Burn severity estimation in a disease affected forest landscape from remotely sensed data: A comparison of empirical, simulation and deep learning models

Authors: Yinan He*, University of North Carolina - Charlotte, GANG CHEN, University of North Carolina at Charlotte, Feng Huang, Yango University; Fuzhou University
Topics: Landscape, Sustainability Science, Remote Sensing
Keywords: Burn Severity; Empirical Model; RTM Inversion; Deep Learning; Landsat; Spectral Weighting Analysis; OBIA
Session Type: Paper
Day: 4/11/2018
Start / End Time: 8:00 AM / 9:40 AM
Room: Studio 7, Marriott, 2nd Floor
Presentation File: No File Uploaded


Forest ecosystems are subject to recurring fires as one of the most significant disturbances. Accurate burn severity estimation is a crucial conditioning factor for post-fire land management and vegetation regeneration monitoring. Mapping burn severity in a pre-fire, disease-affected forest often faces challenges because the post-fire spectral responses are joint effects of fire and disease disturbances, leading to a confusion in the relationship between spectral reflectance and burn severity. In order to accurately map burn severity and investigate the performance of recent methods in a forest landscape jointly affected by disease and fire, we compared three types of models, including empirical, radiative transfer, and cutting-edge Deep Learning models (Recurrent Neural Network) in this study. Meanwhile, remotely sensed images from multiple sensors (i.e., AVIRIS, MASTER, and Landsat) were analyzed to test the effects of varying spatial and spectral resolutions on model performance.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login