Result2019.tif

Water overuse in agriculture poses a significant threat to ecosystem degradation in drylands. Consequently, striking a delicate balance between water resources allocated for agricultural development and ecosystem preservation has emerged as an urgent concern for sustainable development in arid regions. To address this challenge effectively, precise quantification of cropland becomes essential in accurately measuring water consumption in agriculture. In this study, we focused on the Ebinur Lake Basin and developed a method to accurately identify cropland. Our approach integrated phenology information of major crops, multi-remote sensing data, and cropland layers from various land cover products. We carefully selected 46 metrics using a rigorous process involving random forest Out-of-Bag (OOB) and co-linearity tests. Additionally, we created two key phenological metrics of maize and cotton based on NDVI data, allowing us to capture critical temporal patterns in crop growth. Using a random forest classifier, we successfully identified cropland in the region for 2019. The results show that: (1) Cropland mapping in the Ebinur Lake Basin for 2019 achieved a good overall accuracy (OA) of 92.39%, with user accuracy (UA) and producer accuracy (PA) at 96.47% and 94.13%, slightly outperforming other existing cropland products; (2) The inclusion of phenology metrics led to an OA increase of 1.40%, with gains of 0.75% in PA and 0.97% in UA. These enhancements suggest the significance of leveraging crop phenological characteristics in the identification of cropland areas; (3) Notably, two phenology metrics used ranked among the top 10, further affirming their substantial contribution to the accurate identification of cropland regions. These results emphasize the importance of considering crop growth patterns and temporal dynamics in the mapping process and provide valuable insights for understanding water usage patterns and promoting sustainable land management in drylands.