Methodology for determining runoff coefficients using open-access images and classification open-source algorithms

Resumen

This study proposes a methodology using Landsat-8, Sentinel-1, and Sentinel-2 images to determine Landcover and Runoff Coefficient. It considers geographic and climatic characteristics of the region, resolutions of sensors, and accuracy obtained using four classifiers: Minimum Distance, Spectral Angle Mapper, Maximum Likelihood, and Support Vector Machines. The methodology was applied to regions in Mexico with diverse characteristics. The overall accuracy, commission, and omission were obtained for each classifier-sensor combination. These indicators were analyzed by region, sensor, and algorithm. Presence, accuracy, error of commission, and omission of cover were analyzed according to land cover class. Landsat-8 had the highest average accuracy (76.1 %), followed by Sentinel-2 (75 %), and Sentinel-1 (38.8 %). Minimum distance had the highest accuracy (71.2 %), followed by Support Vector Machines with 70.5 %, Spectral Angel Mapper with 60.9 %, and Maximum Likelihood with 50.7 %. This methodology makes it possible to use open-access satellite images and programs to define criteria for determining land cover with good accuracy and high spatial resolution. The results of our research show that to estimate runoff coefficients, open-access satellite images must be selected based on criteria such as sensor resolutions, classification algorithm to be used, period and scale of the study, climate, types, and dynamics of land coverage, as well as the dominant coverage class.