Combining UAV remote sensing data to estimate daily-scale crop water stress index: Enhancing diagnostic temporal representativeness

Abstract

Using thermal infrared remote sensing from unmanned aerial vehicles (UAVs) to obtain crop canopy temperature and calculate the crop water stress index (CWSI) is a promising method for monitoring field water conditions. However, such endeavors are often constrained to instantaneous scales due to the diurnal variability of thermal infrared data. To address this limitation, we developed a daily-scale CWSI suitable for UAV remote sensing, enhancing the temporal representativeness of crop water stress diagnostics. We focused on spring maize in the Hetao Irrigation District of Inner Mongolia and investigated four key growth stages. UAV thermal infrared was used to obtain multiple instantaneous statistical CWSI (CWSI_s) values during the day. UAV multispectral data and the Penman–Monteith model were combined to obtain the actual evapotranspiration and daily-scale CWSI (CWSI_{t_day}). A temporal upscaling model from instantaneous CSWI to daily-scale CWSI was established by comparing the relationships between the CWSI_s and CWSI_{t_day} at different times. Results show that compared to the fluctuations of the CWSI_s values throughout the day, those of the CWSI_{t_day} values were smaller, with values of 0.13, 0.09, 0.03, and 0.03 during the ninth leaf (V9), tasseling (VT), silking (R1), and milk (R3) stages, respectively. The CWSI_{t_day} demonstrated a higher correlation with the measured stomatal conductance ($g_s$) at different time periods, thereby being more stable and temporally representative. However, both indices may incorrectly interpret the decline in leaf physiological activity due to aging as water stress at the end of maize growth, leading to overestimated CWSI values. The temporal upscaling model, which was developed by combining CWSI_s values observed at 12:00, 14:00, and 16:00 with the random forest regression algorithm, achieved coefficient of determination of 0.794 and root mean square error of 0.04. Hence, multiple instantaneous observations can be used effectively instead of daily-scale observations, providing key insights into the popularization and application of the CWSI_{t_day}. Overall, this study presents a new method for obtaining continuous CWSI values with high temporal and spatial resolutions based on a UAV platform.