Site index estimation using airborne laser data in Eucalyptus dunnii Maide stands in Uruguay
Autor
Rizzo Martín, Iván Gabriel
Tutor
Hirigoyen, AndrésNavarro Cerrillo, Rafael M.
Editor
Universidad de CórdobaFecha
2022Materia
LiDARIntensive silviculture
Eucaliptus spp.
Linear model
Random Forest
Total height
Site index
Stand segmentation
METS:
Mostrar el registro METSPREMIS:
Mostrar el registro PREMISMetadatos
Mostrar el registro completo del ítemResumen
The estimation of forest variables to support a forest inventory can be approached through the use of different technologies. Although field sampling is the most widely implemented technique, the development of remote sensing techniques increases the possibilities of action in this field. One of these technologies is the airborne LiDAR scanner (ALS). In this study, linear models and non-parametric models with Random Forest imputation were generated to estimate the total height (HT) and site index (SI) of Eucalyptus dunnii Maide, based on LiDAR metrics. High spatial resolution continuous rasters for HT and SI were created with these models. The use of a semi-automatic object-oriented segmentation algorithm for stand delimitation based on the SI raster was then carried out. To evaluate the performance of these models, the One Leave One cross-validation technique was implemented, determining for each model the ratio between the RMSE of the model and the RMSE of the cross-validation (RMSEcv). Linear models for HT estimation presented a better fit (R2=0.84, RMSE=0.84 m, MAPE=0.039, Bias=0.002) than the Random Forest (R2=0.85, RMSE=1.26 m, MAPE=7.19, Bias=-0.173) model including only independent variable the 99th percentile. The RMSE/RMSEcv ratio presented a higher value for the linear model (0.93) than Random Forest (0.75). For the estimation of the site index (SI), the Random Forest model was applied, which included the LiDAR metrics corresponding to the 99th percentile and the 80th bicentile. This model had an R2 value of 0.65 and an RMSE value of 1.62 m. Then, on the SI raster generated by the Random Forest model, automatic segmentation was applied, generating segments with high internal homogeneity and low homogeneity between segments. The methodology developed in this work provides accurate estimates and mapping of HT and SI at stand scale based on LiDAR data. In addition, an automatic segmentation method was applied, generating stands based on the SI. This segmentation is very useful for the sector as it is a tool that will improve forest management in terms of harvesting and future plantations.
Descripción
Premio extraordinario de Trabajo Fin de Máster curso 2020/2021. Máster en Geomática, Teledetección y Modelos Espaciales Aplicados a la Gestión Forestal