Forest ecosystems play a crucial role in regulating eco-climate and carbon cycling, with forest height serving as a key metric for ecosystem functionality. This study aims to accurately survey the spatially continuous woodland canopy stature within the research area. Here, we employ a stacking algorithm approach. We utilize four machine learning models-Multiple Linear Regression (MLR), Gradient Boosting Machine (GBM), k-Nearest Neighbors (kNN), and Random Forest (RF)—as primary learners. A wide neural network (WNN) serves as the secondary learner in a stacked ensemble approach. The study integrates Sentinel-2 measurement data obtained from the Google Earth Engine (GEE) and ICESat-2 ATL08 measurement data downloaded from the National Snow and Ice Data Center (NSIDC). We assess forest canopy height in the research area. The results show that compared to the MLR, GBM, kNN, and RF models, the improved stacking model achieves the highest prediction accuracy for forest canopy. The RMSE values are reduced by 27.1%, 25.39%, 26.54%, and 20.08%, respectively. This study aims to achieve continuous mapping of forest canopy height in the research area and to lay the foundation for accurate mapping of forest canopy height in other regions.