How do Water Quality Models Respond to Growing vs. Non-growing Season Landuse Inputs? An Exploration with Google Earth Engine and Dynamic World
Abstract
Studies relating water quality to landuse/landcover (LULC) typically use an LULC input representing growing season conditions. Now, real-time remotely sensed LULC products make it easier to study non-growing season LULC-water quality impacts. We used Dynamic World LULC data from Google Earth Engine, and a long-term water quality dataset for 37 streams in the National Park Service National Capital Region, to compare how water quality models respond to growing vs. non-growing season LULC inputs. We included simple regression models and a more complex Soil and Water Assessment Tool (SWAT) that has built-in seasonal processes for leaf area and crops. Watersheds typically had more tree cover during the growing season by approximately 5-10% of watershed area, and less built area and shrubland, possibly due to actual vegetation changes and/or LULC classification inaccuracies. Stream water specific conductance and concentrations of nitrate and total phosphorus were positively correlated with built LULC. Simple regression models comparing growing and non-growing season LULC with these parameters did not show large differences in relationships between seasons. However, an uncalibrated SWAT model for the Bush Creek watershed, Maryland, with a large 20% change in LULC proportions between seasons (possibly due to more visibility under deciduous trees in mixed-LULC areas), simulated greater surface runoff (+19%), soluble phosphorus (+42%), and nitrate yield (+51%), when non-growing season instead of growing season LULC input was used. This was over the same period objectively keeping the same parameters, but there was 8% more built area in non-growing season LULC data, which increased modeled runoff. We conclude that when large sub-annual changes in LULC data occur, water quality models using growing season LULC input could be substantially different from those using non-growing season LULC, because of their sensitivity to watershed landcover proportions rather than differences in simulated processes. This is important to know for future modeling of water quality impacts with different seasons of LULC data, as real-time LULC can help identify water quality discrepancies with high temporal precision. Finally, our SWAT model has a new lookup table to read Dynamic World LULC data that is transferable to other studies.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H13D..01M