Global　Ecosystem　Dynamics　Investigation　(GEDI)　and　Ice,　Cloud,　and　land　Elevation　Satellite-　2　(ICESat-　2)　products　provide　reliable　global　references　for　the　accuracy　evaluation　and　correction　of　Global　Digital　Elevation　Model　(GDEM).　However,　existing　DEM　correction　methods　mainly　address　the　signal　of　vegetation　in　DEM　errors　and　mostly　use　linear　regression　models.　So,　we　first　analyze　the　relationship　between　Advanced　Spaceborne　Thermal　Emission　and　Reflection　Radiometer　(ASTER)　GDEM　v3　data　accuracy　and　the　land　cover　type,　elevation,　slope,　relief　amplitude,　and　vegetation　coverage.　Based　on　this,　this　paper　proposes　a　Digital　Elevation　Model　(DEM)　error　correction　method　that　takes　into　account　various　influencing　factors　and　combines　Extreme　Gradient　Boosting　(XGBoost)　machine　learning　and　spatial　interpolation　to　model　the　errors.　The　analysis　of　the　results　shows　that　the　overall　error　of　the　original　ASTER　GDEM　has　a　normal　distribution　with　a　large　negative　offset　(average　error　of　-　3.463　m).　The　Mean　Absolute　Error　(MAE)　and　Root　Mean　Squared　Error　(RMSE)　of　original　ASTER　GDEM　are　12.930　m　and　16.695　m,　respectively,　and　the　elevation　accuracy　decreases　with　the　increase　of　elevation,　slope,　relief　amplitude,　and　vegetation　coverage.　After　correction,　the　Mean　Error　(ME)　of　ASTER　GDEM　is　reduced　to　-0.233　m,　which　means　the　negative　deviation　is　effectively　removed　and　the　overall　MAE　and　overall　RMSE　are　reduced　by　26.04%　and　23.56%,　respectively.　The　MAE　and　RMSE　of　DEM　for　cultivated　lands,　forests,　grasslands,　wetlands,　water　bodies,　and　man-made　surfaces　are　all　reduced　by　different　degrees.　The　DEM　accuracy　evaluation　and　correction　method　proposed　in　this　paper　models　the　non-linear　relationships　between　multiple　feature　elements　and　terrain　errors　and　achieves　better　correction　results　in　the　study　area.　©　2023　Journal　of　Geo-Information　Science.　All　rights　reserved.