Integrating Satellite-Derived Data and Machine Learning Algorithm for Assessing Winter Wheat Evapotranspiration

Evapotranspiration (ET) is the loss of water from both the soil and plants, and it is an important component of the hydrologic cycle. In the recent decades, ET estimation has improved due to developments in remote sensing technologies, particularly in the agricultural domain. ET is affected by a variety of factors, including weather and crop conditions, which are difficult to estimate for larger regions at fine resolution. Therefore, the current study intends to employ the Surface Energy Balance Algorithms for Land (SEBAL) model using satellite images to estimate and provide spatial ET variation using crop growth biophysical parameters such as land surface temperature, albedo, Normalized Difference Vegetation Index (NDVI), Soil Adjusted Vegetation Index (SAVI), Leaf Area Index (LAI), net radiation, sensible heat flux, and soil heat flux. The pixel-based SEBAL technique was used for the Haridwar district (area = 2360 sq. km) of Uttarakhand, India. The study utilized 5 cloud-free harmonized Landsat 8 and sentinel 2 satellite data for winter wheat crop at the beginning, middle, and end of the season. The area of cultivated wheat fields was initially identified using a machine learning support vector machine technique based on time series-threshold values of NDVI. This showed a wheat area of 526.86 sq. km, while the observed wheat acreage was 446.44 sq. km. The results showed that, for the research region, the support vector machine produced a significantly accurate assessment, with a kappa coefficient of 0.89, producer accuracy of 0.89, user accuracy of 0.82, and overall accuracy of 0.84. The estimated mean actual ET values were found to be 3.7 mm/day, 3.0 mm/day,  4.1 mm/day, 0.6 mm/day, 0.8 mm/day, and potential ET calculated by FAO-56 Penman-Monteith method were 4.4 mm/day, 4 mm/day, 4.1 mm/day, 3.7 mm/day, 2.1 mm/day dated 14^th and 6^th March, 2023, 26^th and 18^th February 2023, 17^th January 2023, respectively.  Based on the findings, ET maps and NDVI maps showing spatial variation were developed for the study area. These maps can be helpful for hydrological modeling, drought management, crop yield estimation, and irrigation scheduling.