Otoro et al. (2025) Integration of Machine Learning and Remote Sensing to Evaluate the Effects of Soil Salinity, Nitrate, and Moisture on Crop Yields and Economic Returns in the Semi-Arid Region of Ethiopia

Identification

• Journal: Agriculture
• Year: 2025
• Date: 2025-11-18
• Authors: Gezimu Gelu Otoro, Katsuaki KOMAI
• DOI: 10.3390/agriculture15222378

Research Groups

• Graduate School of Engineering, Kitami Institute of Technology, Kitami, Japan

Short Summary

This study integrated machine learning and remote sensing to evaluate the combined effects of soil salinity, nitrate, and moisture on crop yields and economic returns for banana, cotton, and maize in semi-arid Ethiopia. It found that soil salinity was the most critical factor reducing crop yields and economic profitability, with Random Forest models demonstrating high predictive accuracy for these outcomes.

Objective

• To estimate crop yields under varying soil salinity, moisture, and nitrate conditions in semi-arid Ethiopia.
• To assess the economic impacts of banana, maize, and cotton cultivation among smallholders in southern Ethiopia under these conditions.
• To identify key predictors and algorithms for sustainable land management in salinity-affected semi-arid regions.

Study Configuration

• Spatial Scale: Sille irrigation scheme, southern Ethiopia (37°28′30″ to 37°30′50″ E and 5°50′0″ to 5°54′10″ N, altitude 1118 m). The scheme covers three villages (Eligo, Cahfe, and Mage).
• Temporal Scale: In-situ data collected from 10 October 2023 to 20 October 2023. Satellite images collected between 10 October 2023 and 30 January 2024, corresponding to the main cropping season.

Methodology and Data

• Models used: Ridge Regression (RR), Decision Tree Regressor (DT), Random Forest Regressor (RF), Gradient Boosting Regressor (GB), Support Vector Regressor (SV), K-Nearest Neighbors Regressor (kNN), Multi-Layer Perceptron (MLP) Regressor. SHapley Additive exPlanations (SHAP) for feature importance.
• Data sources:
  • In-situ observations: Soil moisture (gravimetric method, %), soil salinity (EC in dS/cm, pH, TDS in mg/L), soil nitrate (NO3- in mg/L), and crop yield (kg/m^2, converted to tons/hectare) for banana, cotton, and maize.
  • Remote sensing: Landsat 8, MODIS, and Sentinel-2 satellite images. Derived indices include Normalized Difference Vegetation Index (NDVI), Normalized Differential Salinity Index (NDSI), and Normalized Difference Moisture Index (NDMI).
  • Ancillary data: Arba Minch Meteorological Station (AMS) data for temperature (17 to 33 °C) and annual precipitation (less than 850 mm). Farm-gate prices (USD/kg) for economic loss calculation.

Main Results

• Crop Yield and Soil Variability: Significant variability was observed in crop yields (cotton: 0.14–2.05 tons/hectare; maize: 2.53–6.17 tons/hectare; banana: 4.85–8.71 tons/hectare) and soil conditions (soil moisture: 0.08–0.51; soil nitrate: 1.05–21.67 mg/L; soil salinity: 7.63–25.61 dS/cm).
• Impact of Stressors: Higher soil salinity negatively impacted crop yields, while sufficient soil moisture and nitrate improved productivity. Salinity was identified as the dominant limiting factor.
• Model Performance for Yield Prediction: Random Forest (RF) consistently performed well, achieving high accuracy for cotton (R^2 = 0.998), banana (R^2 = 0.922), and maize (R^2 = 0.793). Gradient Boosting (GB) also showed strong performance for maize (R^2 = 0.781) and banana (R^2 = 0.901).
• Model Performance for Economic Loss Prediction: RF, Ridge Regression (RR), and GB models achieved high accuracy in estimating economic losses for cotton and maize. GB showed the highest accuracy for banana economic loss (R^2 = 0.901). Support Vector Regressor (SV), K-Nearest Neighbors (KNN), and Multi-Layer Perceptron (MLP) consistently underperformed for economic predictions.
• Feature Importance (SHAP Analysis): Soil salinity was the most influential factor for both crop yield and economic loss across all crops (e.g., mean SHAP values for salinity in yield prediction: cotton 0.378, maize 0.769, banana 0.575; for economic loss prediction: cotton 213.4, maize 291.4, banana 277.8). Soil moisture and nitrate also contributed but were less influential than salinity.

Contributions

• Addresses critical gaps in literature by focusing on salinity-affected semi-arid regions in Ethiopia, quantifying productivity gaps for banana, maize, and cotton under combined stresses, and providing economic evaluations for smallholders.
• Integrates advanced machine learning algorithms with remote sensing data to provide robust predictions of crop yields and economic losses under varying soil conditions.
• Identifies soil salinity as the primary determinant of yield reduction and economic loss, highlighting the urgent need for targeted salinity mitigation strategies.
• Offers actionable insights for farmers and decision-makers to enhance crop productivity, improve livelihoods, and strengthen food security in vulnerable semi-arid agricultural systems.

Funding

• American Society of Mechanical Engineers (ASME), Environmental Systems Division (ESD) Education Support Program (Grant No. 222009).
• Japanese Government (MEXT) Scholarship (for G.G.O.).

Citation

@article{Otoro2025Integration,
  author = {Otoro, Gezimu Gelu and KOMAI, Katsuaki},
  title = {Integration of Machine Learning and Remote Sensing to Evaluate the Effects of Soil Salinity, Nitrate, and Moisture on Crop Yields and Economic Returns in the Semi-Arid Region of Ethiopia},
  journal = {Agriculture},
  year = {2025},
  doi = {10.3390/agriculture15222378},
  url = {https://doi.org/10.3390/agriculture15222378}
}