Calibration of Water Yield Model

Lyndsy

Dear InVEST users and developers,

I have set up my water yield model but am unsure about how to proceed with calibration. Can values for long-term average streamflow be compared directly to the water yield results once the units are converted?

In my case, I have three guages in a watershed with 17 subcatchments, including a final one at the outlet. Would it be appropriate to use a cumulative subtotal of the model values for comparison at each of the guages or how would you suggest proceeding?

Your suggestions and guidance about the calibration procedure would be greatly appreciated! Thank you!

swolny

Hi Lyndsy -

I have a bit of experience with calibration, but am not a hydrologist, so they may wish to weigh in here. Also, is this for the annual or seasonal water yield model? I'll assume annual for the moment.

Yes, you will want to use long-term average annual observed streamflow, and it's great if you have gages with consistent, daily flow, preferably over at least 10 years that overlap the year of your LULC map (as well as your precipitation/ET climate inputs.) You'd then create the watersheds that drain into the location of your gages, and use them to summarize the per-pixel wyield.tif and compare that with your observed data at those gage locations.

There are other things to consider, such as whether any water is being removed from the stream before it makes it to the gage. (For a municipal water supply, or water transfer to another location, etc.) If so, you should subtract that from the summarized model results before comparing. 

~ Stacie

Marye

Hello Stacie;

How about if it is seasonal water yield. What I mean is how can I calibrate seasonal water yield and validate the result of the model with   long-term average  observed streamflow data.

Thank you!!

Marye 

Rich

Hi @Marye, FYI I forwarded your question to our hydrologists. I think they will post something here soon.

Marye

Thank you very much @Rich!!!

Perrine

Hi Marye, 

This is a good question. As background, here is a reminder of the tool's scope and objectives: "The InVEST seasonal water yield model seeks to provide guidance regarding the contribution of land parcels to the generation of both baseflow and quick flow. The model computes spatial indices that quantify the relative contribution of a parcel of land to the generation of both baseflow and quick flow. Currently, there are no quantitative estimates of baseflow (only the relative contributions of pixels); a separate tool is in development to address this question." (from User's guide)

One subtly here is that on the one hand, the model uses a physics-based approach to modeling baseflow (computing a water balance at the pixel scale, and then routing the water downgradient). On the other hand, the equations are extremely simplified (at the temporal and spatial scales), which increases significantly the uncertainty on the absolute numbers.

This is why we do not suggest to use the absolute values, but mainly the relative values across the landscapes (where we assume that the simplifications matter less, because they apply to the entire landscape).

Now of course, model validation or ground-truthing is always a good thing to do. What I'd recommend is to validate the relative values (i.e. the distribution of values across the landscape). This requires several (at least >3, more realistically >5) stream gauges, which you could compare with the baseflow generation output of the model.

Alternatively, you can compare your results to a different spatially-explicit model, if it is available. This is what we've done in a study that is currently under review (happy to share more info if you are interested).

I hope this helps, let me know if you have questions on the theory or the practical steps of validation.

Best,

Perrine

Marye

Thank you very much @ Rich and Perrine