Please login first
Estimation of tree height in burned areas with GEDI laser data in northern Portugal and Galicia (Spain)
* , * , * , *
1  CECS, University of Minho
Academic Editor: Giorgos Mallinis

Abstract:

The monitoring of burned areas by forest fires is essential to know their dynamics and the use of orbital data and remote sensing techniques are fundamental in this process. The Global Ecosystem Dynamics Investigation (GEDI) program produces high-resolution laser observations of the earth, measuring vertical structure and canopy height, in addition to surface elevation, which are key data for understanding ecosystem services such as the carbon cycle. In the present study, we analyzed the vegetation structure of four areas affected by forest fires in the summer of 2020 in northern Portugal and Galicia (Spain). We used the Google Earth Engine platform to analyze satellite imagery, digital elevation model and GEDI data to measure vegetation height before and after fires. Our results indicated that before the fires the height varied from 5.21 to 20.16 meters, and the training and validation data obtained, respectively, r2 values of 0.82 and 0.67. After the fires, heights of 5.55 to 9.12 meters were recorded, with values of r2 0.47 and 1 in the training and validation data. These r2 values after fires indicate the absence or limitation of sample data. The training data recorded an RMSE value of 3.47 before and 3.36 after the fires. The validation data recorded an RMSE value of 5.23 before and 3.34 after the fires. The most important variables for the measurement were identified through the Random Forest algorithm and training and validation data, they are: VV_iqr, elevation and B8 of the GEDI, after the fires the most important variables were the bands B2, B3 and B11. We conclude that the GEDI data has great potential to assist in the mapping of areas affected by forest fires, with the potential to measure the height of vegetation and contribute to the monitoring of areas affected by fires.

Keywords: Forest fires; Remote Sensing; GEDI; Sentinel; Random Forest
Top