TL;DR
This paper introduces a novel method for dynamically delimiting sub-watersheds using a "no-fill" Digital Elevation Model (DEM) and defined precipitation levels. By treating sink areas as potentially overflowing features rather than filling them, this approach better preserves natural terrain characteristics, leading to more accurate flood management insights and a stronger correlation with observed waterlogging points compared to traditional methods.
What Problem Does This Paper Address?
Accurate watershed delimitation is crucial for effective flood control and management, especially in urban areas prone to waterlogging. Traditional methods often rely on filling Digital Elevation Models (DEMs) to create continuous flow paths, which can artificially alter the natural "sink" areas (low-lying regions that collect water). This alteration can misrepresent the actual storage capacity of sub-watersheds, making them less suitable for studies sensitive to these storage dynamics. The challenge lies in developing a delimitation method that respects the natural topography, particularly the behavior of sink areas under varying rainfall conditions, to better predict and manage urban flooding.
Methods
The core innovation of this paper is a "dynamical watershed delimitation" method that utilizes a no-fill DEM and defined precipitation levels. Instead of filling depressions in the DEM, the authors treat them as integral parts of the terrain.
Key methodological steps include:
- No-Fill DEM Flow Direction Calculation: This involves calculating flow directions by considering all adjacent cells with lower elevations as potential outlets. A unique aspect is the assignment of a "0" value when no outward flow is possible, identifying potential sink areas.
- Basic Unit Identification: Cells with a "0" flow direction and cells that flow into them are identified as "original basic units." These units are then logically treated as a single "big cell" for flow accumulation purposes.
- "Source Flood" Analysis: The method incorporates a "source flood" concept, where sink areas can overflow when their internal water level exceeds the elevation of their outlets. This is dynamically determined by the input precipitation.
- Precipitation-Driven Merging: Basic units are merged based on defined precipitation levels. If the precipitation exceeds a unit's "limitation precipitation" (its storage capacity relative to its outlet), it can merge with neighboring units, forming larger sub-watersheds.
- Flow Accumulation: This step traces flow paths, considering both regular cells and the "big cells" representing merged basic units.
This approach contrasts with traditional "fill-DEM" methods and the standard D8 algorithm, which often simplify or alter sink area behavior.
Datasets
- Digital Elevation Model (DEM): The study utilized the AW3D30 dataset, chosen for its high accuracy and ability to represent detailed terrain features, even in complex areas.
- Precipitation Data: Four levels of short-duration rainfall intensity were used, corresponding to 1-year, 5-year, 20-year, and 100-year return periods (73.6 mm, 138.1 mm, 205.6 mm, and 336.2 mm, respectively, for a 12-hour duration). This data was sourced from previous studies on Wuhan's rainfall patterns.
- Waterlogging Points: 40 typical waterlogging points within Wuhan's third ring road, recorded during a specific flood event (June 30 - July 6, 2016), were collected. This dataset served as a crucial validation benchmark.
Key Insights and Ideas
- Sink Areas as Dynamic Features: The most significant insight is treating sink areas not as static depressions to be filled, but as dynamic features that can overflow under sufficient rainfall. This "source flood" concept is key to capturing realistic hydrological behavior.
- Preserving Natural Topography: By avoiding DEM filling, the method preserves the subtle elevation differences that define low-lying and flat areas, which are critical for understanding water storage and potential waterlogging.
- Precipitation as a Delimitation Driver: The study demonstrates that precipitation intensity is a powerful factor in dynamically shaping sub-watershed boundaries. Higher rainfall leads to larger, more interconnected sub-watersheds as isolated units merge.
- Improved Correlation with Reality: The proposed method's sub-watershed delineations showed a significantly higher correlation with actual waterlogging points than those derived from the traditional D8 algorithm. This suggests its practical utility for flood risk assessment.
Key Results
- Reduced Isolated Units: As precipitation increased from 73.6 mm to 336.2 mm, the ratio of isolated basic units (those unable to flow out) decreased significantly, from 27% to 1%. This confirms the "overflow" hypothesis.
- Increased Sub-Watershed Area: With increasing precipitation, the average sub-watershed area also increased substantially. For instance, the average area for 73.6 mm precipitation was 187.07 cells, while for 336.2 mm, it rose to 2038.45 cells.
- Better Waterlogging Point Correlation: The proposed method's sub-watersheds showed a more consistent overlap with observed waterlogging points compared to the D8 algorithm. Specifically, 91% of "all cars cannot pass" and 100% of "small cars cannot pass" waterlogging events occurred in sub-watersheds identified as "not containing range of river" by the proposed method, suggesting a better localization of risk areas.
Limitations and Future Work
- DEM Resolution: The study acknowledges that while AW3D30 is high-resolution, urban catchments can have complex terrain features that might benefit from even more refined DEMs. Coarser resolutions can lead to a large number of small basic units.
- Land Cover and Drainage Systems: The current method does not explicitly differentiate runoff production based on land cover types or account for changes in flow direction due to engineered drainage systems.
- Future Directions: The authors suggest future work should focus on incorporating these factors to improve the accuracy of sub-watershed delimitation, particularly for small-scale urban environments, and enhancing quantitative effectiveness.
Summary updated