Evaluation of machine-learning algorithms in estimation of relative water content of sorghum under different irrigated environments

Abstract

Sorghum, a climate-resilient crop widely cultivated in arid and semi-arid regions, is crucial for food and feed security in India. Its ability to tolerate water stress makes it essential for sustainable agriculture under future climatic scenarios. However, the inherent delay in water stress symptom expression in sorghum poses challenges for timely intervention, necessitating precise monitoring methods for assessing crop water status. This study evaluates the performance of different ML algorithms in estimating the relative water content of the sorghum under irrigated and rainfed conditions. The results showed that the distinct spectral curves obtained under the two different irrigated condition. The four different ML algorithms such as RF, XGboost, SVM, PLS were trained with spectral reflectance data covering 320–1100 nm to estimate the RWC of the sorghum. Among the models, SVM (R² = 0.94, RMSE = 1.457, MSE = 2.123) demonstrated the highest accuracy followed by XGboost, RF, and PLS. Under rainfed conditions, all models exhibited reduced performance, with XGboost performing relatively better (R² = 0.70, low RMSE value of 3.707, and MSE of 13.741), while PLS performed the least (R² = 0.60, RMSE = 5.523, MSE = 30.505). This study demonstrates the potential of ML algorithms, particularly SVM and XGBoost, for precise RWC estimation and also highlighted the limitations of models under limited water condition.