Top Banner
1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr ([email protected]) Joe Spencer ([email protected]) Prepared for presentation at the Open Meeting of the Global Environmental Change Research Community, Montreal, Canada, 16-18 October, 2003.
28

1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr ([email protected]) Joe Spencer ([email protected]) Prepared for.

Dec 21, 2015

Download

Documents

Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

1

Uncertainty in Extrapolations of Predictive Land Change Models

R Gil Pontius Jr ([email protected])Joe Spencer ([email protected])

Prepared for presentation at the Open Meeting of the Global Environmental Change Research Community, Montreal, Canada,

16-18 October, 2003.

Page 2: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

2

Major Points

• Validation is a waste of time,– unless you use the validation statistic to express

the level of certainty of predictions of the unknown.

• The prediction’s accuracy approaches random as the prediction’s time interval grows.– We estimate how fast the accuracy approaches

random.

Page 3: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

3

Worcester Massachusettsand nine surrounding towns

Page 4: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

4

Strategy of Three Runs

0

5

10

15

20

25

30

35

40

45

1950 1957 1964 1971 1978 1985 1992 1999 2006 2013 2020

Year

Per

cen

t o

f B

uil

t Real

Run 1: 1971 to 1985

Run 2: 1985 to 1999

Run 3: 1999 to 2013

Page 5: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

5

Percent Built versus Geology 1971

0 5 10 15 20 25 30 35 40 45 50

Sand and Gravel

Till and Bedrock

Floodplain Alluvium

Ge

olo

gy

Typ

e

Lubrication Values

Page 6: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

6

Percent Built versus Geology 1985

0 5 10 15 20 25 30 35 40 45 50

Sand and Gravel

Till and Bedrock

Floodplain Alluvium

Ge

olo

gy

Typ

e

Lubrication Values

Page 7: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

7

Percent Built versus Geology 1999

0 5 10 15 20 25 30 35 40 45 50

Sand and Gravel

Till and Bedrock

Floodplain Alluvium

Ge

olo

gy

Typ

e

Lubrication Values

Page 8: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

8

Percent Built versus Slope 1971

0 5 10 15 20 25 30 35 40 45 50

[0,1)[2,3)[4,5)[6,7)[8,9)

[10,11)[12,13)[14,15)[16,17)[18,19)[20,21)[22,23)[24,25)[26,27)[28,29)[30,31)[32,33)[34,35)[36,37)[38,39)Water

Slo

pe

Lubrication Values

Page 9: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

9

Percent Built versus Slope 1985

0 5 10 15 20 25 30 35 40 45 50

[0,1)[2,3)[4,5)[6,7)[8,9)

[10,11)[12,13)[14,15)[16,17)[18,19)[20,21)[22,23)[24,25)[26,27)[28,29)[30,31)[32,33)[34,35)[36,37)[38,39)Water

Slo

pe

Lubrication Values

Page 10: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

10

Percent Built versus Slope 1999

0 5 10 15 20 25 30 35 40 45 50

[0,1)[2,3)[4,5)[6,7)[8,9)

[10,11)[12,13)[14,15)[16,17)[18,19)[20,21)[22,23)[24,25)[26,27)[28,29)[30,31)[32,33)[34,35)[36,37)[38,39)Water

Slo

pe

Lubrication Values

Page 11: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

11

Real Built 1971

Page 12: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

12

Simulated Built 1971-1985

Page 13: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

13

Real Built 1971-1985

Page 14: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

14

Observed Accuracy Run 1: 1985

0

10

20

30

40

50

60

70

80

90

100

Pe

rce

nt

of

La

nd

sc

ap

e error due to quantity

error due to location

correct due to location

correct due to quantity

correct due to chance

4%

6%

Page 15: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

15

Estimated Accuracy 1971-2013

0

10

20

30

40

50

60

70

80

90

100

19

71

19

78

19

85

19

92

19

99

20

06

20

13

Time

Pe

rce

nt

of

La

nd

sc

ap

e

Error due to quantity

Error due to location

Correct due to location

Correct due to quantity

Correct due to random

Page 16: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

16

Strategy of Three Runs

0

5

10

15

20

25

30

35

40

45

1950 1957 1964 1971 1978 1985 1992 1999 2006 2013 2020

Year

Per

cen

t o

f B

uil

t Real

Run 1: 1971 to 1985

Run 2: 1985 to 1999

Run 3: 1999 to 2013

Page 17: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

17

Expected Accuracy 1985-2013

0

10

20

30

40

50

60

70

80

90

100

19

85

19

92

19

99

20

06

20

13

Time

Pe

rce

nt

of

La

nd

sc

ap

e

Error due to quantity

Error due to location

Correct due to location

Correct due to quantity

Correct due to random

Page 18: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

18

Expected Accuracy Run 2: 1999

0

10

20

30

40

50

60

70

80

90

100

Pe

rce

nt

of

La

nd

sc

ap

e

error due to quantity

error due to location

correct due to location

correct due to quantity

correct due to chance

4%

6%

Page 19: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

19

Real Built 1985

Page 20: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

20

Simulated Built 1985-1999

Page 21: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

21

Real Built 1985-1999

Page 22: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

22

Accuracy Run 2: 1999

0

10

20

30

40

50

60

70

80

90

100

Pe

rce

nt

of

La

nd

sc

ap

e error due to quantity

error due to location

correct due to location

correct due to quantity

correct due to chance

0

10

20

30

40

50

60

70

80

90

100

Pe

rce

nt

of

La

nd

sc

ap

e error due to quantity

error due to location

correct due to location

correct due to quantity

correct due to chance

Observed

1% Quantity Error

9% Location Error

Expected

4% Quantity Error

6% Location Error

Page 23: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

23

Expected Accuracy Run 3: 1999-2013

0

10

20

30

40

50

60

70

80

90

100

19

99

20

06

20

13

Time

Pe

rce

nt

of

La

nd

sc

ap

e

Error due to quantity

Error due to location

Correct due to location

Correct due to quantity

Correct due to random

1%

9%

Page 24: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

24

Simulated Built 1999-2013

Page 25: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

25

Expected Accuracy 1999-2199

0

10

20

30

40

50

60

70

80

90

1001

99

9

20

99

21

99

Time

Pe

rce

nt

of

La

nd

sc

ap

e

Error due to quantity

Error due to location

Correct due to location

Correct due to quantity

Correct due to random

Page 26: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

26

Expected Accuracy 1985-2185

0

10

20

30

40

50

60

70

80

90

1001

98

5

20

85

21

85

Time

Pe

rce

nt

of

La

nd

sc

ap

e

Error due to quantity

Error due to location

Correct due to location

Correct due to quantity

Correct due to random

Page 27: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

27

Major Points

• Validation is a waste of time,– unless you use the validation statistic to express

the level of certainty of predictions of the unknown.

• The prediction’s accuracy approaches random as the prediction’s time interval grows.– We estimate how fast the accuracy approaches

random.

Page 28: 1 Uncertainty in Extrapolations of Predictive Land Change Models R Gil Pontius Jr (rpontius@clarku.edu) Joe Spencer (jspencer@clarku.edu) Prepared for.

28

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.PDF file is available at www.clarku.edu/~rpontius or [email protected]

National Science Foundation funded this via: Center for Integrated Study of the Human Dimensions of Global ChangeHuman Environment Regional Observatory (HERO)

Plugs & Acknowledgements