SDR: calibration difficulties between values of full watershed and subwatershed

Dear NatCap community,

I’m experiencing some difficulties in calibrating the SDR model to balance estimates of sediment export between measured values at the mouth of the full watershed (Whole WS dataset) and observations for 2 sub-basins (Sub WS dataset).

I am able to get similar values between the SDR model and the observations at the scale of the whole watershed but in this case the sediment exportation value of the sub-watersheds estimated by the SDR model were skyrocketing compared to the reality (Sub WS dataset).

Here is the sediment exportation results from my best run (SDR_75, SDR_76 in attachment):
SDR model whole WS: 187306tn/an VS reality: 187862tn/an
SDR model Sub catchment 1: 252tn/an VS Sub WS dataset 1: 111tn/an
SDR model Sub catchment 2: 5020tn/an VS Sub WS dataset 2: 300tn/an

I tried adjusting many of the calibration parameters according to Hamel and all. (2015) (P and C values in my biophysical table, Threshold flow acc, K parameter, IC0 parameter and even max SDR). However, the result of my simulation never works at both scales (whole WS and sub-watersheds).

Do you have any recommendation on which parameters or datasets I should adjust to help me get better results? Or would someone have an explanation about the gap between the watershed and his sub catchments?

Thank you in advance


Hamel, P., Chaplin-Kramer, R., Sim, S., & Mueller, C. (2015). A new approach to modeling the sediment retention service (InVEST 3.0): Case study of the Cape Fear catchment, North Carolina, USA. Science of the Total Environment, 524, 166-177.

InVEST-Sediment-Delivery-Ratio-Model-(SDR)-log-2021-06-16–15_59_16.txt (245.4 KB)
SDR_calibration_CapNat.csv (6.6 KB)

1 Like

Dear Olivier,
It is possible that there is some landscape heterogeneity in the input parameters between the subwatersheds in your AOI that is not reflected by a global parameterization. If you are not able to achieve realistic calibration at both scales, then I would suggest splitting your AOI into different subwatersheds and calibrating each separately. There are at least two ways to do this, depending on whether you are adjusting parameters in the biophysical table versus global parameters:

  1. C, P, and other biophysical table parameters: You can split your land cover/land use classes based on subwatersheds, and calibrate the parameters separately. For example, you might have “Forest - Subwatershed 1”, “Forest - Subwatershed 2”, and “Crops - Subwatershed 1”, “Crops - Subwatershed 2”, etc. Then do not vary all C and P by the same amount, but instead systematically vary C and P by subwatersheds until you achieve a good model fit.

  2. Global parameters: This is a bit more tricky, but you could run the model for each subwatershed separately, splitting the landscape based on known heterogeneity in landforms etc. You then run the model for each piece with different global parameters. Note that if you divide the AOI into pieces then it should be on the basis of subwatersheds, since you need to maintain hydrologic connectivity of each piece. Once you have the results for each piece of the AOI, you can sum the results, or merge the raster outputs. I had to do this once for the Seasonal Water Yield model and was able to achieve decent results.

I hope these ideas help!



Hi Olivier,
thanks for your note. I agree with @adrianvogl comments above, both options are reasonable.

It is not uncommon in hydrologic modeling that there is a trade-off between good model results on large scales, and how well the model represents variability in water, or in your case, sediment, on some smaller scales.
Thus, if you are particularly interested in representing the sediment yield from the two smaller subwatersheds, then you might want to develop and calibrate separate models for them.
If you go down the route of building a model for the whole watershed, have a critical look at the soil data, too. Especially if there is any evidence for much less erodible soils in the subwatersheds where sediment exports are overestimated.