Predicting human activities patterns based on climate and related data

Human activities as agriculture and urban construction have altered a large portion of natural ecosystems and vegetation cover, defining three new land cover classes: croplands, urban and built-up areas and mosaics of croplands and natural vegetation. In a period in which climate is rapidly changing, understanding how climatic conditions are linked to the global distribution of human-modified land covers has fundamental importance. Our goal is to identify the aforementioned relationships and build models that predict realistic fractions of land covers associated with human activities only based on climate data. Decision trees and random forests have been employed to solve the three multiple regression problems, one for each human-modified land cover class. Experiments for optimal model selection have been conducted. Out of three land cover classes, only croplands responded well to modelisation. Decision trees exhibited sensible predictive accuracy and good potential for climatic patterns description, yet little robustness. Whereas, Random forests guaranteed higher accuracy and more stability, proving to be reasonably informative models. They provided valuable insights into the nature of the connections between climate and the distribution of croplands.