Comparison Between Three Methods to Monitor Reservoir Extension in the Brazilian Semi-Arid Region

Abstract

The State of Pernambuco covers an extensive semi-arid area where the Caatinga biome dominates. This region is characterized by long periods of drought, highlighting the need for water resource optimization. This paper aimed to compare three methods to assess reservoir changes: MapBiomas' products, the Normalized Difference Water Index (NDWI), and a support vector machine (SVM) algorithm. Initially, we obtained the monthly precipitation from 1987 to 2019 and calculated the yearly accumulation. Mapbiomas, Landsat 7 ETM, and Landsat 8 OLI data from 2012-2018 were accessed and processed using the Google Earth Engine platform. We obtained the annual image with the median pixel criterion to determine the NDWI and quantify the annual reservoir area. For the supervised classification with SVM, samples from different land-use types of the study area were used to train the algorithm. From 2012 to 2018, a reservoir reduction of 63.42% was observed with MapBiomas images, 69.49% with NDWI images, and 67.69% using the SVM algorithm. The results obtained using NDWI were the most similar to those from the artificial intelligence classification, indicating that NDWI can be used to monitor the reservoir conditions.