I am exploring the impacts of a wetland and riparian buffer restoration scenario against a business-as-usual scenario using the SDR and NDR models. The nodata area created when running the models results in the omission of ~40% of the restoration area I fear compromising any findings. I understand these models produce nodata areas where streams/water bodies are defined but is there a way to mitigate this? Have folks tried to fill nodata areas using scaling factors calculated based on nearby and similar areas with known data? Many thanks.
Please remember that these models do NOT include in-stream processes.
For example, the SDR model only deals with overland erosion and sediment export. As stated in the User Guide,
Other sources of sediment include gully erosion, streambank erosion, and mass wasting from landslides or rockfalls, and glacial erosion.
But none of those processes are captured by SDR.
If your modeled spatial extent is quite flat, these InVEST freshwater models may consider many pixels to be part of a braided stream or river network, and therefore results would not be provided in those areas. Of particular importance is step #5 described in the User Guide here, “Verify the stream network”:
The stream network generated by the model from the DEM should closely match the streams on a known correct stream map.
Please closely review these steps explained under “Working with the DEM” and be sure to pre-process your input DEM with care before running these InVEST models.
If you are interested in representing in-stream processes, InVEST likely does not provide the appropriate models for your use case, yet.
This is caused by the way the MFD flow direction algorithm works. It’s great for defining flood plains, but it is problematic when the resulting modeled stream network is not realistic and cuts into riparian restoration defined in the land use map, it’s happened to me more than once. For what it’s worth, we’re working on adding the option of using the D8 flow direction algorithm to the freshwater models, which is much more likely to produce streams that are just 1-2 pixels wide (except in large, totally flat areas, where most flow direction algorithms have trouble), but it hasn’t been released yet.
Jesse pointed to the Working with the DEM section of the User Guide, which is useful for sanity-checking your DEM. But if it’s just the case that your riparian zone is actually naturally flat, and the DEM is pretty accurate, one thing you could try (that we don’t really talk about in the User Guide) is “burning” a thin (1-2 pixel) stream network into the DEM, which might create an equally thin output stream. Here’s a previous forum post that addresses the topics of wide streams and riparian buffers in some detail.
Many thanks for your thoughtful reply. I fully recognize the point that the models don’t include in-stream processes. My stream network checks quite well against the network observed via satellite imagery so I believe some areas are indeed quite flat. That said, I will have another look at “Working with the DEM”.