Mapping field-scale crop water stress for wheat using satellite remote sensing data by formulating lower baseline using a novel approach

The quantification of crop water stress is very crucial for efficient irrigation water management and sustainable agriculture. The empirically derived crop water stress index (CWSI) is a widely used method for quantifying the crop water status. However, developing lower baseline is a prerequisite for estimating the crop water stress using the empirical approach. Traditionally, the lower baseline is formulated by taking in-situ observations of a well-watered crop canopy using infrared radiometers. In this study, a novel methodology is formulated for estimating the lower baseline using land surface temperature (LST) and normalized difference vegetation index (NDVI) for the wheat crops using Landsat-8, Landsat-9 and Sentinel-2 satellite data. This study is conducted during the 2021-22 and 2022-23 wheat crop seasons, covering approximately 630 acres of agricultural fields, managed by local farmers in the western part of Uttar Pradesh, India. The entire analysis is conducted on Google Earth Engine.  Initially, multi-temporal image classification is performed, employing the synergetic use of Sentinel-2 and machine learning algorithms, to distinguish the wheat and non-wheat fields. The manually collected ground truth data are used to train and test the random forest model. Subsequently, the candidate pixels are selected based on the maximum NDVI range, from (NDVI_max - 0.1) to NDVI_max, which represents dense and healthy wheat patches. These candidate pixels are further refined by selecting the pixels having less than 10^th percentile of the LST values, which account for relatively higher evapotranspiration. The lower baseline is derived using LST values of the refined candidate pixels along with concurrent air temperature (T_a) and relative humidity measurements recorded by an automatic weather station.  Finally, CWSI is mapped for the study area using the empirical approach.

Classification accuracy of 96% and 95% was achieved for the classification of wheat and non-wheat fields during the 2021-22 and 2022-23 seasons, respectively, with corresponding Kappa coefficients of 0.85 and 0.80. For the classified wheat pixels, the lower baseline equation formulated by the proposed methodology are (LST – T_a) = -1.864VPD + 1.325 for 2021-22 season and (LST – T_a) = -4.92VPD + 3.14 for 2022-23 season, where VPD is vapour pressure deficit. The fixed upper baseline of (LST – T_a) = 4°C is taken for empirically deriving and mapping CWSI for both seasons. The minimum and maximum values of the CWSI ranged from 0 to 0.89 during the 2021-22 season and from 0 to 0.78 during the 2022-23 season. The 2021-22 cropping season observed increased CWSI values as compared to 2022-23, primarily due to the heatwave that occurred in the study area from during the latter part of the 2021-22crop season. Significant spatial and temporal variability is obtained in the CWSI values within the study area.  The results suggest that the proposed methodology can be effectively used for mapping crop water stress at field scale without the requirement of tedious in-situ canopy temperature observations.