Variância do erro de predição em um modelo geoestatístico espaçotemporal em dados de albedo obtidos por sensoriamento remoto

Abstract

The albedo of a surface (or reflection coefficient) can be defined as the ratio between the amount of sunlight reflected by the earth surface and the amount of sunlight received.Therefore, the increase or decrease of the albedo can be a good indicator of changes occurring in the composition of the atmosphere and Earth’s surface such as deforestation and desertification. Thus, the use of models capable of modeling the variability of the albedo in a particular region, over a period of time, can be of great importance for scientific research. The main aim of this thesis is to explore further the potential of the geostatistical space-time model (Kyriakidis and Journel (1999)) to analyze the variable albedo. To achieve this goal, this thesis incorporates new statistical methods in this model, such as prediction maps and prediction error variance maps via nonparametric bootstrap. The albedo data set was obtained through two complete regular grids, containing 249 and 499 points, georeferenced in mesoregion South/Southwest of the state Minas Gerais - Brazil, during the 31 days of December 2010, captured by the satellite Meteosat 9, acquired from the EUMETSAT. The results achieved in this study showed that: (1) The statistical methods proposed in this thesis to obtain both the prediction and the prediction error variance maps enable that the geostatistical space-time model spatiotemporal being used in actual data. (2) From the statistical methods proposed in this thesis, it is possible to observe that the number of samples points in the grid does not interfere with the predictions of the albedo produced by geostatistical space-time model.(3) Purely spatial analysis produced a slightly lower prediction error than the geostatistical space-time model. One possible reason for this behavior may be due to the fact that the model by Kyriakidis and Journel (1999) is built based on ad hoc arguments.