Predicción de bofedales en la Reserva de Producción de Fauna Chimborazo a través del tratamiento de imágenes satelitales

Abstract

The present search proposes: to carry out the prediction of bofedales in the Chimborazo Wildlife Production Reserve, through the treatment of Landsat satellite images, for which three Random Forest algorithms were developed. This process was carried out with the calculation of spectral indices that covered abiotic characteristics of the area, generating a predictive model of categorization by means of decision trees: the algorithm chose the class or coverage that obtained the highest value, yielding variables in each level of the tree. The results show that the relevant factors in the categorization for each training are: in the first training the NDWI normalized water differential index with a categorization percentage of 97.88%, being 97.44% bofedal and 99.44% different from bofedal, in the second training was the adjusted vegetation index with the soil brightness factor SAVI 15 with a percentage of 89.45%, being 85.23% bobedal and 90.04% different from bofedal and in the third training the NDVI standardized vegetation index with a 95.95 percentage % being 93.82% bofedal and 99.21% other than bofedal . In order to analyze the percentages and behaviors of the bofedales, a multitemporal analysis of the study area for the years 1991, 2001, 2010 and 2016 was carried out under the same procedure as the Random Forest algorithm; for which we worked with Landsat 8 OLI, 7 ETM and 5 TM satellite images, on the results obtained from the years 1991 and 2016 the IUCN area reduction tool was applied, which project an absolute reduction rate of the ecosystem in the next 16 years, for which I assign the category of critical danger, in this context it is necessary to propose management actions and sustainable use of this ecosystem to avoid its collapse.

Key words: REMOTE SENSORS - ALTOANDINOS WETLANDS - ECOSYSTEMIC PASS-BOFEDALES