PREDICTING SURFACE FOREST FUELS ON THE CERRADO IN CANTÃO STATE PARK FROM AIRBORNE RGB SENSOR IMAGES

Abstract

Forest fuel quantification in a typical Cerrado area is difficult due to the high costs and long field times associated with collecting data. In search of alternatives that facilitate data collection, indirect estimation has been studied, resulting in local equations for predicting the load based on easily obtainable variables. In this context, we sought to develop local equations to estimate the load of forest fuel, in a Cerrado area, in the Cantão Park State – Tocantins. As a function of variables measured in the field and digital variables from the processing of RGB digital images (Red, Green and Blue) acquired through aerial surveys. With the processing of digital images, the following variables were extracted: mean height in the digital model (hMDA) and point density in the three-dimensional model (DPM). After the aerial survey, field sampling was carried out, collecting the following variables: mean height in the field (hc), number of individuals (Qti) and Total Fuel Material (MCT). Subsequently, equations were fitted, adopting the entire sample set. The model selection criterion was based on R²aj, Syx% and residual graph, in which adjusted coefficient of determination (R²aj) from 0.12 to 0.83 and residual standard error (Syx%) of 19.4 were obtained at 44%. It was observed that the use of control points performed by the MDA reduces the occurrence of errors, the correlations between the analyzed variables indicate the importance of considering multiple factors when performing analyzes and predictions in the study areas.