Técnicas de modelagem e geoprocessamento visando ao desenvolvimento de um sistema de alerta precoce para os rebanhos em regiões da caatinga

Abstract

The aim of this study was to analyze the capacity of modeling and geoprocessing techniques in the generation of a protocol with early warning information for livestock farmers living in regions of the Caatinga Biome. The first chapter carried out a temporal analysis of droughts and the spatial distribution of herds in the Brazilian Semi-Arid region using geoprocessing tools. Annual maps of the vegetation index by the normalized difference (NDVI) and the drought index by the normalized difference (NDDI) maps were created using MOD09A1 images processed on the Google Earth Engine (GEE) platform and the QGIS Software (2.18). For the spatial distribution of herds, the municipal livestock production database of SIDRA/IBGE was used. The comparative analysis showed that the classes, extreme drought and exceptional drought were higher in the 2nd historical series, while abnormal drought had greater area in the 1st series. The analysis of the data generated information such as the most favorable places for the creation of sheep and goats in view of the risks of periods of climatic uncertainties. In the second chapter, spectral characterization was carried out through the seasonality of the bi-directional surface reflectance factor and the spectral indices of vegetation of fragments of native pastures, as well as the change in land use and occupation in regions of the Caatinga Biome. Images from the Landsat-8 satellite of the rainy season and the dry period of the year 2018 were used to verify the seasonality of the vegetation through the variation of the bidirectional reflectance of the surface of the spectral bands and the spectral indices of vegetation (NDVI, SAVI, EVI, IAF, MSAVI2 and NDWI). In order to obtain a change in land use and occupation, the Google Earth Engine (GHG) platform and Landsat-8 images were used during the period from 2014 to 2018. The validation of the classification model through the confusion matrix resulted in a global accuracy of 91% and Kappa index of 89% in Tauá, and global accuracy of 90% and Kappa index of 86% in Ouricuri. Image classification using the Google Earth Engine tool proved to be effective in verifying the temporal and spatial change in land use and occupation, making it possible to identify places with the most affected vegetation and susceptible to degradation. The third chapter tested the efficiency of empirical and mechanistic modeling in simulating forage biomass from native pastures of the Caatinga Biome and applying them to forage production maps. The estimate of pasture biomass production was based on the sum of herbaceous and shrub-tree biomass. For empirical modeling, images from the Sentinel 2-A satellite were used and the indices tested: NDVI, SAVI, EVI, IAF, MSAVI2 and NDWI, through the LAB Fit Curve Fitting. The calibration of the PHYGROW model was performed using the PHYWEB virtual platform. The SAVI was the index that presented the best adjustments of the model with a medium capacity to simulate the production of total forage biomass in rangelands of the Caatinga. PHYGROW showed better performance in simulating biomass production, proving to be useful for estimating biomass production in areas of Caatinga with different levels of woody density, making it a useful tool for the rational management of rangelands.