UNET-Driven Water Body Segmentation for Enhanced Hydrologic Model Reliability and Water Resource Distribution

Abstract

Accurately dividing up bodies of water is important for making hydrologic models more accurate. These models are very important for figuring out where and how water resources will be available. This study looks into how UNET-based models can be used to separate bodies of water. It does this by comparing how well three models work: CNN_UNET, EfficientNetB0_UNET, and MobileNet Lightweight UNET. The main goal is to find out how well and accurately these models work to make hydrologic model estimates better. The CNN_UNET model is the basis for this method; it provides a basic but useful framework for dividing areas of water. EfficientNetB0_UNET, on the other hand, uses advanced convolutional neural network (CNN) methods to improve the accuracy of segmentation by finding the best balance between model efficiency and complexity. Finally, the MobileNet Lightweight UNET model is used to see if it is possible to keep the model's accuracy high while reducing its processing needs. This would make it ideal for real-time use in water forecast. A dataset that showed different water body traits was used to train and test the models. The comparison of accuracy shows that the MobileNet Lightweight UNET model had the best training accuracy (81.46%) and validation accuracy (77.14%). The EfficientNetB0_UNET model, on the other hand, was only moderately accurate, with scores of 78.64% for training and 72.73% for validation. With a training accuracy of 74.61% and a validation accuracy of 71.96%, the CNN_UNET model had the worst result. The results show that the MobileNet Lightweight UNET model is the most promising for water body division in hydrologic applications. It has the best mix of accuracy and efficiency, which makes forecasts about water resources more reliable.