高精度DEM重采样及其对土壤侵蚀模拟的影响

发布时间：2018-07-08 18:29

本文选题：沟道侵蚀 + 坡面侵蚀　；参考：《西华师范大学》2017年硕士论文

【摘要】：DEM是土壤侵蚀研究的重要基础数据源之一,是求算侵蚀因子如坡度、坡度坡长因子、径流强度指数等的数据源。一直以来,DEM在土壤侵蚀研究中不断应用和发展,不同比例尺DEM数据不断丰富,促进了基于DEM的土壤侵蚀研究在模拟实验、小区实验、小流域实验、流域乃至区域尺度的广泛开展和应用。作为派生多尺度DEM的常用方法之一,重采样得到广泛应用。然而,现有基于DEM重采样的研究主要集中于重采样的数据精度变化、重采样高程误差及其分布特征等的研究。对于不同形态、不同坡度的地形而言,是否存在不同的最佳重采样方法这一问题仍有待探讨;此外,三种重采样方法中,最邻近(Nearest Neighbor, Near)、双线性(Bilinear Interpolation,Bil)及三次卷积(Cubic Convolution,Cub)方法得到新的DEM之后,对土壤侵蚀模拟存在什么影响?三种采样方法得到的DEM模拟的土壤侵蚀结果相似性如何?对于不同形态的地形,这种相似性是否会发什么变化?是否存在面向土壤侵蚀模拟的最佳重采样方法?这些问题都值得进一步研究和探索。本研究主要基于1-9期不同形态的黄土模拟小流域,分析了三种重采样方法带来的高程中误差的差异特征。结果表明:中误差按大小排序为:Near、Cub、Bil。推荐采用Bil重采样方法。三种重采样高程中误差(y)与格网大小(x)之间的关系可以用y=kx+b型线性函数关系表达。不论哪一种重采样方式,其派生DEM的高程中误差(RMSE)与格网大小(g)、地形平均坡度(s)之间存在函数关系:RMSE=(a×s-b)×g±(c×s d)。式中,a、b、c及d都是常数,其因重采样方法的不同而不同。以Bil重采样为例,其表达式为:RMSE=0.0116×g×s-0.0756×g+0.0676×s -0.7307。该公式以定量的形式,明确了地形表面形态、格网大小及其与重采样之后的DEM高程中误差的关系,有助于在实际运用中依据产生DEM质量要求的不同,根据高程中误差坐标、地形平均坡度等参数,确定合适的重采样方法和格网大小。基于USPED模型,模拟1-9期的潜在土壤侵蚀。对原始10mm×10mm格网的DEM的模拟结果表明,1-9期黄土模拟小流域沟道侵蚀极值在3、5、7期得到极大值;可以据此将9期小流域分为三个不同阶段,大致对应发育初始期、活跃期、稳定期。当采用原始DEM模拟结果作为参照值时,三种重采样DEM的模拟结果都存在约20%的概率,将潜在的侵蚀、沉积错误地模拟为沉积、侵蚀;随着重采样DEM的格网大小由15mm×15mm增大到55mm×55mm过程中,误分概率大致由18%增长至约30%。随着格网尺寸增大,模拟得到的土壤侵蚀、沉积的极值不断降低,坡面侵蚀比沟道侵蚀降低的速度更快。三种重采样方法重构的DEM,在约30mmX 30mm格网DEM左右,其模拟的侵蚀、沉积极大值迅速降低到参考值的50%以下。沟道侵蚀极值、沉积极值降低至参考值的比例(y)、格网大小(g)、地形坡度(平均坡度s或最大坡度sm)之间存在定量关系,其参数随重采样方法的不同而不同。以Cub重采样为例,沟道侵蚀极大值降低至参考值的比例(y)、格网大小(g)及地形平均坡度(s)之间的关系为:y= (0.0098 ×s2-0.517 ×s + 8.9743) ×e(0.0002×s2-0.0095×s+0.182)×g;坡面侵蚀极大值降低至参考值的比例(y)、格网大小(g)及地形平均坡度(s)之间的关系为:y= (0.0187 ×s2 - 0.9944×s +15.207) × e(0.0003×s2-0.0141×s+0.2395)×g。该公式明确了已知地形平均坡度前提下,不同尺度DEM模拟的坡面或沟道侵蚀极大值、沉积极大值与格网尺寸的关系,可以为多尺度土壤侵蚀模拟极值估算提供参考。利用X/Y散点图、相似系数等对比了重采样前后1-9期黄土模拟小流域土壤侵蚀的差异。结果表明,与原始结果相似性的程度与格网大小有关,格网尺寸越接近原始格网大小,得到的结果越相似。三种重采样方法模拟结果与原始DEM模拟结果相似性按从大到小排序为:Cub、Bil、Near。三种重采样方法之间,Cub与Bil相似度很高;二者与Near模拟结果相差较大。当DEM格网不断增大时,三者模拟结果的相似性不断增强,但偏离真实值也越来越远。
[Abstract]:DEM is one of the important basic data sources for soil erosion research, and it is a data source for calculating erosion factors such as slope, slope length factor and runoff intensity index. DEM has been continuously applied and developed in soil erosion research, and the different scale DEM data are constantly enriched, and the soil erosion research based on DEM has been promoted in simulation experiments. Experiments, small watershed experiments, basins and even regional scales are widely used and widely used. As one of the commonly used methods for deriving multiscale DEM, resampling is widely used. However, the existing research based on DEM resampling mainly focuses on the study of data precision change, resampling error and distribution characteristics of resampling. The problem of whether there is a different optimal resampling method is still to be discussed whether there are different optimal resampling methods for different slope terrain. In addition, three kinds of resampling methods, the most adjacent (Nearest Neighbor, Near), bilinear (Bilinear Interpolation, Bil) and the three convolution (Cubic Convolution, Cub) method get the new DEM, the soil erosion simulation exists What does the similarity of soil erosion results from the DEM simulation obtained by the three sampling methods? Are there any changes in this similarity for different forms of topography and the existence of the best resampling method for soil erosion simulation? These problems are worth further research and exploration. This study is based on the 1-9 stages of different shapes. In the Loess simulated small watershed, the difference characteristics of elevation error caused by three kinds of resampling methods are analyzed. The results show that the middle error is classified as Near, Cub, Bil., and Bil resampling method is recommended. The relationship between the three resampling elevation error (y) and the grid size (x) can be expressed by the y=kx+ B linear function relationship. Which method of resampling has a functional relationship between the elevation error (RMSE) of the derived DEM and the grid size (g) and the average gradient of the terrain (s): RMSE= (a * S-B) X G + (C x s d). 0.7307. the formula, in the form of quantitative, defines the surface morphology of the terrain, the size of the grid and the relationship between the error of the DEM Gao Cheng after the resampling, and is helpful to determine the appropriate resampling method and grid size according to the differences of the DEM quality requirements in the actual application, according to the error coordinates of Gao Cheng, the average slope of the terrain and so on. The USPED model is used to simulate the potential soil erosion of the 1-9 phase. The simulation results of the original 10mm x 10mm grid DEM show that the 1-9 loess simulated small watershed gully erosion extremes get maximum value in the 3,5,7 period. According to this, the 9 stage small basins can be divided into three different stages, which roughly correspond to the initial stage, the active period and the stable period. When using the original DEM simulation As a reference value, the simulation results of the three kinds of resampling DEM have about 20% probability, and the potential erosion and deposition is missimulated as deposition and erosion. As the grid size of the resampling DEM is increased from 15mm x 15mm to 55mm * 55mm, the error probability is roughly increased from 18% to about 30%. with the grid size increasing, the simulated soil is obtained. The extremum of soil erosion is decreasing, and the slope erosion is faster than the channel erosion. The reformed DEM of the three resampling methods is around the 30mmX 30mm grid DEM, and its simulated erosion, the active heavy value is rapidly reduced to less than 50% of the reference value. The channel erosion extreme value, the ratio of the sink positive value to the reference value (y), the grid size (and the size of the grid), and the grid size (y) G), there is a quantitative relationship between terrain slope (average gradient s or maximum gradient SM), and its parameters vary with the resampling method. Taking Cub resampling as an example, the maximum ratio of channel erosion to the reference value (y), the relation between the grid size (g) and the average gradient of the terrain (s) is y= (0.0098 * s2-0.517 x s + 8.9743) * e (0.0002 * s). 2-0.0095 * s+0.182) x g; the ratio of the slope erosion maximum to the reference value (y), the relation between the grid size (g) and the average gradient of the terrain (s): y= (0.0187 * S2 - 0.9944 x s +15.207) x E (0.0003 * s2-0.0141 * *) by the formula, which defines the slope surface or channel of different scales under the premise of the known terrain average slope. The relationship between the maximum erosion value, the positive heavy value and the grid size can provide reference for the estimation of the extreme value of the multi scale soil erosion simulation. Using the X/Y scatter plot and the similarity coefficient, the difference of soil erosion in the 1-9 loess simulated small basins before and after the resampling is compared. The results show that the degree of similarity to the original results is related to the grid size, and the grid is related to the grid size. The closer the size of the original grid is, the more similar the results are obtained. The similarity between the simulation results of the three resampling methods and the original DEM simulation results is between the three resampling methods: Cub, Bil, and Near., the similarity between the Cub and the Bil is very high; the two is different from the Near simulation results. When the DEM grid continues to increase, the three analog junctions are increased. The similarity of fruit is increasing, but the deviation from the true value is farther and farther away.
【学位授予单位】：西华师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S157.1

【参考文献】