Evapotranspiration and soil water content estimation of four urban landscape vegetations using UAV-based multispectral and thermal imagery
Accurate assessment of evapotranspiration along with soil water content (SWC) dynamics in a heterogeneous urban landscape is fundamental for developing effective water management practices. The unmanned aerial vehicle (UAV) remote sensing with high spatial and temporal resolution offers a promising method for monitoring SWC and spatial mapping of ET. In this study, UAV-based multispectral and thermal data were acquired in an experimental field with four landscape groundcover species over two years (May-October 2022 and 2023). Two regression models, including multiple linear regression (MLR) and random forest regression (RFR), were used to predict soil moisture at depths of 10 and 30 cm. The results indicated that both regression models, MLR and FRF, exhibited a relatively good SWC prediction accuracy with Pearson’s r ranging 0.62-0.68, root mean square error (RMSE) ranging 0.034-0.048 cm3cm-3, and mean absolute error (MAE) ranging 0.034-0.038 cm3cm-3. Additionally, two energy balance models, a modified version of SSEBop and pySEBAL, were used to estimate ET for four groundcover species. The performances of models were evaluated against measured ET using the soil water balance approach. Model comparisons indicated that ET estimates for both models correlated well with ET measurements, with Pearson’s r ranging from 0.798-0.928 for the modified SSEBop and 0.843-0.961 for the pySEBAL model. However, the pySEBAL model had lower RMSE values (0.660-1.155 mm day-1) compared to the SSEBop model (0.870-1.270 mm day-1). This study shows that high-resolution UAV imagery combined with energy balance models can be used to estimate ET accurately for different urban vegetation types.