Wed, Aug 04, 2021:On Demand
Background/Question/Methods
Wetlands provide a range of ecological services that are critical to human beings, such as flooding regulation, water quality improvement, wildlife habits, etc. The importance of wetlands is usually related to their spatial connectivity with nearby streams within a watershed ecosystem. Scientists commonly combine wetland and stream maps to detect the hydrological connectivity of wetlands. However, the delineation efficiencies rely on the accuracy of input data and the feasibility for automated mapping of fine-scale aquatic features. Recently, the availability of remotely sensed data offers a tremendous advantage in digitalizing landscape features, allowing us to investigate hydrological variability based on detailed morphological information over a large area. This study aims to develop a new semi-auto topographic-based model by combining Random Forest (RF) and linear feature extraction to delineate water-stream connectivity. Fifteen topographic metrics extracted from high-resolution LiDAR-derived DEM were used as inputs for the model. We tested the model in both low-relief and high-relief watershed to evaluate the feasibility of the model at different geomorphologic conditions.
Results/Conclusions This study suggested that the topographic-based RF model is capable of providing an efficient linear delineation in both high and low-relief watersheds. Generally, high misclassifications are reported in flat-terrains due to the low discrimination between flow paths and their surroundings. However, our model can provide relatively high performance with an accuracy of over 88%, even in the low-relief watershed. The derived wetland-stream connectivity can be scientific support for better water and soil conservation and wetland management. Results of this study will also benefit watershed modeling, which is focused on the impacts of wetlands on hydrology and biogeochemistry.
Results/Conclusions This study suggested that the topographic-based RF model is capable of providing an efficient linear delineation in both high and low-relief watersheds. Generally, high misclassifications are reported in flat-terrains due to the low discrimination between flow paths and their surroundings. However, our model can provide relatively high performance with an accuracy of over 88%, even in the low-relief watershed. The derived wetland-stream connectivity can be scientific support for better water and soil conservation and wetland management. Results of this study will also benefit watershed modeling, which is focused on the impacts of wetlands on hydrology and biogeochemistry.