Biometric variable modeling in different vegetation classes using synthetic data from the Landsat satellite
Name: LARISSA GARCIA FERREIRA
Publication date: 27/09/2023
Examining board:
Name |
Role |
|---|---|
| ADRIANO RIBEIRO DE MENDONCA | Advisor |
Summary: The use of passive remote sensing is an alternative to forest inventory to estimate dendrometric variables. Therefore, the main objective of this work was to evaluate the accuracy of estimates of average diameter, average total height and stem biomass at
different stages of succession of the Atlantic Forest biome based on synthetic images from the Landsat satellite. Remote sensing data were obtained from synthetic images using the Continuous Change Detection and Classification (CCDC) algorithm. Spectral bands and NDVI were used as explanatory variables for modeling. Graphs of observed versus estimated variables, the adjusted coefficient of determination (2%) %), and the root mean square error (RMSE%) were used to evaluate the models. Model selection was performed using the F test to verify the significance of model parameters and analysis of variance was used to compare nested models. Furthermore, k-fold cross validation was performed repeated 1000 times with k equal to 10 for the selected model for each variable analyzed. The selected variables were the annual percentiles (10 to 100) of spectral bands 2, 3, 4,
5, 6e 7. The near-infrared (5) and mid-infrared 1 (5) bands were selected for all estimation equations for mean diameter, mean total height and bole biomass at different successional stages. Estimates of average diameter and average total height showed good accuracy in cross-validation. The bole biomass estimates were of low accuracy and are not recommended for estimating the biomass of the different successional stages. The use of metrics extracted from synthetic images obtained from Landsat satellite images, together with traditional forest inventory data, made it possible to estimate the biometric variables of
the study area.
