I’ve run the SWY model using the contribution areas relative to stream gauges stations so I would have data to compare monthly Quickflow data. After compiling the measured data in monthly averages, I transformed the data from m³/s to mm/month, using the formula:
Qf(mm/month) = (Qf(m³/s)/basin area (m²))x1000x3600x24x31 (or 30 or 28, depending on the month).
To get monthy MODELED Qf data, I simply summed all the Qf pixels in the basin, so as to get total Qf per month.
Although I got fair correlation between modeled Qf and measured Qf, the order of magnitude of modeled Qf is very different, being in the magnitude of millions of mm, whereas the measured data is in the magnitude of thousands of mm.
Can someone help?
Just curious, were you able to resolve this concern?
I am also doing a validation with the observe versus the model results . I also computed the total water yield in the basin using the observed flows per month through the same formula as yours and made it as a benchmark. I compared with the sum of all the pixels in Qf plus pixels in B results and it seems the SWY model calculation is very low as compared to the observed. I tried adjusting the alpha (1/12 to 12/12) and beta (1, 0.01, 0.05, 5, 10, etc) but the results are still very low even with extreme values. However, when I used alpha = 1/12, beta = 1, but adjusted the gamma to 5, it seems the SWY results are getting close to the observe.
I’m not sure if this is the right approach since I don’t see literature that changed the gamma values in SWY but it would really be helpful to know if this kind of approach makes sense?
Any thoughts on this would be really helpful. Thanks!
Hi @carlureta and @Victor_Maluf -
First, apologies for the lack of response to your previous post, Victor.
We are actually working on a similar process for a NatCap project at the moment, and while I’m not a hydrologist (nor the model author), I’ll throw out a few ideas for consideration.
Regarding the B results, note that although the values are given in millimeters, it is recommended to use this as an index of where more or less baseflow is produced, it’s not intended to provide an accurate absolute value.
That said, when using either B or Qf absolute values (and I have to admit that we’re also seeing what happens to use B values along with Qf to do a calibration with observed data), since the results are in millimeters, if we simply sum these up over the whole area, the result is likely to be huge, and doesn’t represent the total water volume properly. Instead, we are using the mean B or Qf value across the watershed, converting mm to meters, then multiplying by watershed area to get a value in cubic meters, which we’re comparing against observed flow data.
As part of the calibration, we’re also doing a sensitivity analysis, changing some of the parameters, including gamma, to see how they impact the results.
Thanks for that.
I did what you suggested using the mean and multiplying to the area and seems it’s still too low. However, the gamma coefficient though seems to be driving things closer to the observed. I don’t know if this will also be true for your case, but so far in my case, the gamma values between 2-2.5 gets me closer to the observed.
Hope this info will also be useful for others.
Dear @carlureta and @swolny,
Thank you for you response! Unfortunately, I haven’t solved this problem, though I haven’t tried calibrating the model by adjusting alpha, beta and gamma parameters due to lack of time. I did try a wide range of formulas to transform mm/year to m³/s, but without success in achieving the same order of magnitude between observed and modeled flow. Since I’m trying to use the InVEST pack (mainly SDR and SWY) for guiding environmental policy makers, sometimes I work under pressure to present tangible results, even if these are not ideal from a scientific method perspective. So I confess that I switched to the Annual Water Yield Model, which already gives the total annual discharge output in m³/s. Even without propper calibration, the AWY model can give me a rough behavior of the watershed in different land use scenarios in comparable discharge units. Nonetheless, it is in my plans to do a sensitivity analysis of the calibration parameters.
@carlureta (I aslo add here @swolny) , it would be nice to hear if you solved the problem.
Do you know what could be the underliyng factor for improvments in the modeled flow when you change the gamma parameter?
I wonder if this has relation to the fact thet the curve number method does not include topography and that the SWY model is based in the curve number method calculations.
I am having the same issue of underestimation (a tenfold underestimation).
Hi @Victor_Maluf -
Hopefully you tried calculating the monthly Qf data differently, since simply summing Qf will indeed give very high values, which do not represent total flow properly. As noted later in this thread, you’ll need to either use the mean Qf value in the watershed and multiply by watershed area, or calculate volume for each pixel and sum the volume.
As for Annual Water Yield, juts a note that the model itself does not produce an output with units of m3/s. Its main wyield output is in mm, and if you do the second step of water scarcity, results are in m3/year, not m3/second.
@felipebenra, it’s interesting that your modeled values are much too low. Just making sure that your input precip and ET0 have units of mm/month. Are you only using quickflow for comparison? And, what is the observed data that you’re comparing against? It can get tricky to compare this model against observed, since in real life, flow will usually be some combination of quickflow and baseflow, yet we’re separating them, and baseflow in particular isn’t meant to be quantitative.
I am not sure if it was because the curve number method does not include the topography, it maybe so. In my case, I did a trial and error to get into the benchmark value as close as possible. alpha was default but beta and gamma was through trial and error combination. It seems their combination makes a significant difference. I ended up using alpha = 1/12, beta = 1.5, gamma = 2.3492 or 2.3493.
Stacy is right, you need to convert the results first to unit/time/area. In my case, I used a m/s/pixel (where 1 pixel = 1sqm). The process for that computation was also mentioned by Victor in this thread.
Hope this helps.
@swolny Thanks for your answer, yes my input precip and ET0 are in mm/month and I am using only QF to compare. Re observed data to compare aginst I am using runoff data (also in mm/month).
@carlureta Also for you a big thanks. I think I am handling ok my units, but I will try to use diferent parameters. I calculated the alpha based on the antecedent precipitation, so I don’t use default 1/12. In the case of beta I also calculated it thorugh the wetness index (shortly described in the SWY model page) and I came up with a value of 6.32. However running the model with those changed parameters didn’t change much. I am going to try your gamma value.
But I am curious about the kind of basin where you applied that gamma parameter.
If you don’t mind to tell, I would like to know how large is the basin and what main tophographic conditions does it have (e.g. hilly, flat, slope)?
The watershed area is 54,400 acres. With elevation range from 536 to 1166 masl and slope range from 0 to 63 degrees, mean slope is 10 degrees but with sd of 10 degrees as well. This is a pretty slopey area.
When I did a validation in a lower watershed with a flatter area, the gamma values that I had still works. But maybe this will differ in very different geographic areas because both watershed is within the state of South Carolina. One is upstate, and one is lowland area.
Hope that helps.
@carlureta Thanks for your answer!