I have successfully modeled SWR for the southeast and northwest regions of the US. I am now planning to validate the result with the observed data. Do you know any useful data/source or map for the US? I highly appreciate it if you could share those references/links.
Hello @hkarimi -
We have used historical data from USGS Water Data for the Nation - https://waterdata.usgs.gov/nwis. It’s a bit time-consuming to work with, you need to review the data to see how consistent and complete the data is on a monthly (or yearly) basis, and you’ll need to process the values to calculate average monthly (or yearly) values for comparison with the model. But it’s a good central spot to look for data, especially if you want to consider sites across a large area.
If anyone knows of observed US flow data that’s easier to work with, please share it!
Thank you once again for providing such valuable information. I’ve reviewed the historical data from the USGS Water Data, and they have numerous monitoring stations across my study area (perfect!). The question arises regarding is the USGS data are station-based. In contrast, the SWY results are pixel or watershed-based. Do you have any recommendations on how to match the station data with my results? Would it be advisable to calculate the average for each watershed basin and concurrently determine the average streamflow for all stations within that watershed?
Thank you again,
Hi Hazhir -
What we usually do for calibration is select one or more points of particular interest. Usually these are reservoirs or something, which often have gauge data. But if you don’t have particular sites of interest, you can choose whatever location(s) work for you. One way to choose them might simply be the ones with the longest/most complete station data.
Once we choose the site, we use the station point to generate the watershed draining to that point, and aggregate the model results within the generated watershed. Then the aggregated results can be compared with the station data at that point.
All of this makes it important that the river network generated by the DEM and model represents reality as well as possible, so the watershed that’s generated also corresponds with reality as well as possible.
Thank you again for the invaluable information. I will do that as you suggested. Thanks.
I have also other two questions about the total water yield in my region:
- I am working on the entire US southeast which is a pretty large area and modeled for 4 periods from 2000-2020 (SWY). I want to see how the total water yield in the entire region changed. Since each pixel of the raster (for my case 90*90 m) shows a value of water yield I used the sum of all pixel values as total water yield. Do you recommend this way as an approach to show the total water yield? The odd thing was total water from 2000 to 2020 doubled while the precipitation did not experience such a big change.
- at first, I modeled for every single year (2000, 2010, 2020) to show the spatiotemporal changes across my case study. But I found that there was huge spatial changes between 2000 and 2010 or 2020 and I think this is because taking into consideration only a year (e.g., only 2010 to 2020) does not show the real change. So, I decided to use an average of each five years as one period (i.e., I averaged precip and ET for 2001, 2002, 2003, 2004 and 2005 as period 1 and added to the SWY model. For other periods same approach). Do you have any comments about this approach?
Hi Hazhir -
First, I want to make sure that you’ve looked at the Calibration section of the SWY chapter of the User Guide, since it talks about your point #1.
For the second question, I got some input from our hydrologists, who made the following points:
It is very possible for there to be significant spatial and magnitude changes between time periods, for example if one period contains drought years versus wet years.
The time period of your precip/ET0/etc input data should match. For example, if you use precip for 2005 only, then use ET0 for that same year, if you use precip averages from 2005-2020, then use ET0 for that same period.
If you’re evaluating climate change over time, ideally the LULC should change through time as well (unless the goal is to isolate the influence of changing climate).
One of the limitations of the SWY model that it only uses a lumped parameter to control the lag of water flow from month to month, but it does not actually track water through time. This is important when considering change over time - the years do not build on each other.
For the purpose of evaluating change through time, it would be ok to break the rainfall record into shorter periods - say 5 or 10 years - and then compare SWY results between these periods. Even better, use a moving window approach of 10 years to see if there were changes through time. For example, driving the model with the average monthly P and ET of 1990-99, 1991-2000, 1992-2001, etc. This would avoid the problem of creating artificial breaks in the data of 5- or 10-year periods that may have apparently random changes in rainfall that are actually due to normal variability year-to-year.
Once again, thank you so much for your assistance.
First, I am really sorry for lots of question I have! In the previous message you said “Once we choose the site, we use the station point to generate the watershed draining to that point, and aggregate the model results within the generated watershed. Then the aggregated results can be compared with the station data at that point”. I can not understand well this statement well. Would you please explain a bit more of “generate the watershed draining to that point, and aggregate the model results within the generated watershed.” How do you do that?
Thanks a lot.
Hi @hkarimi ,
What @swolny is referring to is using the DEM to delineate or define the upslope catchment that drains to your point. To do this, you could use an InVEST helper tool called DelineateIt. Once you define a sensible watershed you’re happy with, you can sum the values of the pixels within that contributing watershed and compare those results to observed data from the station at that point.
Thank you so much for valuable guidance.
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.