Key points are not available for this paper at this time.
The accuracy of machine learning models in predicting soil water content, actual evapotranspiration (ETa), and biomass, as well as the development of a regression model between the Normalized Difference Vegetation Index (NDVI) and adjusted crop coefficient (Kcaj), has not been adequately assessed for watermelon production in the Arba Minch region of southern Ethiopia. This study evaluated the performance of two machine learning algorithms Random Forest (RF) and Artificial Neural Network (ANN) to predict soil water content, ETa, and biomass, and to develop an NDVI-based Kcaj regression model under four irrigation treatments: 100 % irrigation (T1), 100 % irrigation with straw mulching (T2), 50 % irrigation (T3), and 50 % irrigation with straw mulching (T4). Soil water was measured using a time-domain reflectometer, and applied irrigation was monitored using partial flumes. ETa was estimated through the soil water balance method, while NDVI values were extracted from Sentinel-2 images using Google Earth Engine. Model accuracy was evaluated using R 2 , adjusted R 2 (R 2 aj), root mean square error (RMSE), Nash–Sutcliffe efficiency (NSE), mean absolute percentage error (MAPE), and median absolute error (MedAE). Results revealed that straw mulching increased soil water storage by 5.1 % under T2 and 10.9 % under T4. The ANN model outperformed RF across all variables. For soil water, ANN yielded R 2 = 0.75, RMSE = 0.36 %, NSE = 0.74; for ETa, R 2 = 0.80, RMSE = 7.15 mm, NSE = 0.81; and for biomass, R 2 = 0.92, RMSE = 0.89 ton/ha, NSE = 0.92. The NDVI-Kcaj regression followed a second-degree polynomial, with Kcaj ranging from 0.3 to 0.8 and NDVI from 0.5 to 0.7 across treatments. These findings demonstrate the potential of ANN for improving predictive irrigation scheduling and enhancing water use efficiency in semi-arid agricultural systems.
Birara Gebeyhu Reta (Thu,) studied this question.