利用随机森林回归算法校正ASTER GDEM高程误差

Correcting Elevation Error of ASTER GDEM Using Random Forest Regression Algorithm

通过构建ASTER GDEM(advanced spaceborne thermal emission and reflection radiometer global digital elevation model)高程误差与影响因子间的关系模型,可对其高程精度进行有效校正。选取陕北黄土高原境内长武、宜君、甘泉、延川4个不同地貌类型的样区,以1∶5万DEM(digital elevation model)作为参考数据,经过数据预处理后,计算各点位高程误差值及相关地形因子和地表覆盖指数;提取一定数量的采样点和检验点,引入随机森林回归算法,建立高程误差预测模型,以对高程精度进行校正,并与多元回归模型进行比较分析。实验结果表明,ASTER GDEM的高程误差特征与地形条件有较强的相关性;随机森林回归预测模型整体上优于多元回归模型,具有较好的适用性与误差校正效果,可分别将长武、宜君、甘泉、延川的高程误差均值减小3.08 m、3.00 m、3.61 m和6.95 m。

Objectives: Effective correction of ASTER GDEM(advanced spaceborne thermal emission and reflection radiometer global digital elevation model) elevation error is of great significance to the quality and application of ASTER GDEM data. Terrain parameters such as slope and aspect derived from digital elevation model(DEM) data have a significant impact on the accuracy of the data. Therefore, the relationship model between the ASTER GDEM elevation error and each influencing factor can be constructed to effectively correct its elevation accuracy. Methods: Four geomorphological types of Changwu, Yijun, Ganquan and Yanchuan in the Loess Plateau of northern Shaanxi are selected. With 1∶50 000 DEM as reference data, after data preprocessing, the elevation error values and related topographic factors and the surface coverage index of each point are calculated, by extracting a certain number of sampling points and check points,a random forest regression algorithm is introduced to establish an elevation error prediction model to correct the elevation accuracy, and compared with multiple regression model. Results: ASTER GDEM elevation error characteristics are closely related to the terrain conditions;the contribution of each impact factor to the model is different in different areas;the random forest regression prediction model is better than the multiple regression model overall, and has good applicability and error correction effect. It can reduce the error mean of Changwu, Yijun, Ganquan and Yanchuan by 3.08 m, 3.00 m, 3.61 m and 6.95 m. Conclusions: This research is helpful to understand the ASTER GDEM elevation error characteristics and its influencing factors in different geomorphological areas of the Loess Plateau in northern Shaanxi. At the same time, suitable regression models can be selected for areas with different terrain conditions to effectively correct the error values, further improve the application accuracy of ASTER GDEM data.