agreement at multiple resolutions for real and categorical maps chris ayres – george kariuki...

23
Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera [email protected], [email protected], [email protected] GEOG 360 Quantitative Modeling Final Project - Spring 2004 REALMULTIRES, SOFTMULTIRES

Post on 19-Dec-2015

216 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

Agreement at Multiple Resolutions for Real and

Categorical Maps

Chris Ayres – George Kariuki Kristopher Kuzera

[email protected], [email protected], [email protected]

GEOG 360 Quantitative ModelingFinal Project - Spring 2004

REALMULTIRES, SOFTMULTIRES

Page 2: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

2

Lessons

• Currently, there is no known program available to perform and interpret a MULTIPLE-RESOLUTION ANALYSIS for real and categorical data.

• Automation of these processes allows for efficient production and replication of methodologies, with a minimization of human error.

Page 3: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

3

Multiple Resolution Analysis

• Multiple resolution analysis compares corresponding pixels of two maps over varying resolutions ranging from fine to coarse.

• This helps locate spatial patterns within the dataset and allows for ease in interpretability of the data.

• The analysis can distinguish disagreement of quantity from disagreement of location.

Page 4: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

4

Forest 1999Urban 1971

Multiple Resolution Example

Maps of categorical disagreement from fine to coarse resolutions

Worcester County, MA

Page 5: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

5

MULTIRES Programs

1971

1999

Actual

Predicted

• REALMULTIRES– Real variables

• SOFTMULTIRES– Categorical variables

Page 6: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

6

Characteristic REAL-MULTIRES

SOFT-MULTIRES

Performs MULTIPLE RESOLUTION ANALYSIS of three different SOFT-CLASSIFIED OPERATORS to compare two categorical maps.

Yes

Compares two maps of a common real variable (NDVI, SST) at multiple spatial-resolutions using various components of two measures of accuracy: (1) Root Mean Square Error (RMSE) (2) Mean Absolute Error (MAE)

Yes

Uses IDRISI to carry out raster functions: CONTRACT, RECLASS, OVERLAY, EXTRACT, WINDOW, CONVERT, TRANSFORM

Yes Yes

Calculates COEFFICIENTS OF AGREEMENT from generated cross-tabulation matrices of agreement and disagreement.

Yes

Graphically displays results over all resolutions. Yes Yes

Page 7: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

7

NDVI deviation at 1x1 km

Null model would predict zero everywhere.

Drought Prediction in Southern Africa

Actual Map Predicted Map

Page 8: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

8

NDVI deviation at 4x4 km

Null model would predict zero everywhere.

Drought Prediction in Southern Africa

Actual Map Predicted Map

Page 9: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

9

NDVI deviation at 16x16 km

Null model would predict zero everywhere.

Drought Prediction in Southern Africa

Actual Map Predicted Map

Page 10: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

10

Program Implementation: RMSE

Perfect Posterior PriorINFORMATION OF QUANTITY

Pe

rfe

ctP

erf

ect

Po

ste

rior

Un

iform

Un

iform

Glo

ba

lIn

-Str

atu

mP

ixe

lIn

-Str

atu

mG

lob

al

INF

OR

MA

TIO

N O

F L

OC

AT

ION

0

E

1e

Nre

1n

E

1e

Nre

1n

Wren

XjrenYjrenWren2

E

1e

2

Nre

1n

Nre

1n

Wren

XjrenYjrenWren

E

1e

Nre

1n

E

1e

Nre

1n

2

Wren

Wren XjrenYjren

E

1e

Nre

1n

E

1e

Nre

1n

2

Wren

Wren XjrenjeY

E

1e

Nre

1n

E

1e

Nre

1n

2

Wren

Wren XjrenjY

E

1e

Nre

1n

E

1e

Nre

1n

2

Wren

Wren XjrenjY~

Page 11: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

11

Program Implementation: MAE

Perfect Posterior PriorINFORMATION OF QUANTITY

Pe

rfe

ctP

erf

ect

Po

ste

rior

Un

iform

Un

iform

Glo

ba

lIn

-Str

atu

mP

ixe

lIn

-Str

atu

mG

lob

al

INF

OR

MA

TIO

N O

F L

OC

AT

ION

0

E

1e

Nre

1n

E

1e

Nre

1n

Wren

XjrenjY~

Wren

E

1e

Nre

1n

E

1e

Nre

1n

Wren

XjrenYjrenWren

E

1eNre

1n

Nre

1n

Wren

XjrenYjrenWren

E

1e

Nre

1n

E

1e

Nre

1n

Wren

XjrenYjrenWren

E

1e

Nre

1n

E

1e

Nre

1n

Wren

XjrenjeYWren

E

1e

Nre

1n

E

1e

Nre

1n

Wren

XjrenjYWren

Page 12: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

12

Interface for REALMULTIRES• Inputs

– Working Folder– Actual map– Prediction map– Mask Map– No. of Rows– No. of Cols– Range of resolutions

Page 13: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

13

Budget Results:Null versus Prediction

0

0.1

0.2

0.3

0.4

0.5

1 2 4 8 16 32 64 128

Resolution

Mea

n A

bso

lute

Err

or

. Agreement dueto LocationDisagreementdue to LocationDisagreementdue to Quantity

0.4

0.5

1 2 4 8 16 32 64 128

Resolution

Mea

n A

bso

lute

Err

or

. Agreement dueto LocationDisagreementdue to LocationDisagreementdue to Quantity

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

1 2 4 8 16 32 64 128

Resolution

Ro

ot

Mea

n S

qu

are

Err

or

.

Agreement due toLocation

Disagreement due toLocation

Disagreement due toQuantity

0.465

0.47

0.475

0.48

0.485

0.49

0.495

0.5

1 2 4 8 16 32 64 128

Resolution

Ro

ot

Mea

n S

qu

are

Err

or

.

Agreement due toLocation

Disagreement due toLocation

Disagreement due toQuantity

Page 14: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

14

REALMULTIRESAutomation Benefits

• Huge time savings– 2 WEEKS manually becomes 1½ MINUTES automatically.

• Minimize Chance of Error, Maximize Efficiency– Repetitive tasks are prone to human error such as typos that

could have a big unwanted impact on the results.

• RMSE and MAE have great potential for accuracy assessment (drought prediction)The two methods are better than regression at giving useful information to evaluate drought prediction in Africa.

Page 15: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

15

SOFTMULTIRESFlow Chart

Whichoperator?

MULTIPLICATION COMPOSITE

MINIMUM

Calculate minimum valuesof similar classes atresolutions. (MINIMUM)

Create gains and lossesmaps at resolutions.(ADD / SUBTRACT)

Multiply gains and lossesover 1 - agreement

Multiply each map withevery other at resolutions.(MULTIPLY)

Calculate minimum valuesof comparing maps atresolutions. (MINIMUM)

Calculate sumof pixels. (EXTRACT)

Calculate sumof pixels. (EXTRACT)

Calculate sumof pixels. (EXTRACT)

Calculate sumof pixels. (EXTRACT)

STARTRead the following variables frominput sheet: Comparison Year,Reference Year, Map files, Operators,Number of Categories

Create soft-classified resolution maps for eachclass at each year. (CONTRACT)

Read RDC file to determinenumber of resolutions, rows,and columns.

END

Sums entered in contingencytable at resolutions.

Derive statisticsfrom matrices.

Graph statisticsover resolutions.

IDRISI Kilimanjaro

Another operator?

NOYES

Page 16: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

16

SOFTMULTIRESSoft-Classified Operators

• Soft-classified operators allow for multiple class membership per pixel. The pixels are considered “soft”.

• Different operators have different interpretations of class membership and location within a pixel.

• SOFTMULTIRES allows user to choose any or all operators for their analysis.

OPERATOR AGREEMENT DISAGREEMENT

Multiplication multiply multiply

Minimum minimum minimum

Composite minimum multiply / ratio

Page 17: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

17

• Coefficients of agreement determine agreement based on the proportions of categories correctly classified.

• These vary depending on which cells of the cross-tabulation matrix are used for calculation.

• SOFTMULTIRES derives the above coefficients from generated cross-tabulation matrices.

OVERALL COEFFICIENTS OF AGREEMENT

CATEGORICAL COEFFICIENTS OF AGREEMENT

Overall Proportion Correct User’s Accuracy

Nishii-Tanaka Producer’s Accuracy

Cramer’s VConditional Kappa

(by row and column)

SOFTMULTIRESCoefficients of Agreement

Page 18: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

18

Interface for SOFTMULTIRES• Inputs

– Comparison map year– Reference map year– Number of Categories (maximum 15)– Operator (Multiplication, Minimum, and/or Composite)– Path Directory– Boolean raster images for each category

Page 19: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

19

Cross-Tabulation Matrix

Page 20: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

20

Graphical Output

Page 21: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

21

SOFTMULTIRESAutomation Benefits

• SOFTMULTIRES allows for simple recreation of methodology for any two categorical maps.

• SOFTMULTIRES produces many coefficients for ease of interpretability.

• SOFTMULTIRES reduces processing time.– 40 HOURS manually becomes 10 MINUTES automatically.

• SOFTMULTIRES produces hundreds of maps quickly.4 reference categories x 4 comparison categories

x 10 resolutions x 3 operators =

480 images

Page 22: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

22

Lessons

• Currently, there is no known program available to perform and interpret a MULTIPLE-RESOLUTION ANALYSIS for real and categorical data.

• Automation of these processes allows for efficient production and replication of methodologies, with a minimization of human error.

Page 23: Agreement at Multiple Resolutions for Real and Categorical Maps Chris Ayres – George Kariuki Kristopher Kuzera cayres@clarku.edu, gkariuki@clarku.edu,

23

• REALMULTIRES method is based on:• Pontius. 2002. Statistical methods to partition effects of quantity and location during

comparison of categorical maps at multiple resolutions. Photogrammetric Engineering & Remote Sensing 68(10). pp. 1041-1049.

• SOFTMULTIRES method is based on:• Kuzera, K. and Pontius, R. G. Jr. 2004. Categorical Coefficients of Agreement for Assessing

Soft-Classified Maps at Multiple Resolutions. In proceedings TIES 2004.

• Special thanks to: • Clarklabs (www.clarklabs.org) who is incorporating the validation method and the multiple-

resolution analysis of categorical maps into the GIS software Idrisi.• Ron Eastman who supplied data of NDVI images.• Human-Environment Regional Observatory Network for providing Worcester categorical data.• Gil Pontius for advising.• Olufunmilayo E. Thontteh for collaboration. More information available at the presentation of

her Masters Thesis work: “Verification Of Vegetation Index Predictions Using Multiple Resolution Images”.

Plugs & Acknowledgements