Uso das séries temporais de índices de vegetação para o monitoramento agrícola no Estado de Mato Grosso

Abstract

The agriculture is one of main sectors of the Brazilian trade balance, especially from the century begin. In this scenario, the State of Mato Grosso stands out. Called the "Barn of the World", its landscape changed with the agricultural frontier expansion over the Cerrado and Amazon biomes, becoming a world power of the sector, with large areas of production in its territory. Considering the state dimensions, distribution, and dynamics of agricultural practices, it is important to highlight the usefulness of remote satellites and sensors for the monitoring of large agricultural territorial extensions, for their ability to collect detailed and reliable information with a high revisit frequency, associated with conventional agricultural data collection systems. The time series from vegetation-oriented composite products such as MOD13Q1 and MYD13Q1 from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor make it possible to evaluate vegetation dynamic by identifying seasonalities and trends inherent to phenological cycles of agricultural crops. In the spatial part, geostatistical techniques combined with in situ and census data allow to break down the scale barriers and create the downscaling effect, a factor that presents potential to identify agricultural crops and to estimate yield with less uncertainty. In face of technological innovations, this study presents the use of geostatistics techniques and vegetation indices time series analysis for the derivation of phenological cycle parameters, in order to identify, interpret and monitor the evolution of agricultural areas in Mato Grosso between 2000 and 2012, period of exponential agricultural expansion after the turn of the century. At the state level, it was possible to identify soybean areas and their intensification throughout the period with a Global Accuracy of 92.1% and a Kappa index of 0.84, as well as productivity in five agglomerations of farms of different mesoregions with 95.09% of accuracy, considering t he standard deviation and probable error. In order to disseminate the area and yield maps obtained, a web platform was developed entitled SojaSAT. At the level of an agglomerate of farms and field, it was possible to identify crops in the harvest and in the safrinha with Global Accuracy of 89.5% and Kappa index of 0.80. The results obtained show the usefulness of the MODIS time series combined with geostatistical techniques and time series analysis to monitor the phenological cycle of the crops, making it possible to identify areas for planting different crops according to the detected spectro-temporal responses.