVM, RFR, MLR Relies on satellite and sensor data; potential inaccuracies in remote sensing data Yadav et al. [ 43 ] Assessing soil fertility and predicting crop yield based on soil features SVM, random forest, Naive Bayes, linear regression, MLP, ANN Limited to soil characteristics; assumes uniform crop management practices Motia et al. [ 64 ] Exploring ML techniques for predicting soil properties and enhancing soil health management Various ML algorithms Challenges in algorithm selection and integration with existing agricultural practices Burdett et al. [ 56 ] Analyzing crop yield responses to soil and topographic characteristics in Southwestern Ontario, Canada MLR, ANN, decision trees, RFR Limited to specific geographical area; potential biases in dataset sampling Sidhu et al. [ 60 ] Comparing LR and BRTs for predicting climate change impacts on crop yields LR, BRTs Challenges in capturing complex interactions and non-linear relationships Li et al. [ 61 ] Integrating DLM, RF, and GGCM for improving crop yield projections under future climates DLM, RF, GGCM Relies on global gridded crop models; uncertainties in climate projections Jhajharia et al. [ 66 ] Crop yield prediction in Rajasthan, India Random forest Limited to Rajasthan region; no mention of scalability Pokhariyal et al. [ 67 ] Review of RS and ML applications in Indian agriculture Various ML models Limited to studies from 2015 to 2022, no mention of future directions 2. Materials and Methods 2.1. Yield in India The first aim of this study is to develop a predictive model for crop yield using a dataset that includes agricultural data from various states in India spanning from 1997 to 2020 [ 69 ]. The dataset comprises several key features, including crop types, cropping seasons, states, areas under cultivation, production quantities, annual rainfall, fertilizer usage, pesticide usage, and calculated yields.

[Find the meaning and references behind the names: Nasa, Better, Million, December, Main, August, Tur, Market, Green, April, Autumn]

Sustainability 2024 , 16 , 9437 6 of 29 During the process of feature engineering and data preprocessing, several techniques were applied to enhance the model’s predictive performance. Initially, an encoding scheme was developed for the cropping seasons by associating each season with specific months to better capture the temporal effects on crop yield. The mappings are as follows: ‘Whole Year’ included all twelve months; ‘Kharif’ included June, July, August, and September; ‘Rabi’ included October, November, December, January, February, and March; ‘Autumn’ included September, October, November; and ‘Summer’ included March, April, May, June; and ‘Winter’ included December, January, and February. This encoding was intended to incorporate seasonal variations more effectively into the model Climate data from NASA, averaged by Indian state, were integrated to account for environmental factors affecting crop yield according to the season in which the crops were grown. This climatic data, including temperature and precipitation metrics, was merged with the primary dataset based on the state where each crop was cultivated. It should be noted that due to the absence of specific geographic coordinates for the locations where the crops were grown, climate data averaged by state had to be used, which represents a limitation of the data Categorical features, such as crop and season, were transformed using one-hot encoding to convert them into numerical values suitable for ML algorithms. This conversion enabled the effective processing of categorical information within the model The ‘Year’ and ‘Production’ features were excluded from the predictive modeling because ‘Production’ correlates with yield. Instead, the focus was placed on predicting ‘Yield’, as it represents a numerical variable (production per unit area) which was defined as the ratio of production to the area under cultivation. The primary objective was to predict this ‘Yield’ variable DNN architecture was employed to model yield predictions. The design and hyperparameters of the DNN were optimized using GA to maximize the R 2 score. This optimization process involved iterative adjustments of model parameters to enhance performance. The GA explored various hyperparameter configurations, resulting in a refined and effective predictive model The integration of feature engineering, climate data, and advanced ML techniques facilitated the development of a robust model for accurate crop yield predictions. This methodology highlights the significance of combining diverse data sources and optimizing model parameters to achieve high predictive accuracy in agricultural forecasting. The schematic pipeline of this process is shown in Figure 1 . Between 1997 and 2004 (Figure 2 ), the main trends in crop yields in India included several key points. Rice remained the primary cereal crop with the largest cultivation areas, ranging from 40 to 43 million hectares. The area under wheat cultivation also remained stable, averaging around 23–24 million hectares. There was an increase in the cultivation areas of pulse crops such as gram, arhar/tur, and moong (green gram). For instance, the area under gram cultivation grew from 2.4 million hectares in 1997 to 5.5 million hectares in 2004. Soybean and rapeseed/mustard showed significant growth in cultivation areas. Soybean cultivation increased from 4.5 million hectares in 1997 to over 7 million hectares in 2004. The areas under sorghum (jowar) and pearl millet (bajra) cultivation also showed significant fluctuations but remained important crops. The area under pearl millet ranged from 4.3 to 5.2 million hectares depending on the year. The area under sugarcane cultivation remained relatively stable, fluctuating around 3.3–4.4 million hectares. The area under maize cultivation increased from 4.8 million hectares in 1997 to 6.2 million hectares in 2004. Overall, the period from 1997 to 2004 saw significant changes in the structure of cultivated areas in India, indicating an adaptation of agriculture to changing market conditions and climatic factors From 2005 to 2012 (Figure 3 ), India saw a variety of trends in crop production and yield. During this period, the cultivation area for many crops fluctuated, with notable increases and decreases in specific years. For example, the area under rice cultivation showed a consistent increase from 43.4 million hectares in 2005 to 50.8 million hectares

[Find the meaning and references behind the names: Soya, New, Trend, Drop, Play, Given, Pea, General, Peak]

Sustainability 2024 , 16 , 9437 7 of 29 in 2011, before slightly decreasing to 43.4 million hectares in 2012. Wheat cultivation also saw a general upward trend, with the area increasing from 25.6 million hectares in 2005 to 27.2 million hectares in 2011, followed by a slight drop in 2012 Crop Yield Prediction Crop Type Crop Year Season State Area Production Annual Rainfall Fertilizer Pesticide Yield Integrated averaged climate data by state Aggregated climate data by season Excluded the crop year, production and annual rainfall Applied OHE to Crop and Season Initial Data Yield Prediction Model Deep Neural Network (DNN) GA for hyperparameter tuning Maximization of R ² as the objective function Model Interpretation Explainable AI (XAI) using LIME for model interpretation Identified that features like Crop play a significant role in predicting yield New Dataset for Crop Classification Climate data: Temperature, Humidity, pH, Rainfall (averaged as optimal for each crop) Soil data: Nitrogen Phosphorus Potassium Conclusions Supporting the Approach Soil and climate data for classifying crops optimal for given conditions proved effective Predicting and classifying crops based on climate and soil data allows for more accurate agricultural planning Figure 1. Summary of crop yield prediction and the classification process Significant changes were observed in the cultivation of pulses. The area under Arhar/Tur (pigeon pea) cultivation fluctuated, peaking at 4.3 million hectares in 2009 before declining to 2.9 million hectares in 2012. Moong (green gram) and Urad (black gram) also experienced similar fluctuations Oilseed crops such as soya beans, saw an increase in cultivation area, growing from 7.6 million hectares in 2005 to 9.6 million hectares in 2011, then dropping to 7.3 million hectares in 2012. Groundnut cultivation showed some volatility, with the area rising to over 6 million hectares in 2006 and 2008, but then dropping to around 4.4 million hectares in 2011 and 2012 Cotton cultivation experienced significant growth, especially from 2009 onward, with the area increasing from 8.7 million hectares in 2005 to a peak of 11.6 million hectares in 2010, before settling at around 8.2 million hectares in 2012 Sugarcane cultivation remained relatively stable, with a slight increase from 4.6 million hectares in 2005 to around 4.8 million hectares by 2012. Other crops such as potato, maize, rapeseed, and mustard also saw variations in their cultivation areas but remained key crops in Indian agriculture during this period.

[Find the meaning and references behind the names: Seed, Castor, Ragi, Small]

Sustainability 2024 , 16 , 9437 8 of 29 Arhar/Tur 3268737 Bajra 4762888 Coconut 928288 Cotton(lint) 4358475 Gram 2386696 Groundnut 5321839 Jowar 9772737 Jute 979874 Maize 4882848 Moong(Green Gram) 1424493 Moth 981694 Potato 985733 Ragi 1381326 Rapeseed &Mustard 2620634 Rice 4 0082643 Sesamum 962409 Small millets 1076680 Soyabean 4532203 Sugarcane 3293842 Urad 1802652 Wheat 23500898 Crop Area Distribution for 1997 Arhar/Tur 3230418 Bajra 5152852 Coconut 1262327 Cotton(lint) 8686973 Gram 5614096 Groundnut 7309299 Jowar 9367107 Jute 882996 Maize 4895988 Moong(Green Gram) 2210760 Oilseeds total 2929337 Other Rabi pulses 840859 Potato 1135037 Ragi 1775820 Rapeseed &Mustard 3051617 Rice 43724135 Sesamum 1351664 Small millets 1281231 Soyabean 5771468 Sugarcane 3766705 Sunflower 1713277 Urad 2155362 Wheat 24431921 Crop Area Distribution for 1998 Arhar/Tur 2976048 Bajra 4950189 Coconut 1317836 Cotton(lint) 8134495 Dry chillies 762191 Gram 5174877 Groundnut 6797635 Horse-gram 899385 Jowar 9785766 Jute 855034 Maize 5542914 Moong(Green Gram) 2692895 Other Rabi pulses 1598782 Other Kharif pulses 1194985 Potato 1167330 Ragi 1789508 Rapeseed &Mustard 3117341 Rice 40988027 Sesamum 1484210 Soyabean 5486504 Sugarcane 4225350 Sunflower 1054177 Urad 3073278 Wheat 24503856 Crop Area Distribution for 1999 Arhar/Tur 3599342 Bajra 5278532 Castor seed 969156 Coconut 1437107 Cotton(lint) 8040548 Gram 4504700 Groundnut 6483421 Jowar 9323728 Jute 836253 Maize 5587217 Masoor 940026 Moong(Green Gram) 2523547 Other Rabi pulses 1005016 Other Kharif pulses 1386553 Potato 1073955 Ragi 1723960 Rapeseed &Mustard 2579149 Rice 43061434 Sesamum 1451264 Small millets 1308796 Soyabean 5721126 Sugarcane 4314727 Sunflower 1052638 Urad 2618695 Wheat 23338921 Crop Area Distribution for 2000 Arhar/Tur 3279088 Bajra 4386705 Coconut 1466568 Cotton(lint) 8528475 Gram 5436208 Groundnut 5988911 Jowar 9160599 Jute 871600 Maize 5339202 Masoor 942641 Moong(Green Gram) 2112018 Other Rabi pulses 1082844 Other Kharif pulses 981826 Potato 1103885 Ragi 1586080 Rapeseed &Mustard 2879561 Rice 42633509 Sesamum 1278108 Small millets 1193263 Soyabean 5646035 Sugarcane 4394411 Sunflower 1168441 Urad 2629306 Wheat 24027129 Crop Area Distribution for 2001 Arhar/Tur 2945928 Bajra 4523284 Coconut 1798981 Cotton(lint) 7247249 Gram 5443194 Groundnut 5688112 Jowar 8750638 Jute 862961 Maize 5586452 Masoor 878800 Moong(Green Gram) 2432658 Other Rabi pulses 1018360 Other Kharif pulses 1191696 Potato 1273397 Ragi 1392053 Rapeseed &Mustard 2714063 Rice 40743133 Sesamum 1167171 Small millets 1264576 Soyabean 5596785 Sugarcane 4491181 Sunflower 1626803 Urad 2686156 Wheat 23410674 Crop Area Distribution for 2002 Arhar/Tur 3419360 Bajra 4798159 Coconut 1804244 Cotton(lint) 7207162 Gram 5949613 Groundnut 5758885 Jowar 8568903 Jute 847669 Maize 6016937 Masoor 889206 Moong(Green Gram) 2617918 Other Rabi pulses 958575 Other Kharif pulses 1204416 Potato 1231746 Ragi 1675109 Rapeseed &Mustard 3005949 Rice 43116666 Sesamum 1343627 Small millets 1028775 Soyabean 6023027 Sugarcane 3995065 Sunflower 1993802 Urad 3140324 Wheat 24734430 Crop Area Distribution for 2003 Arhar/Tur 2893968 Bajra 4740140 Coconut 1794989 Cotton(lint) 8310720 Gram 5599202 Groundnut 6322851 Jowar 8435113 Jute 771105 Maize 6216055 Masoor 846703 Moong(Green Gram) 2429342 Potato 1231178 Ragi 1503732 Rapeseed &Mustard 3367181 Rice 42546933 Sesamum 1418424 Small millets 947293 Soyabean 7011087 Sugarcane 3500136 Sunflower 2479641 Urad 2366180 Wheat 24789592 Crop Area Distribution for 2004 Figure 2. Trends in crop cultivation areas in India (1997–2004). The areas are indicated in hectares.

Sustainability 2024 , 16 , 9437 9 of 29 Arhar/Tur 2972447 Bajra 4685073 Coconut 1840118 Cotton(lint) 8197732 Gram 5737547 Groundnut 6313904 Jowar 8062042 Jute 756270 Maize 6446694 Masoor 865999 Moong(Green Gram) 2186739 Potato 1301756 Ragi 1510591 Rapeseed &Mustard 3358579 Rice 43406782 Sesamum 1324004 Soyabean 7250368 Sugarcane 4208605 Sunflower 2314456 Urad 2211386 Wheat 24101857 Crop Area Distribution for 2005 Arhar/Tur 3277817 Bajra 4611502 Coconut 1825392 Cotton(lint) 8711315 Gram 6614883 Groundnut 5390718 Jowar 7776645 Jute 802194 Maize 6648386 Masoor 1451113 Moong(Green Gram) 2304232 Other Kharif pulses 1008164 Potato 1253611 Ragi 1193546 Rapeseed &Mustard 3318294 Rice 43395708 Sesamum 1397064 Small millets 874621 Soyabean 7558463 Sugarcane 4645383 Sunflower 2128469 Urad 2360629 Wheat 25628234 Crop Area Distribution for 2006 Arhar/Tur 3416274 Bajra 4502801 Coconut 1682671 Cotton(lint) 8982311 Gram 6432780 Groundnut 6016699 Jowar 7135090 Jute 813994 Maize 6876147 Masoor 1305316 Moong(Green Gram) 2561031 Potato 1433048 Ragi 1390420 Rapeseed &Mustard 3062390 Rice 43431956 Sesamum 1450774 Soyabean 8081856 Sugarcane 4929097 Sunflower 1863169 Urad 3038239 Wheat 25687889 Crop Area Distribution for 2007 Arhar/Tur 3110800 Bajra 3570577 Coconut 1752448 Cotton(lint) 9034741 Gram 6550347 Groundnut 5827684 Jowar 6949140 Jute 785628 Maize 6567748 Masoor 1320740 Moong(Green Gram) 1853125 Potato 1475794 Ragi 1379057 Rapeseed &Mustard 3208872 Rice 44887837 Sesamum 1231841 Soyabean 8756922 Sugarcane 4313036 Sunflower 1761923 Urad 2464262 Wheat 25550851 Crop Area Distribution for 2008 Arhar/Tur 3220961 Bajra 3768596 Coconut 1764837 Cotton(lint) 9540100 Gram 6379883 Groundnut 5158689 Jowar 7072045 Jute 813555 Maize 6985274 Masoor 1408654 Moong(Green Gram) 2080784 Potato 1305199 Ragi 1263863 Rapeseed &Mustard 2678460 Rice 41880921 Sesamum 1451258 Soyabean 8928212 Sugarcane 4146000 Sunflower 1427692 Urad 2784359 Wheat 26099513 Crop Area Distribution for 2009 Arhar/Tur 4266477 Bajra 4106411 Coconut 1365649 Cotton(lint) 10594030 Gram 7079887 Groundnut 5604462 Jowar 6385451 Maize 7256368 Masoor 1575581 Moong(Green Gram) 2457533 Potato 1344034 Ragi 1262952 Rapeseed &Mustard 3078125 Rice 42165993 Sesamum 1597430 Small millets 5070598 Soyabean 8798857 Sugarcane 4795686 Sunflower 895525 Urad 3011074 Wheat 27098934 Crop Area Distribution for 2010 Arhar/Tur 3742207 Bajra 3774412 Castor seed 1167829 Coconut 1839997 Cotton(lint) 11685386 Gram 6334218 Groundnut 4865437 Jowar 5577173 Jute 806039 Maize 7590076 Masoor 1451902 Moong(Green Gram) 2039990 Potato 1381847 Ragi 1169967 Rapeseed &Mustard 3073758 Rice 43591035 Sesamum 1408237 Soyabean 9313046 Sugarcane 4786115 Urad 2952769 Wheat 27177143 Crop Area Distribution for 2011 Arhar/Tur 3514554 Bajra 3266356 Castor seed 1006426 Coconut 1390132 Cotton(lint) 11343542 Gram 6689721 Groundnut 4362253 Jowar 5466301 Maize 7557683 Masoor 1302317 Moong(Green Gram) 1782154 Potato 1363729 Ragi 1119609 Rapeseed &Mustard 3241054 Rice 50766758 Sesamum 1218971 Soyabean 9615365 Sugarcane 4977254 Urad 2798236 Wheat 27145258 Crop Area Distribution for 2012 Figure 3. Trends in crop cultivation areas in India (2005–2012). The areas are indicated in hectares.

[Find the meaning and references behind the names: Range, Plant, Transport, Long, Present, Lose, Point, Offer, Cover, Factor]

Sustainability 2024 , 16 , 9437 10 of 29 Observations reveal that rice and wheat consistently occupy the largest areas among all crops across the years. Cotton, soybeans, sugarcane, and chickpeas also cover significant areas, indicating their importance in agriculture. Trends over the years show that the area under rice fluctuated, with the lowest point in 2015 at 42,993,617 hectares and peaking in 2019 at 47,114,882 hectares. Similarly, the area under wheat varied, reaching its highest point in 2021 at 29,718,900 hectares. The cotton area displayed notable fluctuations, with a peak in 2019 at 12,723,411 hectares. Soybean cultivation increased until 2019, reaching 11,090,458 hectares, after which it began to decline. These trends suggest that the area dedicated to rice and wheat has remained substantial despite variations, highlighting their enduring significance. The fluctuations in cotton and soybean cultivation might be influenced by factors such as weather conditions, government policies, and market dynamics. The analysis of data from 2013 to 2020 (Figure 4 ) shows that, despite the changes, rice and wheat continue to dominate in terms of cultivated area, reflecting their role in agriculture. The decrease in the area under rice and wheat in 2020, with rice falling to 257,251 hectares and wheat to 297,189 hectares, likely reflects incomplete data, as the area under cotton was not available and soybean cultivation dropped to 7246 hectares. Examining these trends over a longer period could offer deeper insights into the long-term shifts in agricultural practices and the factors driving them 2.2. Crop Dataset Optimal crop cultivation requires selecting the most suitable environmental and soil conditions for each specific crop, ensuring that each crop thrives under the best possible circumstances for its growth and yield. The dataset [ 70 ] enables users to develop predictive models that recommend the most suitable crops for cultivation based on a variety of soil and environmental parameters. Figures 5 and 6 present histograms illustrating the distribution of cultivated crops in relation to key environmental characteristics One important climatic characteristic is ambient temperature (Figures 5 a and 6 a), measured in degrees Celsius. Temperature is an environmental factor influencing plant growth, development, and yield [ 71 , 72 ]. Each crop species has an optimal temperature range in which it flourishes, and deviations from this range can adversely affect growth and productivity. By closely monitoring temperature, farmers can select crops that are best adapted to the climatic conditions of a particular region, thereby enhancing agricultural efficiency and output Humidity (Figures 5 b and 6 b) refers to the relative humidity of the environment, expressed as a percentage [ 73 ]. It influences the rate of transpiration—the process by which plants lose water vapor through their leaves. High humidity can reduce transpiration, potentially affecting water uptake and nutrient transport. Conversely, low humidity increases transpiration, which may lead to water stress. Understanding the impact of humidity levels is essential for effective irrigation management and maintaining optimal plant health pH (Figures 5 c and 6 c) measures the potential of hydrogen in the soil [ 74 , 75 ] and indicates its acidity or alkalinity. Soil pH significantly affects nutrient availability and microbial activity. Each crop has specific pH requirements for optimal growth. Maintaining the correct soil pH ensures that nutrients are accessible to plants and supports a healthy microbial community Rainfall (Figures 5 d and 6 d) quantifies the amount of precipitation in millimeters and serves as a primary water source for crops [ 76 , 77 ]. It impacts soil moisture levels and irrigation needs. Adequate rainfall is a key factor for crop growth, while excessive or insufficient rainfall can lead to issues such as waterlogging or drought stress. Understanding rainfall patterns aids in planning irrigation strategies and choosing crops suited to specific rainfall conditions.

[Find the meaning and references behind the names: Barley]

Sustainability 2024 , 16 , 9437 11 of 29 Arhar/Tur 3524300 Bajra 3352130 Castor seed 795972 Coconut 1845205 Cotton(lint) 11571082 Gram 7158764 Groundnut 5023980 Jowar 5968463 Maize 7725949 Masoor 1217759 Moong(Green Gram) 2111370 Potato 1449965 Ragi 1187235 Rapeseed &Mustard 3274540 Rice 43978180 Sesamum 1259213 Soyabean 10767771 Sugarcane 5004841 Urad 2671192 Wheat 28431210 Crop Area Distribution for 2013 Arhar/Tur 3472907 Bajra 3242268 Castor seed 828469 Coconut 1839138 Cotton(lint) 12340743 Gram 6771274 Groundnut 4204756 Jowar 5486990 Maize 8137394 Masoor 1303036 Moong(Green Gram) 1960377 Potato 1405150 Ragi 1187109 Rapeseed &Mustard 2942300 Rice 43685941 Sesamum 1398193 Soyabean 9972336 Sugarcane 5125404 Urad 2900127 Wheat 27843326 Crop Area Distribution for 2014 Arhar/Tur 3584485 Bajra 3083095 Castor seed 844690 Coconut 1848078 Cotton(lint) 11691933 Gram 7265810 Groundnut 4046112 Jowar 5439183 Maize 7763773 Masoor 1259603 Moong(Green Gram) 2727998 Oilseeds total 884779 Peas & beans (Pulses) 882223 Potato 1664633 Ragi 1137999 Rapeseed &Mustard 3021484 Rice 42993617 Sesamum 1575733 Soyabean 10394713 Sugarcane 4892321 Urad 3275011 Wheat 27235725 Crop Area Distribution for 2015 Arhar/Tur 4977484 Bajra 3291905 Coconut 1834273 Cotton(lint) 10613198 Gram 7860906 Groundnut 4749202 Jowar 5589256 Maize 8469517 Masoor 1357204 Moong(Green Gram) 2411785 Oilseeds total 1332304 Peas & beans (Pulses) 1036810 Potato 1764913 Ragi 1002146 Rapeseed &Mustard 3281877 Rice 43909738 Sesamum 1296920 Soyabean 10117910 Sugarcane 4421445 Urad 3850438 Wheat 28435104 Crop Area Distribution for 2016 Arhar/Tur 4478184 Bajra 3243723 Coconut 1825288 Cotton(lint) 11825702 Gram 9202032 Groundnut 4249882 Jowar 5735853 Maize 8503971 Masoor 1504642 Moong(Green Gram) 2492248 Oilseeds total 914698 Potato 1692856 Ragi 1193935 Rapeseed &Mustard 3755382 Rice 44379061 Sesamum 1323941 Soyabean 9437926 Sugarcane 4767467 Urad 4436745 Wheat 27554664 Crop Area Distribution for 2017 Arhar/Tur 4466178 Bajra 2933564 Coconut 1995280 Cotton(lint) 11640727 Gram 7987138 Groundnut 4069788 Jowar 4379165 Maize 8242087 Masoor 1329959 Moong(Green Gram) 2123066 Oilseeds total 1056488 Potato 1695965 Ragi 923871 Rapeseed &Mustard 3735999 Rice 45115931 Sesamum 1173607 Soyabean 10200593 Sugarcane 5100854 Urad 4822446 Wheat 28938746 Crop Area Distribution for 2018 Arhar/Tur 4712278 Bajra 3332941 Coconut 1994616 Cotton(lint) 12723411 Gram 7316421 Groundnut 4103115 Jowar 4348367 Maize 9029751 Masoor 1286183 Moong(Green Gram) 2403034 Oilseeds total 853137 Potato 1660013 Ragi 1042104 Rapeseed &Mustard 3681687 Rice 47114882 Sesamum 1342716 Soyabean 11090458 Sugarcane 4592681 Urad 4042663 Wheat 32311773 Crop Area Distribution for 2019 Barley 19280 Garlic 5345 Horse-gram 13149 Maize 19895 Masoor 9246 Other Kharif pulses 14208 Peas & beans (Pulses) 4919 Potato 11447 Ragi 85427 Rapeseed &Mustard 13150 Rice 257251 Small millets 44769 Soyabean 7246 Sugarcane 90184 Urad 12456 Wheat 297189 Crop Area Distribution for 2020 Figure 4. Trends in crop cultivation areas in India (2013–2020); 2020 has incomplete data. The areas are indicated in hectares.

[Find the meaning and references behind the names: Bean, Vary, Musk, Heat]

Sustainability 2024 , 16 , 9437 12 of 29 15 20 25 30 35 0 2 4 6 8 10 12 (a) Histogram of temperature rice maize chickpea kidneybeans pigeonpeas mothbeans mungbean blackgram lentil pomegranate banana 20 30 40 50 60 70 80 90 0 2 4 6 8 10 12 (b) Histogram of humidity rice maize chickpea kidneybeans pigeonpeas mothbeans mungbean blackgram lentil pomegranate banana 4 5 6 7 8 9 10 0 2 4 6 8 10 12 (c) Histogram of ph rice maize chickpea kidneybeans pigeonpeas mothbeans mungbean blackgram lentil pomegranate banana 50 100 150 200 250 300 0 2 4 6 8 10 12 (d) Histogram of rainfall rice maize chickpea kidneybeans pigeonpeas mothbeans mungbean blackgram lentil pomegranate banana Figure 5. Histograms depicting the distribution of cultivated crops across various environmental factors: temperature ( a ), humidity ( b ), pH ( c ), and rainfall ( d ). The data represents crops including rice, maize, chickpea, kidney beans, pigeon peas, moth beans, mung bean, black gram, lentil, pomegranate, and banana Histograms of crop distribution reveal how climatic factors such as temperature and rainfall vary among different crops. For example, crops like bananas and coconuts require higher rainfall, between 250 and 300 mm, while pigeon peas and moth beans thrive with as little as 50 mm. Papayas and grapes exhibit high heat tolerance, with grape temperatures ranging widely from 8 to 43 degrees Celsius. Humidity preferences also differ, with mangoes requiring humidity levels above 70%, whereas coconuts and musk melons thrive in conditions exceeding 90%.

[Find the meaning and references behind the names: Mango, Orange]

Sustainability 2024 , 16 , 9437 13 of 29 10 15 20 25 30 35 40 45 0 2 4 6 8 10 (a) Histogram of temperature mango grapes watermelon muskmelon apple orange papaya coconut cotton jute coffee 50 60 70 80 90 100 0 2 4 6 8 10 12 (b) Histogram of humidity mango grapes watermelon muskmelon apple orange papaya coconut cotton jute coffee 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 0 2 4 6 8 10 (c) Histogram of ph mango grapes watermelon muskmelon apple orange papaya coconut cotton jute coffee 50 100 150 200 250 0 2 4 6 8 10 (d) Histogram of rainfall mango grapes watermelon muskmelon apple orange papaya coconut cotton jute coffee Figure 6. Histograms depicting the distribution of cultivated crops across various environmental factors: temperature ( a ), humidity ( b ), pH ( c ), and rainfall ( d ). The data represent crops including mango, grapes, watermelon, muskmelon, apple, orange, papaya, coconut, cotton, jute, and coffee Figures 7 and 8 show histograms of nitrogen (N), phosphorus (P), and potassium (K) levels across various crops, highlighting distinct nutrient requirements. For instance, nitrogen levels vary widely, with maize and bananas showing higher concentrations compared to pulses like chickpeas and lentils. Phosphorus levels also differ, with rice requiring more phosphorus than pomegranate. Potassium distribution reveals varying crop needs, with chickpeas having higher potassium levels compared to moth beans.

[Find the meaning and references behind the names: Amino, Vital, Energy, Plays, Flower, Rich]

Sustainability 2024 , 16 , 9437 14 of 29 0 20 40 60 80 100 120 0 2 4 6 8 10 12 (a) Histogram of Ratio of Nitrogen content in soil rice maize chickpea kidneybeans pigeonpeas mothbeans mungbean blackgram lentil pomegranate banana 20 40 60 80 0 2 4 6 8 10 12 (b) Histogram of Ratio of Phosphorous content in soil rice maize chickpea kidneybeans pigeonpeas mothbeans mungbean blackgram lentil pomegranate banana 20 30 40 50 60 70 80 0 2 4 6 8 10 12 14 16 (c) Histogram of Ratio of Potassium content in soil rice maize chickpea kidneybeans pigeonpeas mothbeans mungbean blackgram lentil pomegranate banana Figure 7. Histograms illustrating the distribution of cultivated crops based on key macronutrient features: nitrogen ( a ), phosphorus ( b ), and potassium ( c ). The crops analyzed include rice, maize, chickpea, kidney beans, pigeon peas, moth beans, mung bean, black gram, lentil, pomegranate, and banana Nitrogen (Figures 7 a and 8 a) is a component of chlorophyll [ 78 ], the pigment responsible for photosynthesis, and amino acids, which are the building blocks of proteins Adequate nitrogen levels are vital for healthy crop development, significantly impacting overall growth and yield Phosphorus (Figures 7 b and 8 b) plays a role in energy transfer [ 79 ], photosynthesis, and the movement of nutrients within plants. It is also essential for root and flower development. Maintaining optimal phosphorus levels is crucial for ensuring robust plant health and maximizing productivity Potassium (Figures 7 c and 8 c) regulates various physiological processes, including water uptake [ 80 ], enzyme activation, and photosynthesis. Sufficient potassium levels enhance a plant’s disease resistance, improve water use efficiency, and boost both crop quality and yield 0 20 40 60 80 100 120 140 0 2 4 6 8 10 12 14 16 (a) Histogram of Ratio of Nitrogen content in soil mango grapes watermelon muskmelon apple orange papaya coconut cotton jute coffee 0 20 40 60 80 100 120 140 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 (b) Histogram of Ratio of Phosphorous content in soil mango grapes watermelon muskmelon apple orange papaya coconut cotton jute coffee 0 25 50 75 100 125 150 175 200 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 (c) Histogram of Ratio of Potassium content in soil mango grapes watermelon muskmelon apple orange papaya coconut cotton jute coffee Figure 8. Histograms illustrating the distribution of cultivated crops based on key macronutrient features: nitrogen ( a ), phosphorus ( b ), and potassium ( c ). The crops analyzed include mango, grapes, watermelon, muskmelon, apple, orange, papaya, coconut, cotton, jute, and coffee Soil nutrient levels, specifically nitrogen (N), phosphorus (P), and potassium (K), showcase variability, spanning deficient to nutrient-rich soils. Nitrogen, which ranges from

[Find the meaning and references behind the names: Ideal, Broad, File, Risk, Large, Favor, Cost, Grid, Arid, Lower, Target, Positive, Pre]

Sustainability 2024 , 16 , 9437 15 of 29 0 to 140 mg/kg with a mean of 50.55 mg/kg, is vital for vegetative growth and chlorophyll synthesis, making it relevant for leafy crops like rice and maize. Phosphorus levels, with a range from 5 to 145 mg/kg and a mean of 53.36 mg/kg, influence root development, especially in the early growth phases. Phosphorus is strongly correlated with potassium (correlation coefficient of 0.74), indicating that these nutrients often co-exist in balanced soils. This synergy could favor crops with combined nutrient demands, such as root vegetables or legumes, which require robust root systems for optimal growth. Potassium, varying from 5 to 205 mg/kg with a mean of 48.15 mg/kg, is essential for water regulation and disease resistance. Crops like bananas and other tropical fruits, which are highly potassium-dependent, could benefit from this range of potassium levels The temperature range captured in the dataset, from 8.8 ◦ C to 43.7 ◦ C, with an average of 25.6 ◦ C, reflects diverse climatic conditions that accommodate both cool and warmweather crops. Warmer temperatures, for instance, could support heat-tolerant crops like maize and cotton, while cooler temperatures below 20 ◦ C would be ideal for fruits like apples and grapes. Humidity, ranging from 14.3% to nearly 100% and averaging at 71.5%, further enhances this diversity. High humidity levels may support tropical crops, while lower levels favor drought-resistant varieties like chickpeas and pulses. The positive correlation between potassium and humidity (0.19) suggests that soils in humid environments often have higher potassium levels, potentially supporting potassium-dependent crops that thrive in moist conditions Soil pH, ranging from 3.5 to 9.9 with a mean of 6.47, represents an extensive spectrum from acidic to moderately alkaline soils. This variation is significant, as pH influences nutrient availability and soil microbial activity. Neutral to slightly acidic soils (pH 6–7) are generally ideal for a broad range of crops, while strongly acidic soils (below pH 5.5) are favorable for crops like coffee and citrus, which are well-suited to such conditions. Conversely, moderately alkaline soils can support crops that thrive at a higher pH, like barley and some legumes Rainfall spans from 20.2 to 298.6 mm, with an average of 103.46 mm, covering arid to monsoon-like conditions. This variability allows the dataset to account for crops with high water demands, like rice and sugarcane, as well as those adapted to low rainfall, such as chickpeas and lentils. Additionally, the positive correlation between rainfall and humidity (0.09) underscores the interdependence of these conditions, which can impact crop resilience to water availability 2.3. DNN Methodology An approach to optimizing hyperparameters for a DL model applied to a regression and classification task using GA [ 81 ] was employed GA offers a robust advantage in hyperparameter optimization by adaptively exploring complex, high-dimensional search spaces, reducing the risk of local optima, and handling mixed parameter types more efficiently than, for example, grid search or Bayesian optimization. Grid search [ 82 ], though effective in exploring a pre-defined parameter space, becomes computationally expensive when the parameter space is large. It exhaustively evaluates each possible combination of parameters, which can be infeasible for models with high-dimensional hyperparameter spaces, resulting in a substantial time and resource cost. Bayesian optimization [ 83 ], while more efficient, relies on probabilistic modeling (e.g., Gaussian processes) to select promising parameters and can struggle with highly non-linear, irregular search spaces due to its reliance on the assumptions embedded in its surrogate model. Bayesian methods also have limited flexibility in adjusting to dynamic fitness landscapes, which can lead to suboptimal performance in scenarios with complex, multi-modal parameter relationships Initially, data were loaded and preprocessed. The dataset, imported from a CSV file, was divided into feature and target variables. The target variable was encoded using the LabelEncoder method and transformed into a one-hot encoded format suitable for the classification model.

[Find the meaning and references behind the names: Tune, Elu, Standard, Relu, Gelu, Selu, Adam, Mse, Hard]

Sustainability 2024 , 16 , 9437 16 of 29 The data were then split into training and testing sets, and the StandardScaler method was applied to standardize the features, ensuring a mean of zero and a standard deviation of one The model architecture employed was a TensorFlow Keras DNN. The hyperparameter space included a wide range of configurations, with the number of layers varying from 1 to 20 and neurons per layer ranging from 1 to 128. This flexibility enabled fine-tuning of the network’s depth and density to determine the optimal setup. Activation functions [ 84 ] used included ReLU, Sigmoid, Tanh, Softmax, Softplus, Softsign, ELU, SELU, GELU, hard Sigmoid, and linear, enabling the selection of the most suitable functions for the data Various optimization algorithms [ 85 ] were employed to adjust the learning rate during training, including Adam, SGD, RMSprop, Adagrad, Adadelta, Adamax, and Nadam. Learning rates were parameterized at 0.0001, 0.001, 0.01, and 0.1 to fine-tune gradient descent for optimal convergence. Categorical cross-entropy was used as the loss function for multiclass classification, determining how the network penalizes errors The overall hyperparameter space H for a TensorFlow Keras DNN can be described mathematically as follows: • Number of layers: L ∈ { 1, 2, . . . , 20 } where L denotes the total number of layers in the network • Number of neurons per layer: N l ∈ { 1, 2, . . . , 128 } for l ∈ { 1, 2, . . . , L } where N l represents the number of neurons in the l -th layer • Activation functions: A ∈ { Linear, Hard Sigmoid, GELU, SELU, ELU, Softsign, Softplus, Softmax, Tanh, Sigmoid, ReLU } where A denotes the activation function applied in each neuron • Optimizers: O ∈ { Nadam, Adamax, Adadelta, Adagrad, RMSprop, SGD, Adam } where O represents the optimization algorithm used for training the network • Learning rates: η ∈ { 0.0001, 0.001, 0.01, 0.1 } where η is the learning rate parameter used in the optimization process • Loss function for classification task: L f c = Categorical cross-entropy where L f c is the loss function employed for evaluating the performance of the classification model • Loss function for the regression task: L f r = MSE, MAE, logcosh, huberloss where L f r is the loss function employed for evaluating the performance of the regression model The overall hyperparameter space H for a TensorFlow Keras DNN can be described as follows: H = { ( L , N , A , O , η , L f ) }

[Find the meaning and references behind the names: Step, Class, Train, Dor, Size, Held, Fixed, Batch]

Sustainability 2024 , 16 , 9437 17 of 29 The fitness function [ 86 ] evaluated the quality of models trained with different hyperparameter combinations. It involved building and compiling the model with the selected hyperparameters, training it on the training data with fixed epochs and batch size, and testing it on a held-out test set. Model predictions were compared with true class labels to calculate accuracy, which served as the primary criterion for the fitness function. Additionally, loss history and accuracy at each training step were recorded, providing a detailed evaluation of the training process and stability Results from each generation of the algorithm were stored, allowing analysis of the hyperparameter evolution [ 87 ] and the effectiveness of the GA. This comprehensive assessment guided the evolutionary process toward identifying the best hyperparameters for the classification task Model quality was assessed using accuracy. The evaluation involved compiling the model with the specified optimizer and loss function, training it on the training set, and making predictions on the test set. Prediction accuracy was computed using the accuracy score function from the scikit-learn library The described GA process could be defined mathematically as follows: • Generate an initial population P 0 of candidate hyperparameter sets { h i } N i = 1 , where each h i includes the following: h i = ( L i , N i , A i , O i , η i , L f ) • For each hyperparameter set h i , train the model on the training dataset for E epochs and batch size B Train ( h i , E , B ) • Test the model on a held-out test set to compute the DOR for the classification task: Accuracy = Number of Correct Predictions Total Number of Predictions and, for the regression task: R 2 = 1 − ∑ M j = 1 ( y j − ˆ y j ) 2 ∑ M j = 1 ( y j − ¯ y ) 2 where y j are true values, ˆ y j are predicted values, and ¯ y denotes the mean of the true values • Define the fitness function F ( h i ) as accuracy or R 2 for the required task: F ( h i ) = Accuracy F ( h i ) = R 2 • Select individuals with higher fitness scores using a selection strategy (e.g., roulette wheel or tournament selection) • Perform crossover between pairs of selected individuals to create offspring. Let h i and h j be the parents; the offspring h ′ ij is generated by the following: h ′ ij = Crossover ( h i , h j ) • Apply mutation to the offspring to introduce genetic diversity: h ′′ ij = Mutation ( h ′ ij ) • Form a new population P t + 1 by combining the offspring with the best individuals from the current generation: P t + 1 = SelectBest ( P t ∪ Offspring )

[Find the meaning and references behind the names: Top, Final, Arg, Alpha]

Sustainability 2024 , 16 , 9437 18 of 29 • Continue iterations until a stopping criterion is met (e.g., the maximum number of generations G or convergence): t = min ( G , Convergence ) • Identify the best hyperparameter set h best based on the highest fitness score: h best = arg max h i ∈P G F ( h i ) • Analyze the evolution of hyperparameters across generations to evaluate the GA’s effectiveness: Analyze ( P 0 , P 1 , . . . , P G ) 3. Results Figure 9 illustrates the model accuracy evolution across various neural network configurations, sorted in ascending order. Initially, over 250 individuals exhibited low accuracy, below 0.2. Subsequently, a rapid increase in accuracy was observed up to 0.95 by the 350 th individual, stabilizing around 0.97 ± 0.01 up to the 450 th individual. The maximum accuracy of 0.998 was achieved at the final point 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Individuals 0.2 0.4 0.6 0.8 Scor e Metrics Evolution Accuracy R 2 score Generation 0 Generation 1 Generation 2 Generation 3 Figure 9. Evolution of metrics sorted in ascending order Table 2 presents the best hyperparameter configurations obtained after the optimization process. It includes the number of layers, neurons per layer, activation functions, optimizer, learning rate (alpha), and the corresponding metrics achieved by each configuration. Metric scores are derived from a rigorous evaluation process involving 5-fold cross-validation, which ensures that the reported results are robust and generalizable across different subsets of the data. This approach helps to mitigate overfitting and provides a reliable estimate of model performance Figure 10 presents a comparison between actual and predicted yield values for the architecture ‘Yield prediction’ from Table 2 . This analysis aims to illustrate the model’s accuracy in forecasting agricultural outcomes. With an. With an R 2 value of 0.92, there is a strong correlation between the predicted and actual yield predictions The integration of explainable AI (XAI) techniques, such as local interpretable modelagnostic explanations (LIME), into DNNs aimed at yield prediction is important for applications in agriculture. Understanding the contribution of various factors to crop yield can enhance decision-making and policy formulation. The LIME diagram (Figure 11 ) highlights the top 70 features impacting the model’s predictions. These features are categorized into

[Find the meaning and references behind the names: Mpe]

Sustainability 2024 , 16 , 9437 19 of 29 two main groups based on their impact direction and magnitude. Negative impact features negatively influence yield predictions, with significant contributors including various crops (e.g., Soybean, Cotton, Wheat, Rice), seasonal factors (e.g., summer, winter, Kharif, Rabi), and environmental parameters (e.g., temperature, humidity, precipitation). Positive impact features positively influence yield predictions, although the specific positive impact features are not explicitly listed in the provided diagram snippet Table 2. DNN hyperparameter configurations and scores on the test set Task Layers Neurons Per Layer Activation Functions Optimizer Alpha Loss Function Metrics Type Score Yield prediction 2 [24, 126] [Softmax, Softsign] Adamax 0.1 MSE R 2 score 0.92 ± 0.04 RMSE 2.45 ± 0.12 MPE 1.75 ± 0.09 Crop selection 2 [79, 111] [linear, GELU] Adam 0.001 categorical crossentropy Accuracy 0.87 ± 0.11 Crop selection 6 [36, 53, 62, 9, 39, 9] [linear, Tanh, ReLU, Tanh, Sigmoid, ...] Adagrad 0.100 categorical crossentropy Accuracy 0.91 ± 0.07 Figure 10. Comparison of actual and predicted yield values, demonstrating the model’s high accuracy with an R 2 of 0.92, which confirms its effectiveness in forecasting Analyzing the key features, crop-specific features like soybean, cotton, wheat, and rice exhibit significant negative impacts on yield prediction. This could indicate the model’s sensitivity to the areas under these crops, suggesting their critical role in determining overall yield. The impact of different seasons (e.g., summer, winter, Kharif, Rabi) on yield prediction highlights the importance of temporal factors in agricultural productivity. Variables such as temperature, humidity, and precipitation reflect the model’s reliance on climatic conditions to predict crop yields accurately To improve the model, incorporating real-time data from remote sensing technologies, such as satellites, can enhance the model’s responsiveness to dynamic environmental

[Find the meaning and references behind the names: Minor, Inter, Pepper, Temp, Lat, Lon]

Sustainability 2024 , 16 , 9437 20 of 29 changes. While GA has proven effective for HPO, it is computationally intensive. Future research could explore hybrid approaches combining GA with other optimization algorithms to enhance efficiency and scalability. Developing region-specific and crop-specific models can improve prediction accuracy by tailoring models to local environmental conditions and agricultural practices, thereby enhancing their relevance and applicability. Expanding the dataset to include diverse geographical regions and more comprehensive environmental variables can improve the model’s generalizability, ensuring that it performs well across different contexts and is not overly reliant on region-specific data. Ensuring robust and scalable predictions across various regions and contexts necessitates the integration of multiple environmental variables. This includes climatic factors, soil conditions, and agricultural practices, which can be challenging but essential for accurate yield predictions 0.100 0.075 0.050 0.025 0.000 0.025 0.050 0.075 Crop_Coriander <= -0.10 Crop_Oilseeds total <= -0.04 Crop_Khesari <= -0.06 Crop_Cardamom <= -0.06 Crop_Guar seed <= -0.06 Crop_Soyabean <= -0.14 Crop_Cotton(lint) <= -0.16 Crop_Garlic <= -0.12 Crop_Barley <= -0.13 Crop_Groundnut <= -0.20 Crop_Sannhamp <= -0.10 Crop_Sesamum <= -0.19 Crop_Sweet potato <= -0.12 Crop_Onion <= -0.15 Crop_Linseed <= -0.13 Crop_Tobacco <= -0.14 Crop_Niger seed <= -0.10 Crop_Turmeric <= -0.13 Season_Autumn <= -0.14 Crop_Mesta <= -0.10 Crop_Other Rabi pulses <= -0.13 Crop_other oilseeds <= -0.08 Crop_Masoor <= -0.13 Season_Summer <= -0.25 Crop_Sunflower <= -0.15 Crop_Other Kharif pulses <= -0.14 Crop_Urad <= -0.19 Crop_Banana <= -0.11 Crop_Peas & beans (Pulses) > -0.13 Crop_Jute <= -0.10 Crop_Moong(Green Gram) <= -0.20 Crop_Sugarcane <= -0.18 Crop_Coconut <= -0.09 Crop_Castor seed <= -0.12 Crop_Wheat <= -0.17 Top 35 Features 0.015 0.010 0.005 0.000 0.005 0.010 0.015 Crop_Gram <= -0.16 Crop_Rapeseed &Mustard <= -0.16 Crop_Cowpea(Lobia) <= -0.08 Crop_Bajra <= -0.17 Crop_Other Cereals <= -0.09 Crop_Safflower <= -0.09 Season_Whole Year <= -0.48 Crop_Horse-gram <= -0.14 Crop_Small millets <= -0.16 Crop_Maize <= -0.23 Season_Winter <= -0.14 Crop_Moth <= -0.08 Crop_Dry chillies <= -0.15 Crop_Ginger <= -0.13 Crop_Cashewnut <= -0.08 -0.85 < Season_Kharif <= 1.18 Crop_Jowar <= -0.16 Crop_Other Summer Pulses <= -0.02 Crop_Tapioca <= -0.10 Crop_Arhar/Tur <= -0.16 Crop_Rice <= -0.25 temp <= -0.32 humidity <= -0.80 pressure <= -0.10 precipitation <= -0.79 Crop_Ragi <= -0.16 Pesticide <= -0.22 Crop_Arecanut <= -0.09 Season_Rabi <= -0.64 lat > 0.64 Crop_Black pepper <= -0.08 Fertilizer <= -0.25 Area <= -0.24 Crop_Potato <= -0.18 lon <= -0.80 Next 35 Features Figure 11. LIME explanations of the ‘Yield prediction’ DNN architecture from Table 2 . The confusion matrices for crop selection models (from Table 2 ) are depicted in Figure 12 , illustrating minimal inter-class misclassification, as evidenced by the near absence of off-diagonal elements. The labels used in the tables are as follows: apple (0), banana (1), black gram (2), chickpea (3), coconut (4), coffee (5), cotton (6), grapes (7), jute (8), kidney beans (9), lentil (10), maize (11), mango (12), moth beans (13), mung bean (14), muskmelon (15), orange (16), papaya (17), pigeon peas (18), pomegranate (19), rice (20), and watermelon (21) The difference between the architectures is particularly notable for moth beans (13), where the two-layer model correctly identifies 21 entries, while the six-layer model achieves 24 correct identifications. There is also some minor confusion observed with the lentil class (10).

[Find the meaning and references behind the names: Modern, Solar]

Sustainability 2024 , 16 , 9437 21 of 29 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Predicted 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 True 23 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 26 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 27 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 22 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 11 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 23 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 22 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 23 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 Confusion Matrix 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Predicted 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 True 23 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 26 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 27 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 22 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 11 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 24 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 23 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 23 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 Confusion Matrix Figure 12. Confusion matrices for the model’s crop selection (from Table 2 ) for the test set 4. Discussion Yield prediction and the selection of optimal conditions for crop growth are vital components of modern agriculture. The application of ML techniques to these tasks has the potential to transform agricultural practices by providing accurate, data-driven insights that can improve productivity and sustainability. The workflow diagram in Figure 13 outlines the structured approach for crop yield prediction and optimization using ML. Divided into key stages, this framework highlights the steps from data acquisition to interpretation and future directions ML algorithms are increasingly being used to predict crop yields based on various environmental, climatic, and agricultural factors. These algorithms analyze historical data to identify patterns and make predictions about future yields. Commonly used ML techniques in this domain include DNNs, support vector machines [ 88 ] (SVMs), and random forests [ 89 ] (RFs). DNNs are capable of modeling complex, non-linear relationships between input features and yield outcomes. They are particularly effective when large datasets are available, as they can learn intricate patterns from the data. SVMs are used for regression and classification tasks and have been applied to yield prediction due to their ability to handle high-dimensional spaces and define clear decision boundaries. RFs are ensemble methods that combine multiple decision trees to improve predictive accuracy and control overfitting [ 90 ]. They are useful in yield prediction due to their robustness and interpretability ML models utilize a wide array of features to predict crop yields. These features can be broadly categorized into environmental, climatic, and agricultural factors. Environmental factors such as soil type, pH levels, nutrient content, and topography influence crop growth and yield. Climatic factors, including temperature, humidity, precipitation, and solar radiation, are critical determinants of crop performance. Agricultural practices like crop variety, planting density, irrigation practices, and fertilizer usage directly impact yield outcomes This study highlighted the integration of GA optimization with DNN for yield prediction. The model was optimized with specific parameters, including the number of layers, activation functions, optimizer, and learning rate, achieving a high R 2 score of 0.92 ± 0.04 Local interpretable model-agnostic explanations (LIME) were employed to provide in-

[Find the meaning and references behind the names: Cleaning, Trust]

Sustainability 2024 , 16 , 9437 22 of 29 sights into the model’s decision-making process, identifying key features such as crop type, seasonal factors, and environmental conditions. While LIME plots offered insights into the model’s decision-making, future research should explore more advanced methods like Shapley additive explanations [ 91 ] (SHAP). Utilizing SHAP can offer deeper insights into feature contributions and enhance model transparency. Comparative experiments between LIME and SHAP could validate the robustness of explanations and improve the credibility of the model. This focus on interpretability is essential for building trust in machine learning applications, especially in critical domains where understanding model behavior is vital Crop Yield Prediction and Optimization Workflow Data Preparation Modeling & Optimization Interpretation & Future Input Data Sources Feature Extraction and Data Preprocessing ML Models for Yield Prediction Optimization Techniques Interpretability Models Result Interpretation and Communication Challenges and Limitations Future Directions Environmental, climatic, and agricultural data from diverse sources Cleaning, normalizing, and selecting relevant features Using DNNs, SVMs, RFs for prediction GA optimization for hyperparameters and efficiency LIME for initial insights, SHAP for advanced explanation Visuals, reports, and decision support for policy-making Data quality, computational power, generalizability issues Real-time data, expanded datasets, and new optimization methods Figure 13. Workflow diagram for crop yield prediction and optimization using ML techniques Translating complex model explanations into accessible graphics or concise reports is essential for enhancing the usability of research findings among users and policymakers. Such visual representations can demystify intricate analytical results, making them more comprehensible to non-expert audiences. For example, Berget [ 92 ] introduces a spatial multi-agent programming model designed to evaluate policy options for the diffusion of innovations and changes in resource use. Utilizing a multi-agent/cellular automata approach, the model explicitly captures the social and spatial interactions of heterogeneous farm households, incorporating economic and hydrologic processes to analyze the impacts of water use for irrigation and the adoption of agricultural innovations. Agricultural systems science offers insights for tackling complex issues and making informed decisions. As

[Find the meaning and references behind the names: Choice, Niche, Harder, Big, Still, Past, Common, Cloud]

Sustainability 2024 , 16 , 9437 23 of 29 researchers strive to create next-generation models, it is notable to reflect on the rich history of agricultural modeling, which encompasses diverse approaches and collaborations [ 93 ]. By learning from past developments, the community can effectively advance future models and decision support systems while avoiding common pitfalls Despite the successes, several challenges remain in the application of ML to yield prediction and optimal crop growth conditions. High-quality, comprehensive datasets are essential for training accurate ML models. Data gaps [ 94 ] can particularly limit model effectiveness. While complex models like DNNs offer high predictive accuracy, they often operate as “black boxes” [ 95 ]. Techniques like LIME are essential for interpreting these models, but further advancements in explainable AI are needed. Models trained on data from specific regions or conditions may not generalize well to other areas. Incorporating diverse datasets from various geographical regions can improve model robustness. Training complex ML models, especially those involving GA optimization, requires significant computational power. Developing more efficient algorithms and leveraging cloud-based solutions can mitigate this challenge Additionally, the model’s limitations include reduced accuracy in predicting yields for less common crops [ 96 ], such as pigeon peas, moth beans, and other niche varieties, which may not have as extensive historical or environmental data available. Moreover, crops grown in challenging or extreme conditions—such as drought-prone or highly saline areas—are harder to model accurately [ 97 ]. This is due to limited data on crop resilience and the complex interactions between environmental stressors and crop performance that the model may not fully capture Expanding the model’s applicability to developing countries could provide more agricultural benefits. By integrating insights from these findings into decision-support tools, smallholder farmers in resource-limited settings could optimize crop selection and timing. Such models, when combined with mobile technology and IoT-based soil and weather monitoring, could empower farmers with predictive insights, helping them adapt to climate variability, improve yield forecasts, and make informed decisions to increase food security and reduce losses [ 98 ]. To improve the model, incorporating real-time data from remote sensing technologies, such as satellites and IoT devices [ 99 ], can enhance model accuracy and responsiveness to changing conditions. Combining GA with other optimization algorithms can improve efficiency and scalability, reducing computational demands. Additionally, integrating GA optimization significantly enhances the performance of DNNs by fine-tuning their hyperparameters, thereby improving predictive accuracy. While utilizing random choice for HPO can still wield significant influence over DNN performance, their optimal values often remain elusive due to the exponential space and complex interactions involved Traditional methods are often impractical for this purpose, but evolutionary algorithms like GA offer scalable solutions It should also be noted that the dataset used in this study was primarily limited to specific regions and crop types, which presents a potential limitation in the model’s generalizability. This regional and crop-specific focus means that while the model performs well within the confines of the dataset, its applicability to broader geographical areas and a more diverse range of crops may be constrained. Future research could address this limitation by expanding the dataset to include additional regions with varying climates, soil conditions, and a wider variety of crop types. Such an expansion would not only improve the model’s robustness and predictive accuracy across diverse agricultural settings but also enhance its adaptability to different environmental conditions, making it more broadly applicable to real-world agricultural challenges. This broader dataset could enable more comprehensive yield prediction and crop recommendations that are relevant across various regions, ultimately contributing to more sustainable and adaptable agricultural practices Big data applications in agriculture are advancing but face challenges in scalability and real-world readiness due to the inherent complexity of agricultural data (volume, velocity, variety, veracity). Although the adoption of big data in agriculture is growing, practical

[Find the meaning and references behind the names: Ways, Resources, Level, Higa, Beyond, Monte, Kao, Classic, Carlo, Shrestha, Right]

Sustainability 2024 , 16 , 9437 24 of 29 deployment remains limited, as most solutions require domain-specific customization and remain at a low technological maturity level [ 100 ]. Addressing these gaps requires both practical engineering and systems-thinking approaches to deliver scalable, affordable, and user-friendly solutions tailored to agriculture’s unique needs. However, the current study did not delve into the specific challenges associated with existing data processing technologies. Issues such as processing speed, data integration from disparate sources, and the security of data storage and transmission are factors that can impact the feasibility and effectiveness of big data applications in agriculture. Without addressing these challenges, the potential of big data to transform agricultural practices may be constrained, as processing bottlenecks and security vulnerabilities could limit its accessibility and reliability for widespread agricultural use. Future studies should explore these technical challenges in greater depth to provide solutions that enhance data processing efficiency and security, thereby facilitating the broader adoption of big data technologies in the agricultural sector For example, Shrestha et al. [ 101 ] integrated nature-inspired enhancements and Monte Carlo-based methods to enhance DNN architecture for superior accuracy on datasets compared to traditional genetic algorithms. Their study leveraged augmented datasets comprising comprehensive rainfall, climate, and fertilizer data sourced from India, serving as robust training and validation sets for the model. By employing GA, the DNN achieves superior classification results, effectively identifying optimal crop choices tailored to diverse environmental scenarios. Another approach, suggested by Liu et al. [ 36 ], involves a hybrid intelligent genetic algorithm (HIGA) that integrates DNNs and GAs. A two-step training approach refines the DNN with data from the optimization process, enhancing predictive accuracy for efficient truss optimization. Evaluating three classic truss problems validates the method’s effectiveness while exploring various settings demonstrates its robustness and applicability in improving optimization performance and efficiency DNN layers are complex loops that can be organized, tiled, and scheduled in various ways on DNN accelerators. Optimal per-layer mappings are beneficial, yet selecting the right mapping remains challenging due to the vast map space. Kao et al. [ 102 ] introduced GAMMA, a genetic algorithm-based approach tailored for hardware mapping. Unlike previous methods limited to specific accelerators or mappings, GAMMA explores a flexible map space efficiently. Comparative evaluations demonstrate GAMMA consistently outperforms other methods in finding optimized mappings Despite its strengths, this approach exhibits several limitations that warrant consideration for future research directions. Firstly, while the DNN model proves effective in predicting crop suitability based on available environmental data, its reliance on historical datasets may limit its adaptability to real-time or dynamically changing conditions [ 103 ]. Incorporating real-time data, potentially obtained from remote sensing, particularly satellites [ 104 ], could enhance global and regional analysis of crop conditions and production For example, Wu et al. [ 105 ] employ a hierarchical approach that integrates climatic and remote sensing indicators at multiple scales—from global environmental conditions to detailed assessments at the national and sub-national levels. This methodology provides accurate and timely information, supporting food producers with insights into crop health, farming intensity, and production trends The complexity of integrating multiple environmental variables into the model poses challenges in ensuring the robustness and scalability of the predictions across different geographical regions and varying agricultural contexts. Furthermore, the use of GA for HPO, while effective, requires significant computational resources [ 106 ] and time-intensive experimentation. Future studies could explore more efficient optimization techniques or hybrid approaches that combine GA with other classification algorithms to further enhance model performance and scalability. Additionally, incorporating more diverse and comprehensive datasets from global regions could improve the generalizability and applicability of the model beyond specific geographical boundaries By advancing these methodologies, it is possible to enhance the precision and applicability of crop yield predictions across different regions and crops. Exploring the integration

[Find the meaning and references behind the names: Di Natale, Di Giuseppe, Pasqualetti, Board, Sharma, Nair, Capuano, Joshi, Read, Giuseppe, Barba, Castagnini, Original, Taneja, Natale, Catini, Agri, Papale, Author, Soto, Node]

Sustainability 2024 , 16 , 9437 25 of 29 of advanced optimization techniques and leveraging diverse data sources will be pivotal in developing robust, scalable models capable of addressing the dynamic challenges in agriculture. This multifaceted approach holds the potential to significantly improve agricultural productivity and sustainability worldwide 5. Conclusions The application of GA optimization enhances the predictive accuracy and performance of DNNs by fine-tuning their hyperparameters. By utilizing augmented datasets from India that include rainfall, climate, and fertilizer data, the model is effectively trained and validated, achieving a resulting R 2 score of 0.92 ± 0.04. GA significantly improves hyperparameter optimization, allowing the DNN to deliver superior classification results and identify optimal crops suited to diverse environmental conditions Future research could expand the DNN-GA yield prediction model’s scope by adapting it to diverse and less common crops, allowing for better regional customization and climate resilience modeling. Integrating real-time IoT data (e.g., soil moisture, temperature) would support precision agriculture by providing timely insights, while improvements in feature engineering could highlight key variables that most affect yield, enhancing both model simplicity and interpretability. Developing computationally efficient versions of the model would also make it accessible to farmers in low-resource settings, where data and processing power are limited. Additionally, the model could have cross-applications in agricultural insurance and economic risk assessment, offering reliable forecasts that inform crop-specific insurance products and sustainable farming practices. Finally, embedding this model in agronomic advisory systems would help farmers make data-driven decisions, from planting schedules to resource management, bolstering productivity and sustainability in the agricultural sector In conclusion, this study highlights the potential of DNNs and GA in promoting sustainability within precision agriculture. However, ongoing research should address challenges related to data availability, model interpretability, generalizability, and computational resource requirements. Incorporating real-time data, developing hybrid optimization techniques, creating region-specific models, and integrating diverse datasets will further enhance the adaptability, accuracy, and applicability of predictive models. By tackling these challenges, the agricultural sector can advance toward more sustainable and efficient practices on a global scale Author Contributions: Conceptualization, I.M.; formal analysis, A.G. and Y.T.; funding acquisition, A.B., A.G., V.N. and Y.T.; methodology, V.T.; project administration, A.B., A.G., V.N. and Y.T.; resources, I.M.; software, V.T., A.B. and V.N.; supervision, A.B., A.G., V.N. and Y.T.; validation, I.M. and V.T.; visualization, I.M.; writing—original draft, I.M.; writing—review and editing, V.T. All authors have read and agreed to the published version of the manuscript Funding: This research received no external funding Institutional Review Board Statement: Not applicable Informed Consent Statement: Not applicable Data Availability Statement: Data is contained within the article Conflicts of Interest: The authors declare no conflicts of interest References 1 Charania, I.; Li, X. Smart farming: Agriculture’s shift from a labor intensive to technology native industry Internet Things 2020 , 9 , 100142. [ CrossRef ] 2 Catini, A.; Papale, L.; Capuano, R.; Pasqualetti, V.; Di Giuseppe, D.; Brizzolara, S.; Tonutti, P.; Di Natale, C. Development of a sensor node for remote monitoring of plants Sensors 2019 , 19 , 4865. [ CrossRef ] [ PubMed ] 3 Taneja, A.; Nair, G.; Joshi, M.; Sharma, S.; Sharma, S.; Jambrak, A.R.; Roselló-Soto, E.; Barba, F.J.; Castagnini, J.M.; Leksawasdi, N.; et al. Artificial intelligence: Implications for the agri-food sector Agronomy 2023 , 13 , 1397. [ CrossRef ]

[Find the meaning and references behind the names: Zahoor, Foods, Meenakshi, Mol, Wiley, Saraswathy, George, Nepal, Qaisar, Khan, Singh, Selvi, Priyanka, Western, Barman, Duan, Nasir, North, Gupta, Kamal, Gene, Velmurugan, Dar, Wang, Patil, Mitra, Goel, Idris, Nagendran, Zaman, Grain, Int, Malek, Deb, Kumar, Sci, Busato, Sobhan, Bud, Mesta, Devi, Jin, Berlin, Melo, Acharya, Blaise, Abotaleb, Chain, Carbon, Central, Shah, Arya, Narasimha, Germany, Upadhayay, Mandal, Agarwal, Manan, Kaur, Front, Bus, Chand, Pandey, Cham, Parker, Bakr, Bio, Santra, Gowda, Kashmir, Alam, Chelladurai, Pearson, Bhat, Dutt, Pala, Mughal, Web, Senthil, Tiwari, Choudhury, Verma, Jammu, Ramadhan, Reddy, Jena, Premchand, Balaji, France, Cell, Cuisine, Cold, Yakoob, Lupin, Hong, Southern, Sarma, Saroj, Supriya, Semi, Srivastava, Mir, Case, Karthikeyan, Salt, Sagar, Markov, Mishra, Sofi, Gautam, Pareek, Manna, Springer, Kar, Koul]

Sustainability 2024 , 16 , 9437 26 of 29 4 Liakos, K.G.; Busato, P.; Moshou, D.; Pearson, S.; Bochtis, D. Machine learning in agriculture: A review Sensors 2018 , 18 , 2674 [ CrossRef ] 5 Koul, B.; Yakoob, M.; Shah, M.P. Agricultural waste management strategies for environmental sustainability Environ. Res 2022 , 206 , 112285. [ CrossRef ] 6 Mandal, S.; Choudhury, B.; Satpati, L. Monsoon variability, crop water requirement, and crop planning for kharif rice in Sagar Island, India Int. J. Biometeorol 2015 , 59 , 1891–1903. [ CrossRef ] 7 Xu, D.; Duan, X.; Wang, B.; Hong, B.; Ho, T.H.D.; Wu, R. Expression of a late embryogenesis abundant protein gene, HVA 1, from barley confers tolerance to water deficit and salt stress in transgenic rice Plant Physiol 1996 , 110 , 249–257. [ CrossRef ] 8 Ramadhan, A.J.; Tiwari, A.K.; Kumar, B.; Supriya, S.; Mishra, H.; Gautam, S.; Gautam, R.; Abotaleb, M.; Alkattan, H. Comparative Economics of Maize Crop in Kharif and Rabi Season. In BIO Web of Conferences ; EDP Sciences: Hulis, France, 2024; Volume 97, p. 00134 9 Mir, A.H.; Bhat, M.A.; Dar, S.A.; Sofi, P.A.; Bhat, N.A.; Mir, R.R. Assessment of cold tolerance in chickpea ( Cicer spp.) grown under cold/freezing weather conditions of North-Western Himalayas of Jammu and Kashmir, India Physiol. Mol. Biol. Plants 2021 , 27 , 1105–1118. [ CrossRef ] 10 Chilakala, A.R.; Pandey, P.; Durgadevi, A.; Kandpal, M.; Patil, B.S.; Rangappa, K.; Reddy, P.C.O.; Ramegowda, V.; Senthil-Kumar, M. Drought attenuates plant responses to multiple rhizospheric pathogens: A study on a dry root rot-associated disease complex in chickpea fields Field Crop. Res 2023 , 298 , 108965. [ CrossRef ] 11 Khan, A.; Singh, A.V.; Pareek, N.; Arya, P.; Upadhayay, V.K.; Kumar Jugran, A.; Kumar Mishra, P.; Goel, R. Credibility assessment of cold adaptive Pseudomonas jesenni MP 1 and P. palleroniana N 26 on growth, rhizosphere dynamics, nutrient status, and yield of the kidney bean cultivated in Indian Central Himalaya Front. Plant Sci 2023 , 14 , 1042053. [ CrossRef ] 12 Jeevarathinam, G.; Chelladurai, V. Pigeon pea. In Pulses: Processing and Product Development ; Springer: Cham, Switzerland, 2020; pp. 275–296 13 Nasir, M.; Sidhu, J.S.; Sogi, D.S. Processing and nutritional profile of mung bean, black gram, pigeon pea, lupin, moth bean, and Indian vetch. In Dry Beans and Pulses: Production, Processing, and Nutrition ; Wiley: Hoboken, NJ, USA, 2022; pp. 431–452 14 Parker, T.A.; Lo, S.; Gepts, P. Pod shattering in grain legumes: Emerging genetic and environment-related patterns Plant Cell 2021 , 33 , 179–199. [ CrossRef ] [ PubMed ] 15 Rahman, M.M.; Bakr, M.; Mia, M.; Idris, K.; Gowda, C.; Kumar, J.; Deb, U.; Malek, M.; Sobhan, A Legumes in Bangladesh ; International Crops Research Institute for the Semi-Arid Tropics: Hyderabad, India, 2000 16 Saroj, P.; Kumar, R. Recent advances in pomegranate production in India—A review Ann. Hortic 2019 , 12 , 1–10. [ CrossRef ] 17 Sarma, U.; Govila, V.K.; Yadav, A. The traditional and therapeutic use of banana and its plant based delicacies in ethnic Assamese cuisine and religious rituals from Northeast India J. Ethn. Foods 2020 , 7 , 21. [ CrossRef ] 18 Jena, R.C.; Agarwal, K.; Chand, P.K. Fruit and leaf diversity of selected Indian mangoes ( Mangifera indica L.) Sci. Hortic 2021 , 282 , 109941. [ CrossRef ] 19 Kumar, M.; Ponnuswami, V.; Kumar, P.J.; Saraswathy, S. Influence of season affecting flowering and physiological parameters in mango Sci. Res. Essays 2014 , 9 , 1–6. [ CrossRef ] 20 Priyanka, R.; Nagendran, K.; Aravintharaj, R.; Balaji, C.; Mohankumar, S.; Renukadevi, P.; Karthikeyan, G. Characterization and management of watermelon bud necrosis virus infecting watermelon in India Eur. J. Plant Pathol 2019 , 153 , 759–770. [ CrossRef ] 21 Kaur, A.; Sharma, M.; Manan, J. Comparative performance of muskmelon ( Cucumis melo ) hybrids at farmers’ field in district Kapurthala J. Krishi Vigyan 2017 , 6 , 24–31. [ CrossRef ] 22 Zahoor, S.; Dutt, V.; Mughal, A.; Pala, N.A.; Qaisar, K.; Khan, P. Apple-based agroforestry systems for biomass production and carbon sequestration: Implication for food security and climate change contemplates in temperate region of Northern Himalaya, India Agrofor. Syst 2021 , 95 , 367–382. [ CrossRef ] 23 Gupta, G.; George, P. Profitability of Nagpur santra (oranges) cultivation Indian J. Agric. Econ 1974 , 29 , 134–142 24 Koul, B.; Pudhuvai, B.; Sharma, C.; Kumar, A.; Sharma, V.; Yadav, D.; Jin, J.O Carica papaya L.: A tropical fruit with benefits beyond the tropics Diversity 2022 , 14 , 683. [ CrossRef ] 25 Premchand, U.; Mesta, R.K.; Devappa, V.; Basavarajappa, M.P.; Venkataravanappa, V.; Narasimha Reddy, L.R.C.; Shankarappa, K.S. Survey, detection, characterization of papaya ringspot virus from southern India and management of papaya ringspot disease Pathogens 2023 , 12 , 824. [ CrossRef ] [ PubMed ] 26 Devi, M.; Ghatani, K. The use of coconut in rituals and food preparations in India: A review J. Ethn. Foods 2022 , 9 , 37. [ CrossRef ] 27 Gothandan, G.; Selvi, B.S.; Velmurugan, S.; Meenakshi, P. Performance of coriander genotypes ( Coriandrum sativum L.) for herbage yield and quality parameters under coconut during Kharif season Pharma Innov. J 2021 , 10 , 1324–1328 28 Barman, D.; Kar, G.; Mitra, S.; Alam, N.; Manna, T.; Kheroar, S.; Zaman, A.; Jena, S. Dry and wet spell probability analysis using Markov chain model for planning jute-based cropping systems in eastern India Indian J. Soil Conserv 2024 , 52 , 1–8 29 Kar, G.; Blaise, D.; Srivastava, T.; Kar, C.S.; Verma, P.; Reddy, A.; Singh, P. Commercial Crops (Jute, Cotton and Sugarcane). In Trajectory of 75 Years of Indian Agriculture After Independence ; Springer: Berlin/Heidelberg, Germany, 2023; pp. 331–362 30 VA, N.A.; Panakaje, N. Growth of Coffee Cultivation, Consumption & Production in India—An Overview Int. J. Case Stud. Bus IT Educ. (IJCSBE) 2022 , 6 , 755–770 31 Kamal, B.; Acharya, B.; Srivastava, A.; Pandey, M. Effect of different altitudes in qualitative and quantitative attributes of green coffee beans ( Coffea arabica ) in Nepal Int. J. Hortic. Agric. Food Sci 2021 , 5 , 1–7.

[Find the meaning and references behind the names: De Caro, Zhang, Bajpai, Vallejo, Daily, Soufan, Angle, Jafri, Saleh, Aller, Xiong, Raja, Buchanan, Press, Bennett, Sark, Vijayalakshmi, Gebhardt, Bristol, Gallego, Plaza, Fill, Saha, Sahin, Mix, Islam, Zhao, Anjos, Zulfiqar, Math, Road, Afr, Olson, Hall, Datta, Tariq, Gonzalez, Paddy, Bani, Criado, Panel, Caro, Boca, Shearer, Brockman, Hunter, Correa, Aceh, Patino, Cannarozzo, Asia, Sens, Osei, Sun, Major, Hammoud, Rep, Sabagh, Sawicka, Reg, Agron, Allred, Mach, Gunawan, Horta, Raton, Pac, Zarei, Reed, Provenzano, Prod, Shrivastava, Shehadeh, Ciraolo, Benjeddou, Till, Sugianto, Dhaka, Chopra, Ghorbani, Taghizadeh, Chen, Masoudi, Fulton, Mech, Jayasinghe, Basri, Zadeh, Ippolito, Isik, Yan, Anderson, Delgado, Sya, Yang, Stamenkovic, Qiu, Florida, Fayaz, Palacios]

Sustainability 2024 , 16 , 9437 27 of 29 32 Khanal, S.; Fulton, J.; Klopfenstein, A.; Douridas, N.; Shearer, S. Integration of high resolution remotely sensed data and machine learning techniques for spatial prediction of soil properties and corn yield Comput. Electron. Agric 2018 , 153 , 213–225. [ CrossRef ] 33 Olson, D.; Anderson, J. Review on unmanned aerial vehicles, remote sensors, imagery processing, and their applications in agriculture Agron. J 2021 , 113 , 971–992. [ CrossRef ] 34 Weatherington-Rice, J.; Aller, L.; Bennett, T.; Hunter, D.; Fausey, N.; Allred, B.; Tornes, L.; Brockman, C.; Angle, M.; Hall, G.; et al The use of long term, large scale pumping data to determine the regional recharge rates through fractured till and lacustrine materials. In Groundwater 2000 ; CRC Press: Boca Raton, FL, USA, 2020; pp. 515–516 35 Plaza, J.; Criado, M.; Sánchez, N.; Pérez-Sánchez, R.; Palacios, C.; Charfolé, F. UAV multispectral imaging potential to monitor and predict agronomic characteristics of different forage associations Agronomy 2021 , 11 , 1697. [ CrossRef ] 36 Liu, J.; Yang, K.; Tariq, A.; Lu, L.; Soufan, W.; El Sabagh, A. Interaction of climate, topography and soil properties with cropland and cropping pattern using remote sensing data and machine learning methods Egypt. J. Remote Sens. Space Sci 2023 , 26 , 415–426 [ CrossRef ] 37 Zhang, L.; Wu, Z.; Sun, X.; Yan, J.; Sun, Y.; Liu, P.; Chen, J. Mapping topsoil total nitrogen using random forest and modified regression Kriging in Agricultural Areas of Central China Plants 2023 , 12 , 1464. [ CrossRef ] [ PubMed ] 38 Raja, S.; Sawicka, B.; Stamenkovic, Z.; Mariammal, G. Crop prediction based on characteristics of the agricultural environment using various feature selection techniques and classifiers IEEE Access 2022 , 10 , 23625–23641. [ CrossRef ] 39 Rahman, S.A.Z.; Mitra, K.C.; Islam, S.M. Soil classification using machine learning methods and crop suggestion based on soil series. In Proceedings of the 2018 21 st International Conference of Computer and Information Technology (ICCIT), Dhaka, Bangladesh, 21–23 December 2018; pp. 1–4 40 Maesa, S.; Basri, H.; Sugianto, S. Classification of Rice Growth Phases Using the K-Nearest Neighbor Algorithm in the Irrigation area of Seulimeum Sub District, Aceh, Indonesia. In IOP Conference Series: Earth and Environmental Science ; IOP Publishing: Bristol, UK, 2024; Volume 1297, p. 012006 41 Wu, C.; Wang, M.; Wang, C.; Zhao, X.; Liu, Y.; Masoudi, A.; Yu, Z.; Liu, J. Reed biochar improved the soil functioning and bacterial interactions: A bagging experiment using the plantation forest soil ( Fraxinus chinensis ) in the Xiong’an new area, China J. Clean Prod 2023 , 410 , 137316. [ CrossRef ] 42 Devyatkin, D.A. Estimation of vegetation indices with Random Kernel Forests IEEE Access 2023 , 11 , 29500–29509. [ CrossRef ] 43 Yadav, J.; Chopra, S.; Vijayalakshmi, M. Soil analysis and crop fertility prediction using machine learning Mach. Learn 2021 , 8 , 41–49 44 Gunawan, G.; Sya’bani, A.Z.; Anandianshka, S. Expert system for diagnosing diseases in corn plants using the navies bayes method J. Mantik 2024 , 8 , 849–859. [ CrossRef ] 45 Biazar, S.M.; Shehadeh, H.A.; Ghorbani, M.A.; Golmohammadi, G.; Saha, A. Soil temperature forecasting using a hybrid artificial neural network in florida subtropical Grazinglands agro-ecosystems Sci. Rep 2024 , 14 , 1535. [ CrossRef ] 46 Qiu, J.; Tabasi, E.; Hammoud, A.; Benjeddou, O.; Zarei, M.; Khordehbinan, M.W. Determining the fracture stiffness of modified Hot and Warm Mix Asphalt using semi-circular bending (SCB) geometry Theor. Appl. Fract. Mech 2024 , 129 , 104237. [ CrossRef ] 47 Ebrahimzadeh, E.; Fayaz, F.; Rajabion, L.; Seraji, M.; Aflaki, F.; Hammoud, A.; Taghizadeh, Z.; Asgarinejad, M.; Soltanian-Zadeh, H. Machine learning approaches and non-linear processing of extracted components in frontal region to predict rTMS treatment response in major depressive disorder Front. Syst. Neurosci 2023 , 17 , 919977. [ CrossRef ] 48 Wickramasinghe, L.; Weliwatta, R.; Ekanayake, P.; Jayasinghe, J. Modeling the relationship between rice yield and climate variables using statistical and machine learning techniques J. Math 2021 , 2021 , 6646126. [ CrossRef ] 49 Chandrasiri, C.K.; Tsusaka, T.W.; Ho, T.D.; Zulfiqar, F.; Datta, A. Impacts of climate change on paddy yields in different climatic zones of Sri Lanka: A panel data approach Asia-Pac. J. Reg. Sci 2023 , 7 , 455–489. [ CrossRef ] 50 Gatera, A.; Kuradusenge, M.; Bajpai, G.; Mikeka, C.; Shrivastava, S. Comparison of random forest and support vector machine regression models for forecasting road accidents Sci. Afr 2023 , 21 , e 01739. [ CrossRef ] 51 Sahin, G.; Isik, G.; van Sark, W.G. Predictive modeling of PV solar power plant efficiency considering weather conditions: A comparative analysis of artificial neural networks and multiple linear regression Energy Rep 2023 , 10 , 2837–2849. [ CrossRef ] 52 De Caro, D.; Ippolito, M.; Cannarozzo, M.; Provenzano, G.; Ciraolo, G. Assessing the performance of the Gaussian Process Regression algorithm to fill gaps in the time-series of daily actual evapotranspiration of different crops in temperate and continental zones using ground and remotely sensed data Agric. Water Manag 2023 , 290 , 108596. [ CrossRef ] 53 Yang, H.; Gebhardt, W.; Ororbia, A.G.; Desell, T. Predicted Embedding Power Regression for Large-Scale Out-of-Distribution Detection arXiv 2023 , arXiv:2303.04115 54 Diaz-Gonzalez, F.A.; Vuelvas, J.; Correa, C.A.; Vallejo, V.E.; Patino, D. Machine learning and remote sensing techniques applied to estimate soil indicators–review Ecol. Indic 2022 , 135 , 108517. [ CrossRef ] 55 García-López, A.M.; Delgado, A.; Anjos, O.; Horta, C. Digestate not only affects nutrient availability but also soil quality indicators Agronomy 2023 , 13 , 1308. [ CrossRef ] 56 Burdett, H.; Wellen, C. Statistical and machine learning methods for crop yield prediction in the context of precision agriculture Precis. Agric 2022 , 23 , 1553–1574. [ CrossRef ] 57 Osei, E.; Jafri, S.H.; Saleh, A.; Gassman, P.W.; Gallego, O. Simulated climate change impacts on corn and soybean yields in Buchanan County, Iowa Agriculture 2023 , 13 , 268. [ CrossRef ]

[Find the meaning and references behind the names: De Oliveira, Solaiman, Porras, Krok, Shukla, Fish, Wagle, Chandana, Siddique, Mahale, Waters, Shade, Grey, Omar, Govind, Zeng, Ferreira, Turki, Holden, Mikha, Adv, Abdelmageed, Sandy, Fonte, Rehan, Ghimire, Great, Prado, Novel, Slaughter, Viana, Steiner, Kotecha, Godoy, Zhu, Mathur, Pang, Teixeira, Martinez, Jadon, Filho, Luo, Roman, Yao, Patel, Quan, Stat, Kulkarni, Mehrabi, Lett, Feng, Kim, Ruan, Acosta, Thapa, Harrison, Hou, Srinivasarao, Venu, Aguilera, Sengupta, Washington, Bista, Tari, Shen, Shi, Cheng, Oliveira, Gava, Wolf]

Sustainability 2024 , 16 , 9437 28 of 29 58 Yang, Z.; Yang, W.; Yi, J.; Liu, R. WH-DETR: An Efficient Network Architecture for Wheat Spike Detection in Complex Backgrounds Agriculture 2024 , 14 , 961. [ CrossRef ] 59 Tang, L.; Yi, J.; Li, X. Improved multi-scale inverse bottleneck residual network based on triplet parallel attention for apple leaf disease identification J. Integr. Agric 2024 , 23 , 901–922. [ CrossRef ] 60 Sidhu, B.S.; Mehrabi, Z.; Ramankutty, N.; Kandlikar, M. How can machine learning help in understanding the impact of climate change on crop yields? Environ. Res. Lett 2023 , 18 , 024008. [ CrossRef ] 61 Li, L.; Zhang, Y.; Wang, B.; Feng, P.; He, Q.; Shi, Y.; Liu, K.; Harrison, M.T.; Li Liu, D.; Yao, N.; et al. Integrating machine learning and environmental variables to constrain uncertainty in crop yield change projections under climate change Eur. J. Agron 2023 , 149 , 126917. [ CrossRef ] 62 Holden, T.D. Existence and uniqueness of solutions to dynamic models with occasionally binding constraints Rev. Econ. Stat 2023 , 105 , 1481–1499. [ CrossRef ] 63 Li, L.; Wang, B.; Feng, P.; Jägermeyr, J.; Asseng, S.; Müller, C.; Macadam, I.; Liu, D.L.; Waters, C.; Zhang, Y.; et al. The optimization of model ensemble composition and size can enhance the robustness of crop yield projections Commun. Earth Environ 2023 , 4 , 362. [ CrossRef ] 64 Motia, S.; Reddy, S. Exploration of machine learning methods for prediction and assessment of soil properties for agricultural soil management: A quantitative evaluation. In Journal of Physics: Conference Series ; IOP Publishing: Bristol, UK, 2021; Volume 1950, p. 012037 65 Ghimire, R.; Thapa, V.R.; Acosta-Martinez, V.; Schipanski, M.; Slaughter, L.C.; Fonte, S.J.; Shukla, M.K.; Bista, P.; Angadi, S.V.; Mikha, M.M.; et al. Soil health assessment and management framework for water-limited environments: Examples from the Great Plains of the USA Soil Syst 2023 , 7 , 22. [ CrossRef ] 66 Jhajharia, K.; Mathur, P. Prediction of crop yield using satellite vegetation indices combined with machine learning approaches Adv. Space Res 2023 , 72 , 3998–4007. [ CrossRef ] 67 Pokhariyal, S.; Patel, N.; Govind, A. Machine Learning-Driven Remote Sensing Applications for Agriculture in India—A Systematic Review Agronomy 2023 , 13 , 2302. [ CrossRef ] 68 Malik, P.; Sengupta, S.; Jadon, J.S. Comparative analysis of soil properties to predict fertility and crop yield using machine learning algorithms. In Proceedings of the 2021 11 th International Conference on Cloud Computing, Data Science & Engineering (Confluence), Noida, India, 28–29 January 2021; pp. 1004–1007 69 United States Department of Agriculture India Production Summary ; United States Department of Agriculture: Washington, DC, USA, 2024 70 Mahale, Y.; Khan, N.; Kulkarni, K.; Wagle, S.A.; Pareek, P.; Kotecha, K.; Choudhury, T.; Sharma, A. Crop recommendation and forecasting system for Maharashtra using machine learning with LSTM: A novel expectation-maximization technique Discov Sustain 2024 , 5 , 134. [ CrossRef ] 71 Staniak, M.; Szpunar-Krok, E.; Kocira, A. Responses of soybean to selected abiotic stresses—Photoperiod, temperature and water Agriculture 2023 , 13 , 146. [ CrossRef ] 72 Rehan, M.; Al-Turki, A.; Abdelmageed, A.H.; Abdelhameid, N.M.; Omar, A.F. Performance of plant-growth-promoting rhizobacteria (PGPR) isolated from sandy soil on growth of tomato ( Solanum lycopersicum L.) Plants 2023 , 12 , 1588. [ CrossRef ] [ PubMed ] 73 Ratke, R.F.; Zuffo, A.M.; Steiner, F.; Aguilera, J.G.; de Godoy, M.L.; Gava, R.; de Oliveira, J.T.; Filho, T.A.d.S.; Viana, P.R.N.; Ratke, L.P.T.; et al. Can soil moisture and crop production be influenced by different cropping systems? AgriEngineering 2023 , 5 , 112–126 [ CrossRef ] 74 Wang, Y.J.; Yang, T.; Kim, H.J. pH dynamics in aquaponic systems: Implications for plant and fish crop productivity and yield Sustainability 2023 , 15 , 7137. [ CrossRef ] 75 Zhang, H.; Wang, L.; Fu, W.; Xu, C.; Zhang, H.; Xu, X.; Ma, H.; Wang, J.; Zhang, Y. Soil Acidification Can Be Improved under Different Long-Term Fertilization Regimes in a Sweetpotato–Wheat Rotation System Plants 2024 , 13 , 1740. [ CrossRef ] 76 Chen, M.; Shen, Y.; Wang, H.; Cheng, X.; Luo, Y. Analysis of the Rainfall Pattern and Rainfall Utilization Efficiency during the Growth Period of Paddy Rice Agronomy 2024 , 14 , 1332. [ CrossRef ] 77 Hlisnikovsk `y, L.; Menšík, L.; Roman, M.; Kunzová, E. The Evaluation of a Long-Term Experiment on the Relationships between Weather, Nitrogen Fertilization, Preceding Crop, and Winter Wheat Grain Yield on Cambisol Plants 2024 , 13 , 802. [ CrossRef ] 78 Zeng, J.; Liu, F.; Zhu, Y.; Li, J.; Ruan, Y.; Quan, X.; Wang, C.; Wang, X. Degree of shade tolerance shapes seasonality of chlorophyll, nitrogen and phosphorus levels of trees and herbs in a temperate deciduous forest J. For. Res 2024 , 35 , 72. [ CrossRef ] 79 Islam, M.; Siddique, K.H.; Padhye, L.P.; Pang, J.; Solaiman, Z.M.; Hou, D.; Srinivasarao, C.; Zhang, T.; Chandana, P.; Venu, N.; et al. A critical review of soil phosphorus dynamics and biogeochemical processes for unlocking soil phosphorus reserves Adv Agron 2024 , 185 , 153–249 80 Gonzalez-Porras, C.V.; Teixeira, G.C.M.; Prado, R.d.M.; Ferreira, P.M.; Palaretti, L.F.; Oliveira, K.S. Silicon via fertigation with and without potassium application, improve physiological aspects of common beans cultivated under three water regimes in field Sci. Rep 2024 , 14 , 2051. [ CrossRef ] 81 Dabba, A.; Tari, A.; Meftali, S. A novel grey wolf optimization algorithm based on geometric transformations for gene selection and cancer classification J. Supercomput 2024 , 80 , 4808–4840. [ CrossRef ]

[Find the meaning and references behind the names: Eng, Jogi, Law, Zou, Gap, Beltran, Gomez, Acm, Chenu, Silva, Henna, Arch, Basso, Bracho, Bro, Haakana, Gil, Alwan, Foster, Herrero, Kassie, Single, Schmitt, Ayan, Mujica, Keshri, Furman, Merlos, Brand, Krishna, Kasi, Xing, Evol, Conley, Rani, Tripathi, Janssen, Henry, Tool, Aramburu, Llopis, Royo, Khaki, Tenorio, Grondin, Arab, Paudel, Tesfaye, Zheng, Carminati, Aco, Ideas, Ruiz, Chang, Pandya, Naik, Monzon, Dang, Jones, Ewert, Nat, Rathnayake, Berger, Legal, Zek, Segarra, Jakubiec, Box, Chou, Gayo]

Sustainability 2024 , 16 , 9437 29 of 29 82 Tripathi, A.; Rani, P. An improved mser using grid search based pca and ensemble voting technique Multimed. Tools Appl 2024 , 83 , 80497–80522. [ CrossRef ] 83 Wang, X.; Jin, Y.; Schmitt, S.; Olhofer, M. Recent advances in Bayesian optimization ACM Comput. Surv 2023 , 55 , 1–36. [ CrossRef ] 84 Ramadevi, B.; Kasi, V.R.; Bingi, K. Fractional ordering of activation functions for neural networks: A case study on Texas wind turbine Eng. Appl. Artif. Intell 2024 , 127 , 107308. [ CrossRef ] 85 Blekos, K.; Brand, D.; Ceschini, A.; Chou, C.H.; Li, R.H.; Pandya, K.; Summer, A. A review on quantum approximate optimization algorithm and its variants Phys. Rep 2024 , 1068 , 1–66. [ CrossRef ] 86 Alhijawi, B.; Awajan, A. Genetic algorithms: Theory, genetic operators, solutions, and applications Evol. Intell 2024 , 17 , 1245–1256. [ CrossRef ] 87 Ayan, E. Genetic algorithm-based hyperparameter optimization for convolutional neural networks in the classification of crop pests Arab. J. Sci. Eng 2024 , 49 , 3079–3093. [ CrossRef ] 88 Ruiz-Gonzalez, R.; Gomez-Gil, J.; Gomez-Gil, F.J.; Martínez-Martínez, V. An SVM-based classifier for estimating the state of various rotating components in agro-industrial machinery with a vibration signal acquired from a single point on the machine chassis Sensors 2014 , 14 , 20713–20735. [ CrossRef ] 89 Silva, J.V.; van Heerwaarden, J.; Reidsma, P.; Laborte, A.G.; Tesfaye, K.; van Ittersum, M.K. Big data, small explanatory and predictive power: Lessons from random forest modeling of on-farm yield variability and implications for data-driven agronomy Field Crop. Res 2023 , 302 , 109063. [ CrossRef ] 90 Beltran-Royo, C.; Llopis-Ibor, L.; Pantrigo, J.J.; Ramírez, I. DC Neural Networks avoid overfitting in one-dimensional nonlinear regression Knowl.-Based Syst 2024 , 283 , 111154. [ CrossRef ] 91 Makumbura, R.K.; Mampitiya, L.; Rathnayake, N.; Meddage, D.; Henna, S.; Dang, T.L.; Hoshino, Y.; Rathnayake, U. Advancing water quality assessment and prediction using machine learning models, coupled with explainable artificial intelligence (XAI) techniques like shapley additive explanations (SHAP) for interpreting the black-box nature Results Eng 2024 , 23 , 102831 [ CrossRef ] 92 Berger, T. Agent-based spatial models applied to agriculture: A simulation tool for technology diffusion, resource use changes and policy analysis Agric. Econ 2001 , 25 , 245–260 93 Jones, J.W.; Antle, J.M.; Basso, B.; Boote, K.J.; Conant, R.T.; Foster, I.; Godfray, H.C.J.; Herrero, M.; Howitt, R.E.; Janssen, S.; et al. Brief history of agricultural systems modeling Agric. Syst 2017 , 155 , 240–254. [ CrossRef ] [ PubMed ] 94 Tenorio, F.A.; Edreira, J.I.R.; Monzon, J.P.; Aramburu-Merlos, F.; Dobermann, A.; Gruere, A.; Brihet, J.M.; Gayo, S.; Conley, S.; Mourtzinis, S.; et al. Filling the agronomic data gap through a minimum data collection approach Field Crop. Res 2024 , 308 , 109278. [ CrossRef ] [ PubMed ] 95 Bro ˙zek, B.; Furman, M.; Jakubiec, M.; Kucharzyk, B. The black box problem revisited. Real and imaginary challenges for automated legal decision making Artif. Intell. Law 2024 , 32 , 427–440. [ CrossRef ] 96 Khaki, S.; Wang, L. Crop yield prediction using deep neural networks Front. Plant Sci 2019 , 10 , 621. [ CrossRef ] 97 Vadez, V.; Grondin, A.; Chenu, K.; Henry, A.; Laplaze, L.; Millet, E.J.; Carminati, A. Crop traits and production under drought Nat. Rev. Earth Environ 2024 , 5 , 211–225. [ CrossRef ] 98 Naik, A.; Jogi, M.; Shreenivas, B. Assessing the impact of climate change on global crop yields and farming practices Arch. Curr Res. Int 2024 , 24 , 696–712. [ CrossRef ] 99 Amertet Finecomess, S.; Gebresenbet, G.; Alwan, H.M. Utilizing an Internet of Things (IoT) Device, Intelligent Control Design, and Simulation for an Agricultural System IoT 2024 , 5 , 58–78. [ CrossRef ] 100. Osinga, S.A.; Paudel, D.; Mouzakitis, S.A.; Athanasiadis, I.N. Big data in agriculture: Between opportunity and solution Agric Syst 2022 , 195 , 103298. [ CrossRef ] 101. Shrestha, A.D Enhanced Deep Network Designs Using Mitochondrial DNA Based Genetic Algorithm and Importance Sampling ; University of Bridgeport: Bridgeport, CT, USA, 2019 102. Kao, S.C.; Krishna, T. Gamma: Automating the hw mapping of dnn models on accelerators via genetic algorithm. In Proceedings of the 39 th International Conference on Computer-Aided Design, Virtual, 2–5 November 2020; pp. 1–9 103. Bracho-Mujica, G.; Rötter, R.P.; Haakana, M.; Palosuo, T.; Fronzek, S.; Asseng, S.; Yi, C.; Ewert, F.; Gaiser, T.; Kassie, B.; et al Effects of changes in climatic means, variability, and agro-technologies on future wheat and maize yields at 10 sites across the globe Agric. For. Meteorol 2024 , 346 , 109887. [ CrossRef ] 104. Segarra, J. Satellite Imagery in Precision Agriculture. In Digital Agriculture: A Solution for Sustainable Food and Nutritional Security ; Springer: Berlin/Heidelberg, Germany, 2024; pp. 325–340 105. Wu, B.; Gommes, R.; Zhang, M.; Zeng, H.; Yan, N.; Zou, W.; Zheng, Y.; Zhang, N.; Chang, S.; Xing, Q.; et al. Global crop monitoring: A satellite-based hierarchical approach Remote Sens 2015 , 7 , 3907–3933. [ CrossRef ] 106. Keshri, R.; Vidyarthi, D.P. Energy-efficient communication-aware VM placement in cloud datacenter using hybrid ACO–GWO Clust. Comput 2024 , 27 , 13047–13074. [ CrossRef ] Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Other Environmental Sciences Concepts:

[back to top]

Discover the significance of concepts within the article: ‘Predicting Sustainable Crop Yields: Deep Learning and Explainable AI Tools’. Further sources in the context of Environmental Sciences might help you critically compare this page with similair documents:

Ragi, Jute, Urad, Nitrogen, Rice, Bajra, Jowar, Coffea arabica, Lime, Wheat, Mung bean, Pigeon pea, Maize, Sugarcane, Sunflower, Potato, Coconut, Small millet, Common bean, Moth bean, Sesamum, Cucumis melo, Artificial intelligence, Religious ritual, Castor-seed, Food preparation, Mangoes, Soil quality, Historical data, Rainfall, Decision making, Crop-cultivation, Environmental factor, Accuracy, Cross validation, Climate change, Crop growth, Environmental condition, Northeast India, Developing countries, Agricultural productivity, Physiological aspect, Food Security, Microbial activity, Climatic condition, Climate condition, Machine Learning, Deep Learning, Training Data, Soil analysis, Amino acid, Ambient temperature, Phosphorus, Positive correlation, Learning rate, Physiological parameter, Soil Organic Matter, Plant growth, Sustainable agriculture, Soil fertility, Soil classification, Plant growth promoting rhizobacteria, Multiple linear regression, Support Vector Machine, Artificial Neural Network, Mobile technology, Energy transfer, Nutrient availability, Agricultural sustainability, Nutrient Status, Statistical model, Water quality assessment, Optimal Growth, Cation Exchange Capacity, Soil pH, Crop management, Computational power, Drought stress, Predictive Model, Crop yield, Test set, Environmental parameters, Chlorophyll, Linear regression, Plant health, Soil properties, Sustainability, Disease risk, Tropical fruit, Potassium, Crop pests, Remote sensing, Soil parameters, Kharif season, Big Data, Carbon Sequestration, Precision agriculture, Cotton cultivation, Rabi season, Mangifera Indica L., Soil Health, Coriandrum sativum L., Water uptake, Nutrient rich soil, Extreme conditions, Rice yield, Prediction Accuracy, Agronomic characteristics, Policy formulation, Humidity, Nutrient transport, Rainfall pattern, Remote Monitoring, Decision tree, Waterlogging, Cropping pattern, Green coffee bean, Data integration, Predictive modeling, Open access article, Creative Commons Attribution, Commercial crop, Vegetative Growth, Training and testing sets, Road accident, Random Forest, Predictive accuracy, Root development, Agricultural practice, Machine learning approaches, Expert system, Neural Network, Modern agriculture, Watermelon, Real time data, Agricultural system, Sustainable farming practices, Loss function, Hyperparameter tuning, Root vegetable, Legume, Satellite imagery, Data availability, Model prediction, Trend, Climate variability, Enzyme activation, Soil condition, Confusion matrices, Climate data, Hierarchical approach, Climatic variables, Decision support tool, Machine learning model, Irrigation management, Climate change impact, Remote sensing data, Geographic coordinates, Environmental variable, Abiotic stresses, Plant development, Agroforestry system, Dynamic model, Data preprocessing, Explainable AI, Water use efficiency, Nitrogen fertilization, Academic Editor, Evolutionary algorithm, ML techniques, Target variable, Feature importance, Activation function, Seasonal factor, Flower development, Cropping system, Grid search, Optimization algorithm, Comprehensive dataset, Soil acidification, Agricultural waste management, Carica papaya L, Genetic algorithm, Machine learning method, K-nearest neighbor algorithm, Sensor system, Primary Water Source, Seasonal adjustment, Deep neural network, Bayesian optimization, Feature engineering, Remote sensing technologies, Machine learning technique, Crop Yield Prediction, Climatic characteristics, Fitness function, Unmanned aerial vehicle, Classification Task, Vegetation indices, ML model, Local interpretable model-agnostic explanations, Gaussian process regression, Hyperparameter optimization, DNN, Food producer, Crop type, Crop water requirement, Agri-food sector, Soil Nutrient Levels, Agricultural factors, Temporal factor, Feature selection technique, Climatic factor, Boosted regression tree, Soil nutrient, Climate variable, Remote sensing applications, HPO, Agricultural Product, Weather condition, IoT Device, Markov Chain Model, Categorical features, Yield variability, Transpiration, Leafy crops, Corn yield, Sensor node, One-Hot Encoding, Data Processing Technologies, Soil quality indicator, ML algorithm, XAI technique, Shapley additive explanation, Semi-arid tropics, Model Interpretation, Irrigation strategies, Soil indicators, Categorical cross-entropy, Soil health assessment.