68  Group Check In 06APR2026

Date: April 6, 2026

68.1 Attendees

• Nora Pearson
• Patty Burns
• Joe Brennan
• Adam Devlin
• Ava McCune

68.2 Field Sheet Updates for 2026 (Patty)

• Met with Adam at end of March to discuss needed edits
• Made changes and saved in Box. Edits saved over 2025 vintage of field sheet
• Nora uploaded 2025 field sheet and changed name of current version to “2026”
• Added densic contact to list of subsurface features
• Added indicator for whether densic contact is anthrodensic (yes or no)
• Added bulk density sample number column on lower portion of second page
• Column allows tracking of bulk density samples alongside lab samples
• Team asked to review updated field sheet

68.3 Concept Key Development

• Randy made comments on Adam’s key; Patty replied to Randy’s comments
• Need collective discussion on which wetness concepts to identify and build into key
• Options: Split out more concepts now and combine later, or combine now
• Team needs decision on which wetness groups to recognize as concepts

68.4 2025 Pedon Data Entry Status (Patty)

68.4.1 Upload Status

• Approximately 165 pedons from 2025 now in NASIS
• 8 random pedons not included for various reasons:
  • Some outside AOI (Area of Interest)
  • Some with missing field sheets
  • Some incomplete due to maps tour

68.4.2 NASIS Data Work Progress

• Site tab completely populated
• Currently working on pedon tab in NASIS
• Most calculations complete
• Final step: Getting classifications entered (very slow process)
• Adam has classified all pedons and recorded in spreadsheet
• Patty using Adam’s classifications while QC-ing
• QC process slow due to opening multiple documents

68.5 Rescheduling April Meeting with Randy

• Originally scheduled for Tuesday, April 14th
• Randy needs to reschedule to Tuesday, April 21st
• Works for Patty, Nora, and Joe
• Does not work for Adam Devlin (will be coaching at NACTA soil judging contest that week)
• Adam okay with missing meeting and can write notes for someone else to share
• Nora confirmed that Nic is available and emailed Randy

68.6 Modeling Updates (Nora)

68.6.1 General Approach

• Working on predictive modeling for HTM thickness
• Using machine learning random forest algorithm
• Training models on field-collected data
• Making predictions across landscape using covariates

68.6.2 Recent Progress

• Ran multiple models with different covariate combinations
• Current approach: “kitchen sink” method using all available covariates
• Goal: Determine which covariates are most influential for predictions
• Models evaluated using multiple metrics
• Current best model showing promising results

68.6.3 Covariate Testing

• Before kitchen sink approach: Used 6-8 specific covariates thought to be most influential
  • Proportion of impervious surface
  • Distance from road
  • Building age
  • Building perimeter length
  • Difference from mean elevation
• These performed about as well as kitchen sink approach
• Has not yet discussed feature importance findings with Nic since running latest models

68.6.4 Feature Importance Findings

• Consistently important features across last 5 models:
  • Max daily traffic within 100 meters
  • Max daily traffic within 500 meters
  • Both DEMs (digital elevation models): lidar-derived and National Elevation Dataset
  • Relative topographic position
• Nora suspicious of some results, particularly traffic metrics. Believes relative topographic position makes sense (could indicate soil movement)
• Plans to discuss with Nic

68.6.5 Collinearity Concerns (Joe’s Feedback)

• Joe raised concern about having both DEM and LiDAR-derived DEM in single model
• Strong collinearity expected between these covariates
• Question whether both should be included even in kitchen sink approach
• Nora agrees this is concern
• Other potential collinearity issues identified:
  • Land use and land cover datasets (both one-hot encoded as individual binary rasters)
  • Leaf on and leaf off composites for Sentinel-2 imagery
  • Road distance and building-related metrics may be too related
• Uncertainty about next steps beyond returning to choosing most relevant covariates

68.6.6 Building-Related Covariates

• Currently has building perimeter length covariate
• Based on 2008-2012 LiDAR-based DEM building footprints
• Available at 10-meter scale with:
  • Building count per cell
  • Building area per cell
  • Total building perimeter length per cell
• Joe suggested distance weighting from building perimeter
• Could weight larger buildings higher than smaller buildings (or vice versa)
• Mentioned using cost distancing in other projects
• Same approach could apply to roads with unique weighting
  • Example: Four-lane highway versus residential street
• Nora has created distance from road layer with weighting
• Based on functional class road data (likely from Met Council)
• Includes approximately 8 different road classes (highways, local roads, etc.)
• In kitchen sink models, hasn’t been used widely
• Nora considers it very relevant and will apply similar approach to buildings

68.6.7 Next Steps

• Will discuss feature importance with Nic
• Will explore new distance-weighted building covariates
• Will work on refining kitchen sink approach to address collinearity
• Nora will update Joe on findings with new covariates

68.7 Action Items

• Patty: Continue work on pedon tab classifications in NASIS
• Team: Review updated 2026 field sheet and provide feedback
• Team: Schedule discussion about wetness concepts for key
• Nora: Check with Nic about April 21st meeting availability
• Nora: Reply to Randy confirming rescheduled meeting works for everyone except Adam
• Nora: Discuss feature importance findings with Nic
• Nora: Create distance-weighted building covariates
• Nora: Refine covariate selection to address collinearity issues
• Nora: Update Joe on findings with new covariates
• Adam: Prepare notes for April 21st meeting for someone else to share