Digital elevation model of forest areas in Guangdong Province

Satellite-based laser altimetry enhances the accuracy of elevation data in the Global Digital Elevation Model (GDEM) through its wide coverage and high precision. The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2), with its photon detection sensitivity, provides more accurate and denser surface elevation observations. This forested terrain elevation dataset for Guangdong Province, China, utilizes high-accuracy ICESat-2 satellite control points as reference data and improves the quality of Copernicus DEM in forested areas by integrating multiple feature parameters using the LightGBM machine learning method.Specifically, the ICESat-2 control point data covering the forested regions of Guangdong were first extracted, and the ICESat-2 elevation values were compared with the DEM elevation errors in these areas. Next, an attribute set was created to assess the error sources in the Copernicus DEM. These error sources include terrain features (slope , aspect,etc), vegetation data (vegetation height, vegetation coverage, etc.), radar parameters, DEM quality, and 15 other feature parameters, all of which contribute significantly to the model without redundancy. Finally, a regression model was constructed based on the attribute set and the DEM elevation errors within the ICESat-2 coverage area. The model, powered by the LightGBM machine learning algorithm, was used to correct the DEM in areas not covered by ICESat-2 data.Experimental results demonstrate that using RMSE as the accuracy metric, the model improved the RMSE from 2.55m to 1.06m, achieving a 59% improvement in accuracy.The data resolution is 30m,the value represents the height of the forest terrain,nodata represent  not forest area.