Using a Coupled Radiative Transfer Model and Artifical Neural Networks to Retrieve Forest Biochemical and Biophysical Maps from Remote Sensing Data
Over the last decades, forest ecosystems have been increasingly disturbed by a variety of disturbances. One of the major disturbances to the forests have been reported to be insect infestations and drought. As a result, retrieving forest biophysical and biochemical maps from remote sensing data could aid in the study of forest disturbances. In this study, the radiative transfer models PROSPECT5, 4SAIL and FLIM were coupled in order to simulate forest canopy reflectance for the National Park Hunsruck-Hochwald. The simulated data was used to train Artificial Neural Networks (ANN), and the final trained model was used to predict forest biophysical and biochemical parameters using the PRISMA hyperspectral remote sensing image of the study area. Two ANN models were trained, one with the dimensionality reduction technique, PCA and one with simulated reflectance for PRISMA data bands. In general, both trained ANN models performed well. However, it was determined that the ANN that utilized all the available bands was superior in terms of loss and prediction errors for each plant parameter.