Deep Regression Neural Network for Estimating Canopy Height in Vietnam's National Forests

Bao Quoc Bui, Nguyen Viet Le, Anh Duc Vu, Khoa D Doan, Nidal Kamel

Keywords: Canopy height estimation, vegetation structure, multispectral remote sensing

Abstract: Estimating canopy height in forests is an important and necessary step in measuring forest health, biodiversity, and carbon storage. This paper studies the challenges and applicability of using deep neural network models to predict canopy height in the national forests of Vietnam, a country with a rapidly growing economy and a commitment to achieving a net-zero carbon footprint by 2050. We first argue that estimating canopy height in Vietnam presents significant challenges, particularly the need for a low-cost approach and the country's complex forest structures. Then, using wide-coverage and freely accessible Sentinel-2 data, besides GEDI, we systematically study the performance of the existing canopy height regression models in the context of Vietnam. Finally, we propose a new approach that can effectively take advantage of the vegetation indices with original Sentinel bands data and show promising results compared to the existing models on standard evaluation metrics such as Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). The results show the potential of a low-cost canopy-height estimation approach, taking a step towards sustainable forest management and environmental conservation in Vietnam.

Submission Number: 9