Results & Discussion


Of our four selected watersheds, we were able to successfully run three, the Noyo, Trinity, and Eel. Of the three watersheds we modeled, the Noyo watershed resulted in the highest output average for NSE accuracy, with an average of 0.78 and a peak of 0.94 in the year 2010.The Eel had an average NSE accuracy of 0.31, with the peak coefficient being 0.54 in 1998. The Trinity had the lowest NSE accuracy, with an average of -3.44 and a peak of 0.28 in the year 1997. The Russian River watershed simulation experience an unexpected error and has not yet been successfully run.
This chart represents the annual Nash-Sutcliffe score for each watershed. This legend is interactive please click on each watershed. 

The Noyo watershed was the most accurate due to multiple factors. The Noyo has by far the smallest drainage area, only spreading across 106 sq. miles, while comparatively the Eel River watershed has a drainage area of 3,113 sq. miles and the Trinity’s drainage area is 2,936 sq. miles. The Noyo River is also the least changed by humankind, with no dams or major diversions unlike the Eel and Trinity.

The smaller drainage area allows for the digital elevation model (DEM) and sequential files for the Noyo to be run at higher resolutions compared to the other watersheds. The Noyo was ran at 80 meter pixel resolution, whereas the other watersheds were run at 300 meter pixel resolution. There is also likely more accurate precipitation and temperature inputs, since smaller regions will have less climatic variability than a watershed stretching thousands of miles.

To increase overall accuracy of both the Eel and Trinity Rivers, multiple different approaches could be taken. Increasing the resolution of the input DEM and other ASCII files could increase the overall accuracy of the model. Having a higher resolution DEM would produce a more accurate stream network for VELMA-2.0 to process.

VELMA does not account for diversion/disturbance, and instead models the watershed based solely on the input data parameters. On both the Trinity and Eel, water is diverted and dammed for domestic and agricultural use, meaning a significant amount of water within the watershed is not flowing all the way to the pour point. Because of this, VELMA’s simulated flow results were higher than gauge station flow rates. Additional analysis on multiple pour points/gauge stations could help improve scores and analysis. VELMA could be run using a gauge station located before a large water diversion, such as the Trinity Dam located outside of Weaverville in the Trinity watershed. Running the simulation with different pour points could potentially improve the simulation by looking at flow rate located before a drastic change in flow.

Lastly, while some rivers may not be able to reach an acceptable Nash-Sutcliffe value, by analyzing the simulated flow rate in comparison to the actual gauge station flow data, we can still better understand the hydrology within the watershed and gain insight, providing a basis for future analysis on water diversion.

Next: Check out the results for each watershed
Goals for Future Research:

Successfully Model the Russian River Watershed. The original goal for the project was to run VELMA on four Northern California watersheds. Since we were able to successfully run three, running the Russian River would be a great achievement. Of our future goals, this would likely be the one we would focus on next. We are currently experiencing problems with the model getting stuck/the simulation freezing after running through the first year.

Improve our Nash-Sutcliffe Scores. Trying improves our Nash-Sutcliffe results for the Eel River is another goal for the future of the project. In addition to increasing accuracy, it would be interesting to model analyze the Eel Watershed in relation to legal and illegal cannabis cultivation within the watershed.  Cannabis cultivation is a huge source of water diversion that could potentially impacts to the flow rate and stream chemistry of the watershed.

Model Additional Watersheds. In the future, we would like to use VELMA to model other watersheds, specifically in the Pacific North West. This would allow us to look at more watersheds with varying sizes and amounts of human disturbance and analyze the results. Two larger watersheds we would be interested in modeling are the Snake and Columbia watersheds. Analyzing these watersheds, specifically looking at the flows at the confluence of the two rivers, could provide interesting results and insight on water use and quality.

Click on a watershed to learn more!

Noyo River Watershed
Eel River Watershed
Trinity River Watershed
Russian River Watershed