Quantification of Sugarcane Crop Water Footprint Using Remote Sensing and Machine Learning Techniques: Case Study of Kolhapur District, Maharashtra, India

Abstract. Water footprint (WF) analysis assists in measuring the freshwater utilized by the crops, which provides information regarding sustainable water utilization. Growing water scarcity, irrigation requirements, and climatic variability emphasize the need to effectively monitor and regulate water resources. Considering the limited availability of water resources, estimating the water usage of crops such as sugarcane with high water requirements is necessary. This study aims to calculate the blue water footprint (BWF) and green water footprint (GWF) requirements of sugarcane crops using empirical methods and machine learning techniques in the Kolhapur district of India. By employing robust ground truth data and spectral signatures of Sentinel-2 satellite images, the sugarcane crop masks were identified using advanced machine learning techniques: random forest (RF), support vector machines (SVM), and logistic regression (LR). Furthermore, BWF and GWF were quantified for the identified sugarcane crop using empirical methods that utilized precipitation, evapotranspiration (ET), minimum temperature, and maximum temperature data for the years 2018 to 2023. Following these initial estimations, the potential of machine learning techniques was assessed for predicting WF. The efficacy of RF, support vector regression (SVR), gradient boosting regression (GBR), and artificial neural networks (ANN) was assessed by training and validating them based on the identified features. The RF model (R²:92) outperformed the other models in the precise prediction of sugarcane crop WF. The results show a lower WF in the northern and eastern talukas and a higher WF in the southern talukas of the district. This study can aid in the identification of water-stress areas and sustainable water resource management for sugarcane crops.