Cartografía predictiva de ecosistemas dependientes de aguas subterráneas mediante algoritmos de clasificación supervisada

Zusammenfassung

This paper presents a novel methodology for predictive mapping using information based on point-source field data. For this purpose, different layers of spatial information have been collected and compiled in a geographic information system. The supervised classification algorithms are trained with a sample of the original data, trying to find patterns of relationship between the explanatory variables and the target variable, in this case, the presence or absence of groundwater-dependent ecosystems. The factors that will condition the presence or absence of these bodies of water are lithology, slope, water table elevation, fractures, humidity indices, permeability and flow accumulation potential among others. The study area, with an extension of more than 5,000 km2, is located in Castilla-La Mancha and includes the Lagunas de Ruidera Natural Park. Input data have been extracted for wetlands and springs from the Spanish Inventory of Wetlands (DGOH, 1991) and Mon- tero (2000), respectively. Outcomes are above 0.80 in test score and area under the receiver operating characteristic curve (AUC), the most used metrics in machine learning methods.