Method for Forecasting of Changes in Land Use and Land Cover Using Satellite Remote Sensing Techniques

Abstract

In this investigation is proposed a method for forecasting of changes in land use and land
cover using satellite remote sensing techniques. This study includes the following twelve
stages: 1) acquisition of remote sensing data, 2) collection of the reflectance image time
series, 3) preliminary processing of reflectance image time series, 4) transformation of
reflectance image to principal components, 5) modelling of PC1 statistical spatial prediction,
6) calibration of forecasting models for the PC1 SSPM coefficients, 7) calibration of PC1
SSPM, 8) validation of PC1 SSPM, 9) forecasting of PC1 SSPM coefficients and 10)
calibration of CP1 SSPM with forecasted coefficients, 11) application of change detection
techniques and 12) comparison of methods. Sixteen satellite images are acquired from the
Landsat satellite in the period from 1986 to 2016. The study unit is the Pao river basin. The
proposed method is a hybrid combination that includes three types of applied models that are
based on time series of reflectance images in sequence as follows: the principal component
analysis, the statistical spatial prediction models and forecasting models for time series. The
current study proposes a method that contributes to introduce the temporal pattern of LULC
changes captured by the statistical spatial prediction method coefficients and provides results
characterized by a seasonality parameter; which is able to reproduce the spatio-temporal
variation collected by the reception of the reflectance variable by satellite sensor. The
statistics of error predictions indicate gradients of the predicted and observed function
approximated to the unity as well as near to zero for the errors. The samples evaluated in the
validation stage give correlation coefficient upper to 0.6; being a successful adjust between
observed and predicted values. The forecasted changes in the Pao river basin for 2020 and
20130 vary from: 5.54 to 8. 14%, 5.52 to 8. 14%. These changes are equivalent to those
observed from 2000 and 2016 of 5.13% as well as from 1990 to 2016 of 7.05 %.