Hybrid Model for Forecasting of Changes in Land Use and Land Cover Using Satellite Techniques

This paper proposes a hybrid model identified as krigging ordinary-forecasting models that contributes to predict the spatio-temporal of land use and land cover (LULC) changes using a unique predictor variable represented by the surface reflectance derived of satellite images, transformed in the principal component 1 (PC1). The tools used allow knowing the trends of spatial and temporal prediction models of PC1 semivariances and to judge the adjustment between observed and predicted variables by analyzing prediction statistics as: root-mean-squared error, mean absolute error, mean absolute percentage error, mean error, and mean percentage error. From the observation of statistics, the best spatio-temporal adjustment can be selected. The prediction of LULC changes through the PC1 prediction can be followed for different future time into the time series. The samples evaluated of PC1 prediction in the validation stage give a correlation coefficient upper to 0.8 and adjusted determination coefficient upper to 0.7; being a successful adjustment between observed and predicted values allowing to select the hybrid model proposed to forecast the PC1 variable in a future time. Likewise, an extensive time series is not required to get a good prediction, which has been obtained as a result of the test of three annual time series in different period constituted by a minimum of five years (2014-2018) and a maximum of eight years (1991-2003).