Research On Endorheic Drainage Networks Extraction Algorithm Based On DEM

As a unique type of geographical area,the endorheic has a wide range of distribution,and has great research value.The drainage networks not only reflect the features of topography and geomorphology,but also plays an important role in controlling the distribution of energy and material in surface.Therefore,the extraction of drainage networks in the endorheic is an important step in the study of hydrology and topography.At present,the traditional extraction methods based on the digital elevation model(DEMs)is not suitable for the endorheic,because of the theoretical basis of the exorheic convergence model.A series of data preprocessing steps such as filling depression and increment flat have damaged the unique terrain structure of the endorheic,resulting in the inability to extract an accurate and reasonable drainage network.In view of the above issues,based on the terrain-adapted drainage networks extraction algorithm,this paper proposes a corresponding improvement method to solve the specific problems in the application of inland area to meet the needs of drainage network extraction in endorheic basin.The specific research content and conclusions of this article are as follows:1.Design and implementation of drainage networks extraction algorithm in endorheicThere is no hydrological connection between the endorheic and the ocean,and its causes are complex(topography and climate,etc.).However,the topographic features of the depression in the endorheic determine that the internal rivers cannot flow to the sea and can only be eventually collected in the depression to form Lake.The extraction of the drainage networks in the endorheic needs to be carried out under the condition that the depression is preserved.Therefore,this paper selects the R&N algorithm that can handle the special terrain adaptively.The R&N algorithm by introducing water volume height to the calculation of flow direction,which can better simulate the water flow better,can complete the calculation of the flow accumulation without data preprocessing.Through theoretical analysis and experimental verification,it is found that the R&N algorithm and its related improved algorithms have the problem that the extraction result depends on the grid traversal order,which decrease the stability of the extraction results.This paper proposes an improved algorithm based on water synchronization update strategy(SU-TFM-md algorithm).Through the method of delay updating the water,all the grids are synchronized to update the amount of water and the calculation inconsistency caused by the instantaneous update of the water is eliminated.The experimental results of ideal surface show that the improved method in this paper can effectively eliminate the influence of grid traversal order on the extraction results.Compared with the results of D8 and MFD-md algorithm,the proposed algorithm can correctly express lakes and extract river channels in flat terrain.The generated river network is also more accord with topographic features.2.Design and implementation of parallel algorithm based on GPUThe SU-TFM-md algorithm requires a large number of iterations to simulate the continuous flow of water,and the calculation is time-consuming.Therefore,parallel computation is needed to improve the efficiency.This paper selects a GPU that is suitable for data parallelism as the parallel computing hardware platform for this study.The water synchronization update strategy of the SU-TFM-md algorithm solves the problem of inconsistent extraction results and boundary effects caused by the conflict of random scheduling mechanism of the CUDA thread with the grid traversal order in the R&N algorithm.Selecting different DEM dataset for parallel efficiency test and performance optimization.The results show that:when the thread block configuration is 16x 16,the parallel program is the fastest in GTX 1080ti.What limits the performance of the CUDA kernel in this paper is the register,which is limited by hardware resources.Compared with the serial algorithm based on single CPU core,the parallel algorithm using 3584 GPU cores can achieve an acceleration ratio of about 300.The parallel algorithm has obvious advantages in efficiency.3.Algorithm parameters selection method and empirical researchThe initial water height and iterative termination conditions are important factors that affect the extraction results of the SU-TFM-md algorithm.The experimental results show that the initial water height has no significant effect on the extraction results in the endorheic basin.Relatively speaking,iterative termination conditions are more important.Unlike the exorheic area,the algorithm should not use a fixed amount of water as the iterative termination condition in endorheic,and the depressions can cause the transfer water to fail to converge quickly.The experimental results show that when the number of iterations is greater than 500,the extracted drainage networks are stable.