Bias correction of satellite based crop water stress index using machine learning methods

Abstract

Accurate estimation of the Crop Water Stress Index (CWSI) is essential for supporting irrigation scheduling in water-limited regions. Traditionally, CWSI has been computed from field-based canopy temperature and meteorological measurements; however, these approaches are often limited by high costs and sparse spatial coverage. Satellite remote sensing offers a practical alternative by providing large-scale and repeated observations of crop water stress. Nevertheless, uncertainties in satellite-derived inputs, particularly land surface temperature (LST), can introduce significant biases in CWSI computations. This study addresses the challenge of mitigating biases in the CWSI computations, utilizing satellite data within sugarcane fields in southwest Iran. For this goal, twenty-four Landsat 8/9 satellite images were acquired (June–September 2023). Concurrent with the satellite overpass times, data from 18 points across the sugarcane fields, were gathered to compute CWSI based on Idso method as a benchmark for bias correction. Field measurement includes, canopy temperature (T_c) and meteorological variables (i.e. T_min, T_max, RH_min, and RH_max) from sensor-equipped points. LST was calculated using the single-channel algorithm (SC) in google earth engine (GEE) platform. Three distinct machine learning (ML) methods, namely Random Forest (RF), Support Vector Machine (SVM), and extreme gradient boosting (XGBoost), were employed to mitigate bias in LST as main driver of CWSI based on satellite data. The results demonstrated that the RF model performed exceptionally well for LST bias correction. Overall, CWSI enhance accuracy, about with R² = 35 %, nRMSE= 47 % and rMBE= 50 %. This research highlights the effectiveness of ML methods in improving CWSI estimates based on satellite imagery and taking a big step towards calculating CWSI for optimizing irrigation management in arid regions, supporting sustainable water use and food security with minimal ground monitoring.