I want to calibrate the Z factor of the InVEST water yield model. I have LULC data and corresponding climate data for 2002, 2012, and 2022, as well as long‑term observed streamflow.
First, I calibrated Z using only the 2002 dataset and found that Z = 13 gave a modeled water yield close to the observed streamflow. Should I use this same Z value (13) for 2012 and 2022? When I did, the modeled yield for 2022 was higher than the observed value, and I am not sure if this is the right thing to do.
I also tried finding a single Z that fits all three years together. Even when I increased Z up to 30, the modeled yield for 2022 was still higher than the observed value, although 2002 and 2012 was below and close to the observed.
How should I determine an appropriate Z factor when I have three different years of data (2002, 2012, and 2022)? Please i need an answer to this as soon as possible because it is not written in the inVEST guide.
Thank you.
Interesting question. I could use more information about your modeling setup:
I have LULC data and corresponding climate data for 2002, 2012, and 2022, as well as long‑term observed streamflow.
So are you using precipitation and ETo calculated separately for only years 2002, 2012 and 2022? Or are the climate layers calculated with a time series? For example, the 2002 layers calculated using data from 1992-2002, the 2012 layers calculated using data from 2003-2012, etc?
Also, are you using different observed streamflow data for each climate scenario, such that the observed data has been calculated for 2002, 2012 and 2022 separately?
In general, the annual water yield model is intended to be used with long-term annual average values. So we would recommend, for example, averaging over around 10 years of climate data to produce each of the climate scenarios, then comparing that with observed data that has been averaged over those same 10 years.
Thank you for your kind response. This has helped a lot. Let me clarify that I used time series climate data of 1993-2002 for the 2002 layer, and so I did for all other years. However, i want to ask if i am to use a single Z value for all the climate data years (2002, 2012 and 2022) after validating with my observed streamflow of 30 years long term average or i am to use three Z value i.e. validating my climate data years against streamflow data of 10 years long term average each e.g. using 1993-2002,…, 2013-2022 observed stream flow against all the climate years.
Moreso, if i am to use a single z value, will I keep the Z value that I got for first layer (2002) fixed after the water yield volume is closed to the 30 years average observed streamflow value or what is your suggestion?
I think the right approach depends on the purpose for your modeling analysis. Are you trying to calibrate the model in order to run a scenario analysis, or for a retrospective analysis? Is your scenario analysis intended to highlight the impacts of climate change, or land use change, or both? Is it a retrospective analysis of the impacts of historical land use change (or climate change), or is it prospective - will you be projecting scenarios into the future?
If your focus is on the impact of land use change on water yield historically: In this case you would want to hold the climate factors constant between the model runs, in order to isolate the impact of land use. I recommend calibrating your model to find the average Z value that gives the best fit across all three time periods, using the 3 different land use/climate data inputs. Then I would run the model using the same 10-year series of climate data for all three land cover inputs. This way you highlight the impact of land use change while holding climate inputs constant.
If your plan is to calibrate the historical model and then run a land use change scenario into the future, then I would recommend calibrating the model to the most recent 10- or 20-year period and find an average Z value that provides the best fit. Then use that Z value and the same climate inputs from your baseline model to compare water yield between the baseline and projected land uses.
If climate change impacts on water yield are the focus (or a combo of climate and land use): The Z parameter is a lumped parameter that reflects the seasonality of rainfall. Do you have reason to believe that rainfall seasonality has changed over the last 30 years? If yes, then it could be interesting to examine different Z values across time. In that case you might want to calibrate the model to derive three different Z parameters using the 3 land use inputs. Then you would probably want to do multiple model runs holding either the climate inputs constant or the land use inputs constant in order to be able to show the relative impact of land use versus climate on the water yield. You could consider running the final analysis using all three Z values and report the range of results, like an uncertainty envelope around your findings. Just be careful not to represent this as a full sensitivity analysis, as there are many other uncertainties that are affecting your results.
All this being said, unless you have a clear indication of changes in rainfall seasonality over your 30 year period (for example, from a statistical analysis of rainfall data), then I would hesitate to assume that differences in model fit between the 3 periods are due to this parameter alone and would tend to recommend you find a single average value for Z that provides the best fit. There may also be changes in anthropogenic water withdrawals or water transfers that are not reflected in your model, and you might want to look more closely at those before landing on Z parameter as the only calibration lever.
@Habeebboy4real one additional note is that we do recommend calibrating each 10-year climate scenario to its own corresponding 10-year observed streamflow average.