Filling Missing Soil Data in Watershed Workflow Mesh

[Helen Weierbach]

Hi All, 

I am using watershed workflow to generate a mesh for running ats simulations with subsurface heterogeneities. However, I noticed that there is a significant amount of mukey/ats ids that have missing data when pulled from SSURGO across the domain. For example, some mukey values are missing all data that is output as soil_subsurface_properties,  while others are missing only porosity, or contain only permeability values. (screenshot of rows from soil_survey_props with missing data below).

Does anyone have preferred methods for filling gaps in soil property data? I've tried tracing back information in NRCS to see if there are similar mukeys that could be used to fill values but haven't had any success in locating necessary information directly where the soil properties data is being pulled from (https://sdmdataaccess.nrcs.usda.gov/Tabular/post.rest). 

Best,

Helen 

--

Helen Weierbach (she/her)

Environmental Data Science Research Associate

Energy Geosciences Division 

Lawrence Berkeley National Lab 

(571) 730-7903

[Saubhagya Singh Rathore]

Hi Helen,

I encountered a similar issue where half of my watershed domain (HUC12) had missing SSURGO data (represented by 3-4 mukeys).  I was considering the following two options and decided to go with the first:

1) Predicting mukeys (with no missing data) for the cells where data is missing: I used a decision tree classifier (from sklearn.tree import DecisionTreeClassifier) with mukey as the dependent variable. The classifier was trained and tested using cells having mukeys with no missing data. I found these predictors (in the order of feature importance) helpful: elevation, distance from the river, geology, and land cover. All these datasets are downloaded and mapped onto the mesh in the watershed workflow, so it is very convenient to leverage them. There are other supervised ML techniques you can use, I found this appropriate because it is non-parametric and can take both categorical and continuous variables as predictors. 

2) Predicting soil properties directly: Another option could be spatially interpolating soil properties (porosity, permeability, thickness, etc.) using simple geostatistical techniques like co-Kriging (with elevation etc.) or advanced ML-based data imputation techniques. However, the challenging part is to group/cluster these cells (where soil properties are predicted) based on the similarity/dissimilarity of soil properties so that you can assign them "fake" mukeys/ats ids. This mutually exclusive and exhaustive grouping is essential to define regions (and element blocks) in the mesh based on which soil properties are assigned in the ATS input file. 

There could be better ways to deal with this problem, I would love to hear from others.

Thanks

-- 

Saubhagya Singh Rathore

R&D Associate Staff Scientist 

Watershed Systems Modeling Group 

Environmental Sciences Division

 

Oak Ridge National Laboratory 

1 Bethel Valley Road, Oak Ridge, TN 37830

Mobile: 470-263-1189

rath...@ornl.gov 

 

[Coon, Ethan]

Great conversation so far, and I’m looking forward to hearing more ideas here.  I think this is a common problem.  Saubhagya, it would be great to get a generic function into WW with your option #1!

 

Other approaches I’ve used or seen used:

 

1. Use GLHYMPs where SSURGO is missing.  This isn’t my favorite, because often GLHYMPs is more appropriate for deeper, consolidated soils
2. One could use SoilGrids -- though we don’t have great ways of integrating SoilGrids (which is a raster-based product) vertically and horizontally, so you’d have to download all 7 layers globally and then figure out what to do with it.  This would be a nice addition to the options in WW, but we don’t have it set up easily now.
3. Stare at it for a bit – a lot of the time if you look at the SSURGO shapes plotted, you’ll see there is a direct continuation (e.g. one MUKEY is on a ridge, and if you follow the ridge, you see it connects to another MUKEY that continues the same ridge at the same elevation – probably you can use these as synonyms for each other.  Similar on riparian corridors/hillslopes/etc.

 

Ethan

 

-- 

-------------------------------------------------------------------------

Ethan Coon

Senior Research Scientist

Oak Ridge National Laboratory

 

865-241-1296

https://www.ornl.gov/staff-profile/ethan-t-coon

-------------------------------------------------------------------------

 

--
You received this message because you are subscribed to the Google Groups "Amanzi-ATS Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ats-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ats-users/CAHh7dX5Bqe6yR-mDGqb1ZYUSyWhN76s0XYMenq6w8pmEE-OXpA%40mail.gmail.com.

[Chuyang Liu]

Hi,

I also encountered a similar issue. One potential approach is to use the Gridded National Soil Survey Geographic (gNATSGO) Database (https://www.nrcs.usda.gov/resources/data-and-reports/gridded-national-soil-survey-geographic-database-gnatsgo) instead of using SSURGO alone.

Thanks,

Chuyang

-----------------------------
Chuyang Liu

    

Climate & Ecosystem Sciences Division  
Lawrence Berkeley National Laboratory  

[Helen Weierbach]

Hi All, 

Thanks so much for your thoughts on gap-filling soil data and building on/combining the default datasets that are included in watershed workflow. I've started looking at some of the options Saubhagya, Ethan, and Chuyang suggested. 

If I find an elegant solution that fits well with the workflow I'd be happy to contribute some tools to the repo if that's helpful for the community. 

Best,

Helen 

To view this discussion on the web visit https://groups.google.com/d/msgid/ats-users/13896718-c5e1-4952-8d17-463e1beabea6n%40googlegroups.com.