Monitoring water surface of agricultural reservoirs using multispectral remote sensing in semi-arid regions

Master

英语. This academic internship focused on addressing the gap in understanding water dynamics in small-sized agricultural reservoirs in semi-arid regions. Using Sentinel-2 multispectral remote sensing data, the study aimed to estimate water surface levels over the period 2017 – 2023 and analyze filling and discharge patterns in two reservoirs in Tunisia (Lebna watershed) and three in India (Berambadi watershed). The research investigated how a multitude of spectral indices, including Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), Automated Water Extraction Index (AWEI) and Multi-Band Water Index (MBWI), the Sentinel Water Index (SWI), and Normalized Difference Vegetation Index (NDVI), could be utilized to quantify uncertainty in water surface estimation and leverage the multi-temporal frequency of Sentinel-2 to characterize recharge and discharge patterns. The study employed six frequently used indices and assessed their performance against ground data for single-day observations and temporal scales. A Locally Estimated Scatter Plot Smoothing (LOESS) algorithm was applied to determine minimal and maximal surfaces between seasons by comparing indices LOESS to ground data LOESS for accuracy in estimating seasonal patterns. The study revealed suboptimal accuracy in actual surface estimation, and the attempt to use multiple indices to quantify uncertainty for small reservoirs was unsuccessful. However, the identification of seasonal patterns proved feasible, demonstrating the potential of this methodology in understanding interseasonal charge and discharge patterns in small reservoirs. Further analysis is needed to validate these findings across a larger number of test sites and temporal scales.