Upload
others
View
14
Download
0
Embed Size (px)
Citation preview
Esri International User Conference | San Diego, CA Technical Workshops |
Geostatistical Analyst - An Introduction
Steve Lynch and Eric Krause Redlands, CA
July 2012
Presentations of interest…
• EBK – Robust kriging as a GP tool - Tuesday 3:00pm – 3:30pm Demo theater
• Geostatistical Simulations - Tuesday 3:30pm – 4:00pm Demo theater
• Areal Interpolation – Performing polygon-to-polygon predictions - Thursday 9:00am – 9:30am Demo theater
• Surface Interpolation in ArcGIS - Wednesday 1:00pm – 2:00pm Demo theater
• Designing and Updating a Monitoring Network - Thursday 9:30am – 10:00am Demo theater
• Concepts and Applications of Kriging - Tuesday 10:15am – 11:30am 15B
• Creating Surfaces - Wednesday 8:30am – 9:45am 15A
Outline
• What is • geostatistics? • Geostatistical Analyst?
• New in 10 and 10.1 • Interpolation workflow • Demonstrations • Supplementary information • Questions / Answers
Geostatistical Analyst - Overview
• Interactive - Exploratory Spatial Data Analysis tools - Variography - Kriging - Other interpolation methods - Cross validation
• Geoprocessing toolbox - Interpolation - Sampling Network Design - Simulation - Utilities - Conversion
• 12 independent reputable geostatisticians • Given the same data • Asked to perform the same straightforward estimation • Results were widely different • Different
- data analysis conclusions - variogram models and choice of kriging type - searching neighborhoods.
Experiment conducted by the US EPA 20 years ago Isaaks & Srivastava, 1989. An Introduction to Applied Geostatistics.
What’s new in 10 – Optimize buttons
• Local Polynomial Interpolation • Kriging
- Nugget, partial sill and other(s), are optimized using cross validation with focus on the estimation of the range parameter.
What’s new in 10.1
• Areal Interpolation - predictions can be made from one set of polygons to
another set of polygons
• Empirical Bayesian Kriging (EBK) - builds local models on subsets of the data, which are
then combined together to create the final surface.
• Normal Score Transformation - Multiplicative Skewing approximation method
Interpolation workflow
• Exploratory Spatial Data Analysis (ESDA) • Interpolate
- estimation of values at unsampled locations based on known values
• Goodness of fit
Exploratory Spatial Data Analysis
• Where is the data located? • What are the values at the data points? • How does the location of a point relate to its value?
Kriging
• Concepts and Applications of Kriging - Tuesday 10:15am – 11:30am 15B
• Empirical Bayesian Kriging – Robust kriging as a GP tool - Tuesday 3:00 – 3:30pm Demo theater
Outline • Introduction to kriging • Best practices • Fitting a proper model • Variography, transformations, isotropy, stationarity • Comparing models using cross validation • Interpreting results
What is kriging?
• It is a geostatistical interpolation technique • that models the spatial correlation of point measurements • to estimate values at unmeasured locations.
• Associates uncertainty with the predictions
Distance
Cor
rela
tion
Empirical Bayesian Kriging (EBK)
• automates the most difficult aspects of building a valid kriging model
• estimates the semivariogram through repeated simulations • can handle non-stationary input data.
Unlike other kriging methods (use weighted least squares), the semivariogram parameters in EBK are estimated using restricted maximum likelihood (REML).
New in ArcGIS 10.1
Empirical Bayesian Kriging (cont)
• Advantages - Requires minimal interactive modeling - Allows accurate predictions of non-stationary data - More accurate than other kriging methods for small datasets - Geoprocessing tool
• Disadvantages - Processing is slower than other kriging methods. - Cokriging and anisotropy are unavailable.
Cressie, 1990
• Cross validation does not prove that the model is correct,
• merely that it is not grossly incorrect.
Areal Interpolation
• reaggregation of data from one set of polygons to another set of polygons
New in ArcGIS 10.1
Create Spatially Balanced Points
Sampling Network Design
• Monitor road pollution • Convert roads to raster • High value = busy road • Low value = quite road
Densify Sampling Network
Sampling Network Design
• Used kriging to create: - Prediction surface - Standard error of prediction
• Want to add 2 new sites