Análise de séries temporais compostas por imagens sintetizadas a partir da fusão de dados MODIS-TM

Abstract

Monitoring and characterization of remnants of the native forest and anthropogenic areas are essential for the effective land use management. Remote sensing is one of the most important tools which can assist and aid these studies on the Earth surface. The multi-temporal analysis of remote sensing images, integrated to spectral and spatial components, when properly explored, provides important information for environmental monitoring. It also allows the analysis of complex patterns and characterization of Earth surface covering dynamics. Due to some limitations on the sensors currently available, concerning resolutions, some methodologies of images fusion has been developed, such as the STARFM which is used for studies that require either high time frequency (MODIS) or high spatial resolution (Landsat-TM/ETM+). In this study, we aimed to assess a TM time series comprised by synthetic images, generated using the STARFM prediction algorithm. The study area is located in the southern of Minas Gerais State, Brazil, corresponding to the Landsat-TM 218/075 scene. We acquired all Landsat-TM and MODIS (h13v11 scene) scenes between 2000 and 2011. We used the near infrared (NIR) and red band for prediction of TM data. Analysis consisted in comparing reflectance values among original and synthetic TM data by means of linear regression. We also analysed MODIS and TM NDVI time series to compare statistics of the series generated using greenbrown, and of time components of tendencies and seasonality, obtained from time profiles of some classes of land use, generated using bfast. According to results, although some limitations, the STARFM prediction algorithm was found to be quite promising in predicting TM images. The analysis of TM NDVI time series provided better spatial and time details than the MODIS time series. The NDVI time series, comprised by synthetic TM images was found to be quite similar to MODIS time series, showing the majority of breakpoints in the component of trend coincident between both time series. Errors in predicting TM images can cause false detections of changes. The analysis of TM time series provided better spatial and time details than MODIS time series, so that it can be an alternative for Earth surface monitoring on the area of study.