SRTM DEM Correction Using Ensemble Machine Learning Algorithm

Abstract

The Shuttle Radar Topographic Mission (SRTM) digital elevation model (DEM) is a widely utilized product for geological, climatic, oceanic, and ecological applications. However, the accuracy of the SRTM DEM is constrained by topography and vegetation. Using machine learning models to correct SRTM DEM with high-accuracy reference elevation observations has been proven to be useful. However, most of the reference observation-aided approaches rely on either parametric or non-parametric regression (e.g., a single machine learning model), which may lead to overfitting or underfitting and limit improvements in the accuracy of SRTM DEM products. In this study, we presented an algorithm for correcting SRTM DEM using a stacking ensemble machine learning algorithm. The proposed algorithm is capable of learning how to optimally combine the predictions from multiple well-performing machine learning models, resulting in superior performance compared to any individual model within the ensemble. The proposed approach was tested under varying relief and vegetation conditions in Hunan Province, China. The results indicate that the accuracy of the SRTM DEM productions improved by approximately 46% using the presented algorithm with respect to the original SRTM DEM. In comparison to two conventional algorithms, namely linear regression and artificial neural network models, the presented algorithm demonstrated a reduction in root-mean-square errors of SRTM DEM by 28% and 12%, respectively. The approach provides a more robust tool for correcting SRTM DEM or other similar DEM products over a wide area.