Assessment of Water Spectral Indices Using a Landsat-9 scene: a Case Study in a Semi-arid Region

Abstract

Spectral indices for detecting surface water bodies play a crucial role in environmental studies, on the other hand, these indices behave differently depending on the study site or the bands used. Therefore, this study proposed an evaluation of the performance of spectral indices in detecting surface water, using a scene from the Landsat-9 satellite and implementing a new index called NIR-Green Water Index (NGWI). After a visual inspection, positive performance was found in all indices in the detection of surface water, especially those formulated using the green band, such as the Normalized Difference Water Index (NDWI), (NGWI) and Water Ratio Index (WRI). During the Pearson correlation and Separability Index analysis between the "Water" and "Non-water" classes, the Automated Water Extraction Index (AWEI) demonstrated the highest average separability, reaching 1.24, and the lowest correlation, with 0.15, which places it as the index with the best detection estimates. The NGWI stood out especially when compared to other indices, showing a moderate behavior with an average separability of 1.15, surpassing the WRI (1.12) and the MNDWI (1.13), in addition to an average Pearson correlation of 0.17, just behind AWEI (0.16) and WRI (0.15). Although all the indices mentioned have demonstrated usefulness in water detection, it was observed that more complex indices, with a more elaborate formulation and less sensitivity to shadows, such as AWEI, produced results comparable to simpler indices, such as NDWI and MNDWI.

Keywords

• Remote Sensing
• Caatinga Biome
• Water Reservoir