Integração de técnicas geoestatísticas e Sensoriamento Remoto em fragmento da Floresta Ombrófila Mista Montana

Abstract

The spatial characterization of an ecosystem through traditional inventories involves intense, time-consuming and costly work and therefore field measurements cannot be applied on a large scale or over large areas. On the other hand, the integration of multivariate geostatistical techniques with data obtained from space platforms can provide an accurate and low-cost overview of important forest variables. And thus, provide subsidies for forest ecosystem conservation and restoration practices, environmental impact assessment and forest management. In this context, the objective of the present study was to evaluate the performance of ordinary cokriging (OCK) in mapping vegetation metrics of a remnant of Mixed Ombrophilous Forest (MOF), using covariables derived from Sentinel-2A satellite images. For this, data from the continuous forest inventory, carried out in the National Forest of Irati (FLONA de Irati), under the domain of Mixed Ombrophilous Forest, were used, referring to measurements carried out in 2017 in 400 georeferenced sample units of 25 m x 25 m (625 m2 ), totaling an area of 25 hectares. Four metrics related to vegetation were considered in the study: two of diversity: richness (S) and Shannon Index (H') and two structural: basal area (G) and number of trees per hectare (N). Although the field survey contemplated 400 sample units, only 100 plots with the vegetation metrics were used, because it was assumed that the target variables were under-sampled and the covariates over-sampled. To obtain the covariables, multispectral images were obtained from the Sentinel-2A satellite, with a certain level of processing, requiring only an atmospheric correction. After the atmospheric correction, the average values of reflectance of the bands (B2, 0.46 µm to 0.52 µm; B3, 0.54 µm to 0.58 µm; B4, 0.65 µm to 0.68 µm and B8, 0.78 µm to 0.90 µm) and the Normalized Difference Vegetation Index (NDVI) of the 400 plots were obtained. The maps with the vegetation metrics were then obtained, considering two interpolation techniques: one univariate, the ordinary kriging (OK), and the other multivariate, the OCK. To evaluate, by Mean Bias Error (MBE) and Root mean square error (RMSE), the accuracy of the maps obtained were used data with the vegetation metrics of 50 extra plots, randomly distributed in the study area. The vegetation metrics were negatively correlated with band reflectance and NDVI, with low correlation coefficients. The highest correlation coefficients between target and covariable variables were: S and G with band 3, H' with band 4 and the number of trees per hectare with band 2. In general, the two interpolation techniques mapped the vegetation metrics well. However, the hypothesis considered that data derived from Sentinel-2A combined with multivariate geostatistical techniques improve the accuracy of mapping structural attributes and diversity of tree species in Mixed Ombrophilous Forest was rejected. Although the hypothesis tested was rejected, a potential use of auxiliary data derived from satellite images was observed in the mapping of the variables studied, especially the basal area and tree density. And that despite the efforts, there is a vast field to be explored and answered.