A 30 m Canopy Height Map in China Created by Fusion of Multiple Relative Height Metrics

Accurate estimation of canopy height is crucial for monitoring forest health, carbon cycling, biodiversity, and climate change. Existing canopy height mapping methods often integrate global ecosystem dynamics investigation (GEDI) light detection and ranging (LiDAR) data with optical or radar remote sensing images. These methods, however, typically establish relationships between single GEDI relative height (RH) metrics (e.g., RH95 or RH98) and surface reflectance or backscatter signals, overlooking valuable information from multiple RH metrics. In this study, we developed a multiple RH metrics-based canopy height mapping (MRH-CHM) model, which was implemented using deep learning based on the Google Earth Engine (GEE) cloud platform. The MRH-CHM model integrates features from multiple GEDI RH metrics (e.g., RH0–RH100 with an interval of 10 units) and also Landsat-8 and Sentinel-1 images. In order to deal with the distinct features in the multimodal data, in the proposed MRH-CHM method, the convolutional neural network (CNN) and multilayer perceptron (MLP) modules were designed separately before a feature fusion process. Using the MRH-CHM model, a canopy height map of China for the year 2020 was produced at 30 m spatial resolution. The MRH-CHM results present greater accuracy than Lang’s, Potapov’s, and Liu’s products, achieving the lowest root mean square error (RMSE) of 5.74 m and the largest correlation coefficient (r) of 0.78 when validated against 6 168 244 hold-out GEDI validation data points. The produced map provides valuable scientific data for policymakers, researchers, and forest stakeholders to monitor forest health and biodiversity and to guide efforts toward carbon neutrality. The produced canopy height map is publicly available at https://doi.org/10.5281/zenodo.11560195