A novel method for filling the depressions in massive DEM data

It has been universally recognized by the previous researches that depressions in DEM data are the major obstacle of determining hydrologic flow directions, so numerous methods were proposed to address this problem. However, when massive DEM data, such as SRTM and GTOPO30, become available through internet, the conventional approaches are too time- consuming and inefficient in dealing with such massive data for hydrologic analyses. A new innovative approach, unlike all existing methods that process DEM data straightforward without utilizing the topographic features implied in them, was proposed in this study to improve the DEM-processing efficiency. The new approach has two steps: first, classify the initial DEM data into eight groups according to the elevation values using the quantile classification method, and set each category cell values to its quantile and store them in a transient matrix. Second, scan the transient matrix from the minimum category to the maximum one, restore its initial value if its initial value is larger than or equal to its neighbors\´, else set its value to the minimum of its neighbors\´ if its quantile is larger than its neighbor\´s, and repeat this process until all the depressions are filled. Furthermore, unlike the traditional method, which have to scan all the DEM cells in each loop, the new one only scan the data needed to be further processed by storing their location information into two stacks. As a result, the total cells and scanning times are dramatically decreased due to stack\´s nature of "first in last out", and the location information of depressions is all stored in one of the two stacks, which facilitate the subsequent hydrologic analysis. DEMs of various sizes were used for testing and evaluating the new method. The results show the new method is thousands of times faster than the famous Jenson and Domingue\´s (1988) method and is over seven times faster than Planchon and Darboux\´s on average The new algorithm is fully- detailed and pseudo-codes are provided in this paper.