The Study Of Improving DEM Accuracy Based On Multi-source Data Fusion For The Typical Gully In Loess Plateau

The development of Digital Elevation Model(DEM)has been playing an important role in human daily life and working.Improving the accuracy of DEM is one of the important fields for digital terrain analysis.The unique landform feature of the Loess Plateau provides us a perfect region to discuss the method of accuracy improvement.With the public download of multi-source data,more and more researches spring based on these data.And the combination of multi-source data is a meaning attempt to uplift DEM accuracy.In this paper,five typical geomorphic study areas in the Loess Plateau in Shaanxi were selected as the research sample area,and 5m DEM data was used as the basic data.The data of three different data sources,ALOS,SRTM and ASTER,were obtained and processed uniformly through the previous registration.Three methods of GS,weighted fusion,and feature point embedding fusion are used for mutual enhancement of the three data.Taking Suide sample area as an example,the results are as follows:(1)When the ENVI GS method is used to fuse every two different data,the Root Mean Square Error(RMSE)value after fusion is basically between the error value in the data before fusion.The RMSE is reduced by 0.38-5.674 m,and the maximum and minimum error values are optimized by 2-14 m and 32-65 m,respectively.Both the RMSE and the error extreme value have been optimized.In terms of the results after fusion,the fusion effect of ALOS and SRTM is the best,followed by ALOS and ASTER,and the fusion effect of SRTM and ASTER is poor.(2)According to the difference of slope,positive and negative terrains,the samples is divided into three parts: smooth gully bottom,steep area and smooth gully top.According to the different errors of each part,the weight is calculated and then,which is the key step to fuse DEMs.The results show that the median error and error extreme value of the fused data are lower than the median error of the better data in the original data,and improving the accuracy of the original data is very obvious.Among them,in the flat area,the fusion effect of ALOS and SRTM is the best,and the accuracy is improved by 6.432 m and 7.923 m,respectively.In steep areas,the highest accuracy of ALOS and ASTER is up to 9.8m,the median error is reduced by 5.817 m,8.917 m and 5.689 m respectively.The improvement of error extreme value is satisfied as well.(3)Embed the feature points on the ridge line and valley line in the basic data into the three different data of Suide sample area.The results show that the median error and the error extreme value are significantly lowed compared to the data features before fusion.The RMSE is reduced by 1.4-6.73 m,which is of obvious improvement.Among them,the median error value after ALOS fusion is the lowest,and the accuracy is up to 8.55 m.The error extreme values in ASTER are optimized by 6.731 m,20.588 m,and 54.67 m respectively,and the improvement of error extreme value is the best.Combining the effect of three fusion methods in the five sample areas,and compare the eigenvalues before and after the fusion.The main conclusions are as follows:(1)The fusion method of ENVI GS is convenient and simple,and the three combined fusion methods can be better improved.Generally speaking,the fusion of ALOS and SRTM data is the best,but it is greatly affected by the original data.(2)By dividing the sample area into different areas and fusing them separately according to the weights occupied by each area,the accuracy of the data obtained is significantly improved.Comparing the accuracy improvement in each region,it can be found that the fusion of ALOS and SRTM rely on the gentle area.A high accuracy of these two data will receive a better fusion.When ALOS and ASTER are merged,the accuracy improvement is the best.and in areas with large fluctuations,better results can be obtained.And when SRTM and ASTER are merged,the improving is relatively stable with little difference.(3)By embedding feature points into three open-source DEM data,the errors and error extreme value of the data obtained by fusion are significantly improved.Overall,the fusion result is embedded with ALOS data.The original eigenvalues of ASTER are all inferior,and the improvement of the error and the error extreme value after fusion is the most obvious.