Robert Gilmore Pontius Jr rpontiusclarku'edu

1
Uncertainty in the difference between maps of
future land change scenarios

• Robert Gilmore Pontius Jr (rpontius_at_clarku.edu)
• Neeti Neeti (nneeti_at_clarku.edu) and
• Colin Polsky (cpolsky_at_clarku.edu)
• www.clarku.edu/rpontius

2
Major Points

• We compare two approaches to account for the
  uncertainty in land change scenario maps.
• One approach places simple bounds on the possible
  spatial allocation, given the quantity of each
  transition.
• The second approach uses validation to estimate
  uncertainty.
• We prefer the first simpler approach for
  practical and theoretical reasons.

3
Plum Island Ecosystems (PIE)in northeastern
Massachusetts
4
Land Change in the Ipswich River Watershed
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
Logic of Analysis
10
Logic of Analysis
11
Storylines Quantity of Transitions
12
Raw Scenario Maps
Simulated Transition
13
Difference Between Raw Scenario Maps
14
Raw Scenario Maps
Simulated Transition
How much difference can the spatial allocation
make, given the storylines constraint on the
quantity of each transition?
15
Bounds on Difference Between Scenario Maps
16
Accuracy Estimated by Validation
When the model predicts land change, it is
usually wrong.
Simulated Transition
This was the accuracy for a 14-year validation
interval.
17
Raw versus Estimated Maps
Simulated Transition
Probability of Forest

18
Estimated Difference Between Scenario Maps
19
Conclusions and Future Directions

• The method of estimation of uncertainty by
  validation has severe practical and theoretical
  problems.
• It makes little sense to calibrate and validate a
  business as usual scenario when historic business
  has not been usual.
• We recommend the simpler method that separates
  the variation due to the scenario storylines from
  the variation due to the simulation model.

20
Plugs References

• Enroll in my Land Change Modeling Methods
  Workshop, Thursday March 26, 9am-5pm in Capri
  Meeting Room 108.
• Take the CD for the workshop.
• Visit www.clarku.edu/rpontius or write to Gil at
  rpontius_at_clarku.edu to obtain the following
• Pontius and Neeti. In press. Uncertainty in the
  difference between maps of future land change
  scenarios. Sustainability Science.
• Pontius, Versluis and Malizia. 2006. Visualizing
  certainty of extrapolations from models of land
  change. Landscape Ecology 21(7) 1151-1166.
• Pontius, Boersma, Castella, Clarke, de Nijs,
  Dietzel, Duan, Fotsing, Goldstein, Kok, Koomen,
  Lippitt, McConnell, Mohd Sood, Pijanowski,
  Pithadia, Sweeney, Trung, Veldkamp, and Verburg.
  2008. Comparing the input, output, and validation
  maps for several models of land change. Annals of
  Regional Science, 42(1) 11-47.

21
Flow
22
Plum Island Ecosystems- Anderson Landcover Within
Ipswich Parker Watersheds
23

• The GIS-based model predicts the quantity and
  location of deforestation.
• New residential development is responsible for
  most deforestation.
• The model can create a variety of policy relevant
  scenarios.
• We analyze each scenario for its implication for
  nutrient release.

24
Plum Island Ecosystems - LTER
25
Simulated and actual deforestation1985-1991
Simulated
Actual
26
Actual versus simulated deforestation1985-1991
Non-change classified as non-change Change
classified as non-change Non-change classified as
change Change classified as change
27
Actual versus simulated deforestation1985-1991
28
Forest loss 1991-2101 w/ 1991 laws
29
Forest loss 1991-2101 with no laws
30
Land Use of Catchment
31
Nitrate loading 1991
32
Nitrate loading 2101 with constraints
33
Annual Nitrate loadingconstraints versus no
constraints
34
Land Use 1985
35
Proximity to Existing Residential
36
Proximity to Roads
37
Proximity to Hydro Features
38
Likelihood of deforestation
High suitability
Low suitability
39
Zoning Districts
40
Constraints on Development
41
Forest loss 1991-2101 w/ 1991 laws
42
Forest loss 1991-2101 with no laws