error propagation zandbergen · why modelbuilder? • easy to follow visual programming language...

23
Error Propagation Modeling for Terrain Error Propagation Modeling for Terrain Analysis using Dynamic Simulation Tools in ArcGIS Modelbuilder ArcGIS Modelbuilder Paul Zandbergen Paul Zandbergen Department of Geography University of New Mexico

Upload: others

Post on 13-Jun-2020

7 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Error Propagation Modeling for TerrainError Propagation Modeling for Terrain Analysis using Dynamic Simulation Tools in 

ArcGIS ModelbuilderArcGIS Modelbuilder

Paul ZandbergenPaul ZandbergenDepartment of GeographyUniversity of New Mexico

Page 2: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Today’s topic:

Error Propagation Modeling

Page 3: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

OutlineOutline

• Teaching error propagation modelingTeaching error propagation modeling• Modelbuilder for dynamic simulationC d d h d d li i• Case‐study: stream and watershed delineation

• Exercise details and results

Page 4: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

MotivationMotivation• How to teach error propagation modeling to students 

and early career GIS professionals? – Strong general GIS skills– Basic knowledge of data quality– Limited experience with (geo)statistics

• Concepts:• Concepts:– Uncertainty (inputs, outputs)– Spatial autocorrelation of error

Monte Carlo simulation– Monte Carlo simulation

• Applied to stream and watershed delineationpp

Page 5: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Class ExerciseClass Exercise

• Sample datasetSample dataset• Step‐by‐step instructions• Tbx file with modelsTbx file with models

Page 6: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Original DEM Error DEMError term

++

StreamsStream Probability

Page 7: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Modelbuilder StructureModelbuilder Structure

Stream DelineationDEM Error ModelIteration Results

Inputs: Outputs:•Digital Elevation Model•Study area boundary•Root Mean Square ErrorSt th h ld

•Stream probability•Watershed probability

•Stream threshold•No. of iterations

Page 8: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Why Modelbuilder?Why Modelbuilder?

• Easy to follow visual programming languageEasy to follow visual programming language• Typically taught at intro GIS level• Models integrated into ArcGIS processingModels integrated into ArcGIS processing• Advanced functionality: if/then, loops, iteration

Page 9: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

DEM Error ModelDEM Error Model

Random Gaussian Error Spatially Autocorrelated Error

Focal Statistics

Page 10: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Results from IterationResults from IterationIteration 1 CombinedIteration 2

++

Page 11: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Final ResultsFinal Results100 iterations

Page 12: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Effect of RMSEEffect of RMSERMSE 0.1 ft RMSE 0.5 ft RMSE 1 ft

Page 13: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Comparison of Stream NetworksComparison of Stream Networks

Page 14: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Comparison of Stream NetworksComparison of Stream Networks

Page 15: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Watershed DelineationWatershed Delineation

1 iteration 100 iterationsDEM

Page 16: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Effect of RMSEEffect of RMSERMSE 0.1 ft RMSE 0.5 ft RMSE 1 ft

Page 17: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Class ExerciseClass Exercise

• Basic model provided:Basic model provided:– Model already designed as a tool– Explore effect of RMSE, no. of iterations

Page 18: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Assumptions / LimitationsAssumptions / Limitations

• Error model is relatively simpleError model is relatively simple– Spatially autocorrelated, yes– Tested (geo)statistically, no

• Error assumed to be unrelated to terrain complexity• Error assumed to be normally distributedy• Number of iterations is user‐defined

Page 19: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Class ExerciseClass Exercise• Students modify/enhance model:

– Improve error model• Scale of autocorrelation• Related to land cover, slope and/or roughness• Add non‐normal behavior

– Adapt model to different terrain derivatives• Visibility• Wetness index• Runoff model

– Set convergence parameter• Students apply simulation to other type of analysis

– Point‐in‐polygon overlayProximity analysis– Proximity analysis

Page 20: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Alternative Error ModelsAlternative Error  ModelsGaussian Original Error

Slope Land coverGaussian

++

Page 21: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Lessons from Teaching Error PropagationLessons from Teaching Error Propagation

• Pros:– Model tools help to get results quickly– Student “get” the error propagation concept

Leads to questions– Leads to questions• about the methodology, assumptions about the nature of error, applicability to other domains

• Cons:• Cons:– Model runs can take a while to complete– Complications

• Somewhat complicated to replicate approach to other application domains, students get stuck on issues with Modelbuilder

– Student still struggle with spatial autocorrelation

Page 22: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

Longer Term ObjectiveLonger Term Objective

• Simulation approach has broader applicabilitypp y

• Toolbox with many tools• Make error propagationMake error propagation modeling easier…

Page 23: Error Propagation Zandbergen · Why Modelbuilder? • Easy to follow visual programming language • Typically taught at intro GIS level • Models integrated into ArcGIS processing

ContactContactPaul Zandbergen – University of New Mexico

[email protected] – www.paulzandbergen.com