Análisis de árboles de decisión para la valoración de carbono edafico de la provincia de Chimborazo mediante el uso de variables de evaluación nacional forestal MAE - FAO

Abstract

The present research work aimed to evaluate the decision tree technique by means of the best classification algorithm, for the evaluation of edaphic carbon in the province of Chimborazo; considering the MAE - FAO National Forest Assessment database. For the study the data set was cleaned, then the useful variables for the categorization of soil organic carbon (SOC) were determined, obtaining 4 classes: Very High, High, Medium and Low, after which variables were generated spectrals derived from Landsat 8 satellite images (OLI and TIRS sensor), using Geographic Information System (GIS). Twelve variables were found that control the SOC distribution dynamics, these were: Ecosystem, Taxonomy, Texture, Slope, Digital Elevation Models (DEM), Normalized Difference Vegetation Index (NDVI), Soil Adjusted Vegetation Index (SAVI ), Atmospheric Visible Resistance Index (VARI), Normalized Water Differential Index (NDWI), Calcined Area Index (BI), Normalized Burned Area Index 2 (NBR2), Two-band Enhanced Vegetation Index (EVI2). The algorithm that provided a better percentage of efficiency and relevant results was the classification and regression algorithm (CART) using the cross-validation method, the model generated a precision of 65.72% and a prediction error of 34.28%; These results are presented as a new alternative for the quantification of SOC. The calibrated model can be extended without the need for in situ sampling, very useful in complex areas such as the forest ecosystem. The digital mapping of SOC allowed to reveal the existing SOC levels in soils of Chimborazo and the inter-Andean alley. It is recommended to continue with research along this line, which will allow to identify the potential of the decision tree technique, so that they can be applied in situations of national interest.