SDR model pixels with values in a No data zone

Hello to all,

I ran the SDR model for my country, and have, in the sed_retention.tif raster, areas with No Data that have pixels with values in the middle (see image bellow).

Don’t mind the scale in the legend, it serves only to see if the zero values actually correspond to the streams.The pink areas represent the No Data and the grey is the value 0 (streams).

As you can see, there are isolated pixels in the pink areas (No data). Is this standard or a error?

I know that the model is influenced by the DEM, so I filled the sinks before I using it.

I didn’t get any error while running the model (I’ve ran it before for other projects, so I pre-processed all data correctly this time :smiley:)

Meanwhille, I’ve validated my results comparing them with the EU JRC data for soil erosion, so I don’t think this is significant enough to influence the overall results, but I might be wrong.That’s why I would really appreciate any input you can give me!

Thank you!
Susana

I should mention, that this area, in the image, is an estuary and that I ran the model in InVEST 3.6, in case it helps :slight_smile:

Hi @SusanaMM -

Here are a few things to check. First, look at your other output raster layers - do they all have NoData in the same place, or are some different? Then check is to make sure that none of your input layers have NoData in those areas. It’s common for, say, soil data to be missing where there are water bodies (also glaciers, and some others), and if there’s NoData in any of the inputs there will also be NoData in the outputs.

I also usually look at the streams.tif output layer. 3.6 is a rather old version of InVEST, and I don’t recall whether it produced 0s or NoData where there are streams. Looking at your output, perhaps there are 0s in streams, so it isn’t just that all of the streams have a NoData value.

And I recommend trying the latest version of InVEST, for several reasons. There have been changes and a lot of bug fixes since 3.6, and it now has a “sediment deposition” output that you might be interested in, since you’re looking at sediment retention. The sed_retention.tif output by SDR 3.6 does not show the actual amount of soil retained by a pixel, it’s only intended to provide a relative ranking, showing what would happen if the current vegetation was turned into bare soil. InVEST 3.8 added “soil deposition”, which actually quantifies where and how much soil gets deposited as it moves downhill toward a stream.

~ Stacie

1 Like

Hi @swolny,

Sorry for the very late reply.

You are right, the problem is the oficial K-factor map from JRC presents this kind of pixels, therefore the outputs will be like this also.

The recent version of InVEST seems interesting with the sed_deposition, I will try it for sure.

Thank you so much for your help :slight_smile:

Susana

Dear @SusanaMM and @swolny,

I would like to “bump” this message up the chain and return with a follow-up question.

I am running the InVEST SDR Model on a global scale, and using K and R factors also from the JRC.

These raster maps also have “holes” in them and is causing my output rasters to also have holes. I doubled checked those output rasters and most likely the “holes” are caused by these K and R rasters and not any other input.

So in this regard, how was the issue solved? Would it be enough to geoprocesses it with the “fill” tool? Similar to the suggested approach of also filling in empty pixels for the DEM (as seen in the InVEST ArcGIS tutorials)?

I am using the latest InVEST model (3.13.0). An example of the “holes” I mentioned is attached. I give two examples, one is the input (K factor) and the other is the output (usle).

Looking forward to your comments or responses!

Best regards,

Maulana


Hi @maulanapajie -

Yes, soil data in particular often has holes, especially where there are water bodies. We talk about this in the SDR Data Sources section of the User Guide and provide suggestions for how to fill the holes if you want to do that. Just remember that when we fill those holes, especially if they’re large, that’s yet another form of uncertainty.

~ Stacie

Dear @swolny ,

Thanks for your reply! I guess this part of the data sources section and user guide must have slipped my mind. Thanks for the reminder!

After further investigation, I managed to do the following:

Scenario A

Just fill those that have NoData with zero

Scenario B

  1. Reclassify LULC raster map into those with and without water bodies

  2. Raster Calculator: “hilda_plus_2015_states_GLOB-v1-0_base-map_wgs84-nn_16bit_reclassified.tif” * “RUSLE_KFactor_v1.1_25km.tif”

  3. Raster Calculator: Con(IsNull(“RUSLE_KFactor_v1.1_25km_reclassified.tif”), FocalStatistics(“RUSLE_KFactor_v1.1_25km_reclassified.tif”, NbrRectangle(3,3,“CELL”), “MEAN”), “RUSLE_KFactor_v1.1_25km_reclassified.tif”)

  4. Raster Calculator: “Con(IsNull(”“RUSLE_KFactor_v1.1_25km_reclassified_recalculated.tif”“), 0,”“RUSLE_KFactor_v1.1_25km_reclassified_recalculated.tif”" )

With scenario B, I managed to get somewhat ok raster maps compared to scenario A, however the challenge remains with some areas of the globe where there are huge chunks of pixels with no data (for example, the Sahara and Tibet) where Step 3 of Scenario B would not work, considering that you have no data neighbors to begin with.

Based on the user profile, it seems the “only” way forward would then be to calculate these R and/or K raster maps for the SDR model “by hand” based on the different soil profiles. However, since the InVEST model is only one step in a very long chain of modelling that I need to do, I would like to avoid this. Considering the time needed to collect data, verify calculations, etc.

Do you see any other ways or means of my “filling” these empty pixels in the existing K- and R-Factor Rasters that I have? By the way, I took my data from the EU for both R and K. Having a km-scale resolution is also ok in my case.

I look forward to your reply.

Thank you again so much for helping us users out with InVEST.

Best regards,

Maulana