Place Still Matters: Generalizing the NHD by Local Terrain and Climate Barbara Buttenfield, Geography, CU-Boulder Cynthia Brewer, Geography, Penn State E. Lynn Usery, USGS-CEGIS Research Assistants: Chris Anderson-Tarver and Jochen Wendel, CU-Boulder Chelsea Hanchett and Paulo Raposo, Penn State Mamata Akella, PSU ESRI Programming Support: Larry Stanislawski, USGS Rolla TNM data prep: Tom Hale, USGS Much advice: Charlie Frye, ESRI
Scope Scope Data generalization for National Hydrography Dataset (NHD) Multiscale topographic base mapping For USGS web delivery portal The National Map (TNM) Scale range 1:20,000 1:1 million (1:4,800) Why focus on hydrography? Highly sensitive to scale and landscape differences Commonly displayed on most base maps
Objectives 1. Work toward full automation Minimize anticipated times for manual editing Use COTS software or USGS available code 2. Maximize efficiency Make no more generalized data versions than necessary Level of Detail (LoD) data
An Aside: what is LoD Data? Intermediate-scale versions of NHD Pre-compute labor-, time-, or skill- intensive tasks LoD is subset of database layers Include only revised hydrography reduce data volume LoDs not required at every mapping scale Only when symbol redesigns fail Establish smallest number of LoDs required for full scale range -- today, 50K
Objectives (con t.) 3. Preserve important hydrographic / cartographic characteristics and differing impacts on the landscape Terrain differences flat, hilly, mountainous Climate impacts humid, arid Settlement patterns rural, urbanized How? Tailor generalization algorithms, parameters, sequences What is the smallest number of tailored solutions that will apply to any generalization situation, nationally?
Landscape Types (Subbasin Sample) Humid Flat: Upper Suwannee, FL-GA Hilly: Pomme de Terre, MO Mountainous: South Branch Potomac, W V Dry Flat: Lower Beaver, UT Hilly: Lower Prairie Dog Town Fork Red, TX Mountainous: Piceance-Yellow, CO Urban St. Louis, MO; Atlanta, GA Four in Iowa: North and South Raccoon, Middle Des Moines, and Lake Red Rock
Building an LoD 2 case studies A. Isolate a single, continuous centerline Intersect NHD flowlines with NHD areas Dissolve on GNIS name Simplify B. Identify NHD Flowlines by local density Isolate stream networks (higher/lower density) Prune differentially Concatenate and simplify
Evaluating the solutions Contextual Mapping sequences across range of scales Critique by domain experts (hydrologists, cartographers) Cross-Validation Compare against 100k Medium Resolution NHD Metric Summary statistics on retained geometry Channel length, network local density Catchment, upstream drainage, polygon areas
Contextual Evaluation 100K map: with original 24K hydro with 50K LoD hydro Pomme de Terre, MO CEGIS meeting, August 2009
Metric Evaluation Goal: Retain overall channel length AND Preserve proportions between dense/ sparse networks 24k NHD 50k LoD Retained Channel Lengths: 24K 50K Total 3,428.40 km 2,375.26 km 68.49% Pruning limits set Dense 1,507.84 km 1,044.66 km 69.28% By Radical Law Normal 1,920.56 km 1,330.60 km 67.72% CEGIS meeting, August 2009
Case Study A: Isolate a single continuous centerline Pomme de Terre Missouri humid - hilly
Pomme de Terre MO NHD 24k humid - hilly Emphasized Problems Multiple centerlines in NHD polygons Distinguish primary and secondary centerlines Gaps in centerline continuity Submerged stream conflicts with centerline
Pomme de Terre MO NHD 24k humid - flat 3. Simplify flowlines 1. Select and simplify waterbodies and areas Prototype in ModelBuilder; test code in Python 2. Primary and secondary centerlines
1. Select and simplify NHD areas / waterbodies
2. Primary and secondary centerlines Intersect flowlines with selected polys; select on GNIS name; simplify CEGIS meeting, August 2009
Centerline continuity problems NHD Artificial Paths insufficient to produce cartographic centerlines gap gap CEGIS meeting, August 2009
3. Simplify flowlines CEGIS meeting, August 2009
Draft result MO 50K LoD mapped at 100K With original 24K hydro With derived 50K hydro
Draft result MO 50K LoD mapped at 100K High resolution (24K) NHD Derived 50K LoD
Validation Medium Resolution Derived 50K LoD (100K NHD)
Missouri Metrics channel length sum d km channel length /24k length channel density km / km2 NHD 24k 3,014 1.00 1.69 LoD 50k 1,744 0.58 0.87 NHD 100k 1,922 0.64 0.96 catchment area 24k 1,920 100k 2,001 24k 50k 100k CEGIS meeting, August 2009
MO: Original 24K hydro compared to 50K LoD At 300K CEGIS meeting, August 2009
Case Study B: Identify NHD Flowlines by local density S. Branch Potomac River humid - mountainous
S. Branch Potomac W V NHD 24k humid - mountainous Objectives: Preserve local density Protect topology Formalize thresholds for pruning Retain pruned tributaries for mapping CEGIS meeting, August 2009
What is meant by local density CEGIS meeting, August 2009
Upper Potomac W V NHD 24k humid - mountainous 1. Prune differentially, merge, build diffs_file 3. Isolate non-centerline flowlines 4. Simplify 2. Intersect NHD flowlines with NHD areas / waterbodies to delineate centerlines
Prune differentially Selection of dense and non-dense networks (manual, at present) Pruning criterion: Summed remaining stream channel length Summed original catchment area
Formalize thresholds for pruning Pruning thresholds taken from Radical Law: # _ items = t arg et.# _ itemssource * RF source RF t arget Modified as follows: channel _length = t arg et channel _length source * RF source RF t arg et For WVA: target length = 6,048.063km * SQRT(24k/50k) = 4,190.098 km
Radical Law worksheet Optimum pruning levels nor60: 2,643.128 + den40: 1,519.159 4,162.287
1. Prune differentially, merge, build diffs_file CEGIS meeting, August 2009
Draft result WV 50K LoD mapped At 100K: With original 24K hydro With derived 50K LoD
Draft result WV 50K LoD mapped At 100K: With original 24K hydro With derived 50K LoD
Draft result WV 50K LoD mapped At 100K: With original 24K hydro With derived 50K LoD
West Virginia Metrics channel length sum d km channel length /24k length channel density km / km2 NHD 24k 6,050 1.00 1.58 LoD 50k 2,174 0.36 0.61 NHD 100k 1,682 0.28 0.48 catchment area 24k 3,838 100k 3,543 24k 50k 100k CEGIS meeting, August 2009
Summary Status of our Work Physiography and climate impact 3 aspects of generalization Tool sequencing, Algorithm choice, Parameter choice Status: 4 of 8 subbasin solutions prototyped and moved to Python scripts, enter refinement phase this fall Centerline continuity NHD Artificial Paths insufficient to produce cartographic centerlines still unsolved trace the node network? Selection of density classes for pruning experiment in Iowa subbasins to select density classes automatically, scaling up to national fix. Stanislawski, Buttenfield and Roth 2009; ICC paper for meetings in Santiago
Funding acknowledgements USGS Center of Excellence for Geospatial Information Science (CEGIS) Funded through Department of Interior CESU program, January 2007 to present Initial Generalization and ScaleMaster work funded by ESRI, June 2003 to May 2007
Metric Evaluation Other Examples of Metric Evaluation preserved channel length original channel length % length preserved preserved channel length (original) catchment area % density preserved symbol line weight (in ground units) * channel length @ original scale @ LoD scale real estate utilized