Estimation of dam volume using remote sensing and Machine Learning

Abstract

In Mexico, where the most significant water consumption sector is agriculture, and due to the increasing demand for food, it is necessary to maintain a proper balance between agricultural production and water consumption. In this study, with the assistance of Sentinel 2 images, Machine Learning models were developed to estimate the surface area and volume of the Manuel Ávila Camacho (Valsequillo) dam in Irrigation District 030, which receives wastewater, leading to the proliferation of aquatic lily. There are water indices such as NDWI, NDWI_Mcfeeter, NDWI_Gao, NDWI_Xu, AWEI_non-shadow, AWEI_shadow, and ICEDEX, which enable the discrimination of water bodies through the use of satellite images. However, in this case, the use of these indices results in confusion between the lilies and natural vegetation. The training was conducted using the values of the aforementioned indices and the RGB, NIR, and SWIR layers at the pixel level, and the accuracy results obtained were as follows: Linear discrimination model at 98.1 %, decision trees at 99.2 %, and logistic regression at 98.5 %. With these models, it was possible to estimate the surface area of the dam's water body. Using the graph of the dam's capacity areas, a quadratic function with an R-squared value of 0.9988 was generated, where volume is a function of area, and the dam's volume was estimated, with an average difference of 8.5 % compared to the linear discrimination model.