Automated Enforcement of High Resolution Terrain Models April 21, 2015 Brian K. Gelder, PhD Associate Scientist Iowa State University
Problem Statement High resolution digital elevation models (DEMs) should improve peak flow return estimates by better measuring basin characteristics However, high resolution elevation models capture flow impeding features such as roadways and removing them is a time consuming process of manual editing Can this process be automated?
Project Goals Goals for IIHR and Iowa DOT Stream centerline enforcement needed for update of Iowa FEMA floodplain maps Enable improvement of watershed boundaries base on LiDAR Enable classification of depression drainage status for improved StreamStats regression Goals for Iowa Daily Erosion Project 2.0 Continuous flowpaths within agricultural fields
Trial Area Walnut Creek (South Skunk River) HUC 12 in north central Iowa 12274 acres 19 square miles 60 meters of relief Prairie pothole depressions throughout Roads, abandoned railroads, bridges, culverts, lagoons, rivers, trees, small town Some low quality, pocky LiDAR Within USGS Spec!
Trial Area DEM & Hillshade
Trial Area Watersheds
Process Overview 1. DEM generation and initial filtering 2. Hole punching for Major fill regions 3. Fill region Characterization 4. Thin fill region removal 5. Fill region removal by cutting DEM 6. Analyze enforcements for retention
Major Accomplishments Classification of likely drained depressions Slopes >5% at local watershed boundary Embankment that created depression is roughly symmetrical Sounds like a road/railroad/runway right? Removal of embankment reduces depth of impeded flow
Major Accomplishments Automating enforcement process that accommodates data issues Irregular LiDAR return spacing Irregular TIN spacing Regular DEM spacing Resulting in Many depressions don't easily flow based on search criteria Elevation estimates are not always in channels
Major Accomplishments Developed an automated enforcement process that accommodates data issues Maximum effective depth 3 rd minimum elevation in a 3x3 kernel Max breaching distance is 3 x minimum distance from points at 80% of maximum effective depth to watershed boundary Multiple cutting attempts Upstream -> downstream (twice) Downstream -> upstream
Fill Depth at 3 m Resolution
Raw DEM
Fill Depth at 3 m Resolution
Fill Regions at 3 m Resolution (1 st Fill Step)
Fill Regions at 3 m Resolution (2 nd Fill Step)
Fill Regions at 3 m Resolution (From All 5 Fill Steps)
Fill Regions at 3 m Resolution
Upstream Points & Search Buffer
Eligible Downstream Points
Best Connection(s)
Preliminary Enforced DEM
Classification Accuracy Depression Enforcement Accuracy 88 4 290 850 Depressions Correctly Classified as Undrained - 850 Depressions Enforced That Should Not Be - 290 Drained Depressions Enforced Correctly - 88 Drained Depressions That Were Not Enforced - 4
Classification Errors
Remaining Fill Depth
New HUC12 Boundary
Continuing Work Iowa Department of Natural Resources will now use enforced DEMs to Improve watershed boundary accuracy Improve National Hydrographic Dataset 'blue lines' Iowa Institute for Hydraulic Research/Iowa Flood Center is continuing with statewide floodplain risk analysis
Further Research Test if enforced LiDAR DEMs can improve USGS Streamstats regressions Most likely in Des Moines Lobe where regressions are least accurate There are 17 small watershed streamgages suitable for testing
Further Research Test if calculating statistics based on 'effective' watershed area can improve regressions versus entire watershed area Effective watershed area is the subset of the total area that contributes flow during a design storm of a given Annual Exceedance Probability (AEP) Only valid for Des Moines Lobe
Further Research Test if quantitative channelization metrics improve regressions over qualitative NHD 'blue lines' 'Blue line' locations vary by analyst Quantitative channelization metrics include Terrain Position Index Aspect change Profile curvature Other metrics
Questions?
Current Process 2. Fill Region Delineation (Hole Punching) Iterative process to define unique areas where depression is deep or large Set deepest point in each area to null in DEM Null values are holes in DEMs for water to flow out After hole punching determine if deepest depression is still deeper than minimum Code each area uniquely
Oxbow Cutoffs
Deepest Cells
Enforced Filled Flowpaths GT 160 acres
Trial Area DEM & Hillshade
Potential Channel Map
All Fill Regions
After Thin Fill Region Removal
Current Process 1. DEM Generation and Initial Filtering DEM Generation Create DEM using Mean Mild 18 This setting uses tiling to remove some of the pits when creating the terrain Initial Filtering Pit Filling Removes outliers in elevation One cell sinks» Reassign to fill elevation Most needed in pocky areas 17549 watersheds in HUC 12
Current Process 2. Fill Region Delineation (Hole Punching) Filled DEM Original DEM = Difference Same as difference grid depressions of Poppenga et al. 2010 Retain all fill regions greater than depth or area criteria Minimum Basin Depth Currently using 18 cm, RMSE spec is 18 cm Minimum Basin Area Currently 100 m 2 3895 watersheds in HUC 12
Current Process 3. Fill Region Characterization Thickness of fill region Maximum depth and depth range around sink Deepest 3 points in a 5x5 kernel around sink Are deepest points in an area of aspect change, low elevation, and high profile curvature? i.e. In a channel? Distance from deepest points to edge of fill region Convert raster representing deepest cells to deepest points for later processing
Current Process 4. Remove Channelized Fill Regions Remove fill regions from DEM that are: Fully in channelized areas AND Less than 1 meter in depth Removed fill regions are then tested to see if the overflow path is channelized If not, do not remove this fill region Protects small culverts in road ditches from being removed in this step Simplifies analysis for next steps
Current Process 5. Depression Removal by Cutting DEM Define fill regions to attempt to cut Deep points near edge of fill region Set constraints for downstream connections Within allowable search distance What is the distance to edge of fill region?» Multiply by 5x» Minimum of 30 meters Does the fill region abut a wide transportation feature (interstate highway, railyard, runway)?» Add distance to cross feature Must be at lower elevation than fill region
Current Process 5. Depression Removal by Cutting DEM Buffer points by search distance and find cells at lower elevation Thin by in channel status Evaluate all connections for distance, slope within greater of 15m or 50% of minimum match distance within greater of 1% or ½ of maximum match slope Enforce all eligible connections Thin to one connection later
Current Process 6. Analyze Enforcements for Retention Reduce connections to one per fill region First priority easiest connection by cut/fill Evaluate flowpath out of fill region before and after enforcement If enforcement cuts off an oxbow reject enforcement If enforcement does not reduce fill region storage reject enforcement Other issues in multi-pixel wide channels