Progress in Modeling Spatial Population Scenarios. Bryan Jones CGD-IAM, NCAR NCAR IAM Group Annual Meeting, July 19, 2012

Size: px

Start display at page:

Download "Progress in Modeling Spatial Population Scenarios. Bryan Jones CGD-IAM, NCAR NCAR IAM Group Annual Meeting, July 19, 2012"

Eugene Wheeler
5 years ago
Views:

1 Progress in Modeling Spatial Population Scenarios Bryan Jones CGD-IAM, NCAR NCAR IAM Group Annual Meeting, July 19, 2012

2 Agenda 1. Introduction a. Importance b. Research goals 2. Methodology a. Current methods b. Gravity models c. Our approach 3. U.S. Scenarios 4. Producing Global Scenarios

3 Why are spatial population projections important? Spatial Emissions Risk and Vulnerability Assessment Source: Balk et al., 2009

4 Community Demographic Model (CDM) Urbanization Projection (National) Household Projection ( 31 regions, pop in hh by age, size, and rural/urban) Emission Mitigation Analysis Population Projection (31 regions, age, sex, rural/urban) Spatial Downscaling (grid cell, rural/urban) Impacts, Adaptation, Vulnerability Analysis

5 Project Overview Research Goal: To develop an improved methodology for constructing large-scale plausible future spatial population scenarios which may be calibrated to reflect alternative regional patterns of development. Process: Large-scale Data constraints Scale of application Scenarios Storylines (SRES, SSP) 1) Assessment of leading existing methodology. 2) Identify problematic features and test modifications. 3) Apply modified methodology to produce 100-year spatial scenarios for the United States (SRES A2 & B2 scenarios)

Existing Methods Proportional scaling Gaffin et al., 2004 Bengtsson et al., 2006 van Vuuren et al., 2007 Trend extrapolation Balk et al., 2005 Hachadoorian et al.

6 Existing Methods Proportional scaling Gaffin et al., 2004 Bengtsson et al., 2006 van Vuuren et al., 2007 Trend extrapolation Balk et al., 2005 Hachadoorian et al., 2011 Projected economic data Asadoorian, 2005 Gravity-based Grübler et al., 2007 Source: CIESIN, 2004 Smart Interpolation Use of ancillary data to enhance allocation of population (e.g., EPA, 2010) Data and computational issues with global application

7 Gravity-Based Models Gravity I = ij P i D P 2 ij j Flows 2 directions Spatial interaction Migration and transportation Potential v i = n j= 1 P D j 2 ij Influence 1 direction Spatial allocation Accessibility or attractiveness

The IIASA Potential-Allocation Downscaling

8 The IIASA Potential-Allocation Downscaling Methodology v i = m j= 1 P D j 2 ij Projected US Population: IIASA A2 Scenario Gridded Population Distribution Replicates commonly observed patterns of spatial development including: Urban sprawl Development of urban corridors

9 Assessment of the IIASA Methodology Not validated against or calibrated to historical data Limited range of development trajectories Border effects Irregular patterns along the urban/rural border Population loss is misallocated Allocation into areas unsuitable areas for human habitation IIASA A2 Scenario (Gruebler et al., 2007)

10 The Modified Potential Model IIASA potential model: v i = m j= 1 P D j 2 ij v i = NCAR modified potential model: a l i i m j= 1 P α j e β d ij Urban and rural populations coexist within grid cells Allocation occurs as proportional to the inverse of potential during projected periods of population loss

Calibrating the Parameterized Model P P obs i,2000 obs T,2000 = P P mod i,2000 mod T,2000 + ε i ( α, β ) Unconstrained

11 Calibrating the Parameterized Model P P obs i,2000 obs T,2000 = P P mod i,2000 mod T, ε i ( α, β ) Unconstrained minimization problem Generalized Reduced Gradient (GRG2) algorithm (Lasdon and Warren, 1986) Historical data test: Continental US

12 Population and Urbanization: IIASA A2 & B2 Scenarios

13 Geospatial Mask: IIASA A2 & B2 Scenarios Low Protection: A2 Scenario High Protection: B2 Scenario

14 Comparison to the IIASA A2 Scenario NCAR A2 Scenario, 2100

15 Patterns of Spatial Development: Atlanta MSA Observed, 1990 IIASA A2, 2100 NCAR A2, 2100

16 Geophysical Spatial Mask: Everglades Example 1990 Observed IIASA 2100 Predicted IIASA A Predicted NCAR A2

17 NCAR Projections: A2 and B2 Scenarios NCAR A2 Scenario, 2100 NCAR B2 Scenario, 2100

18 Producing Global Scenarios Scenarios consistent with new SSPs Regional/national patterns of spatial change Estimates from historical data Catalog of estimates Families of countries following specific trajectories Output/useful metrics Location-based vulnerability to climate impacts Urban structure

19 Future Work Methodological Higher resolution Calibration: other countries, catalog of parameters Reclassifying urban /rural populations Applications Global scenarios Integrated urbanization

20 EXTRA SLIDES

21 Historical Data Test Observed 2000 Predicted

22 Historical Data Test: Prediction Error Divisional Redistribution Scenario

23 IIASA Greenhouse Gas Initiative (Gruebler et al., 2007) IIASA A2 Scenario:

Distribution t(0) Stability Final Distribution

24 Limited Range of Spatial Patterns The methodology forces a single spatial pattern to evolve over time: dispersion A % Population 4% 3% 2% 1% Normal Distribution, 80 Cells Initial Distribution t(0) Stability Final Distribution 0% This pattern is the result of a fixed distance-decay function

25 Expanding the Range of Spatial Patterns Squared Inverse-Distance v i = m j= 1 P D j 2 ij Negative Exponential (Parameterized) v i = m j= 1 P α j e βd ij Normal Distribution, 80 Cells % Population 4% 3% 2% Initial Distribution t(0) Stability Final Distribution 1% 0%

26 Border Effects High Low Projected Population Distribution: NCAR Model, 2100

27 Modifications: Removing Boundary Effects A. Base-year = = m j d j i i ij e P a v 1 β α = = m j d j i ij e P v 1 β α Original Adjusted m v v v a i i i = min 1 Adjustment Factor

28 Rural/Urban Boundary Problems Normal Distribution 7% 6% Urban/Rural Border 5% t(0) - Urban % Population 4% 3% 2% Initial Distribution t(0) - Rural t(100) - Urban Final t(100) - Rural Distribution 1% 0% The urban/rural cliff effect results from: Classification and separate allocation to urban and rural cells We eliminate this problem by: 1. Allowing urban and rural population to exist in the same cell. 2. Allocating urban and rural change across all cells.

29 Population Loss is Missallocated

30 Allocation to Areas Unsuitable for Human Habitation v i = a l i i m j= 1 P α j e β d ij

Smoothing effect of larger windows Larger windows lead to a

Similar to smoothing that results from the repeated

31 Smoothing effect of larger windows Larger windows lead to a smoother potential surface, a result of the wider influence of large urban populations. The result is a more rapid diffusion of the population. Similar to smoothing that results from the repeated application of a moving average window. The larger the window the faster the movement towards uniformity.

32 Windows

33 Spatial Stability in the 2-D Hypothetical Model

34 Population Change; Adjusted vs. Unadjusted 2D Hypothetical Scenario

35 Potential vs. Inverse Potential Allocation of Population Loss

36 Historical Data Test: Parameter Estimates Interpretation of parameter estimates: Relative measure of the change in variability in the spatial distribution of the population.

37 Border Effects: El Paso Example Population Change (000s) Border Interior Expon. (Border) Expon. (Interior) y = 22593e x R² = y = 11516e x R² = Distance to Urban Center (km) IIASA Population Change (000s) Border Interior NCAR Expon. (Border) Expon. (Interior) y = 16186e x R² = y = 14497e x R² = Distance to Urban Center (km)

38 Boundary Effects Cell Cells j Avg Dij (km) K11 1, Y25 1, The calculation of V(i) for Y25 includes 851 more cells j than that of K11. On average, those cells j included in the calculation of V(i) for K11 are 25.8 km closer. The window effect increases V(i) in Y25 relative to K11, while the positional effect does the opposite. Window effects result from variation in the number of cells j included in the calculation of potential for each cell i, leading to lower potential in cells closer to the boundary. Positional effects result from variation in the average distance from cell i to those cells j included in the calculation of potential.

The Window Effect 2000 Number of cells j contributing to potential 1500 Contribuing cells j 1000 500 Cells 0 A1 B2 C3 D4 E5 F6 G7 H8 I9 J10

39 The Window Effect 2000 Number of cells j contributing to potential 1500 Contribuing cells j Cells 0 A1 B2 C3 D4 E5 F6 G7 H8 I9 J10 K11 L12 M13 N14 O15 P16 Q17 R18 S19 T20 U21 V22 W23 X24 Y25 Cell i The window effect increases potential in red cells relative to blue cells.

The Positional Effect Average distance to contributing cells j 205 200 Average Dij 195 190 185 180 175 A1 B2 C3 D4 E5 F6 G7 H8 I9 J10 K11 L12

40 The Positional Effect Average distance to contributing cells j Average Dij A1 B2 C3 D4 E5 F6 G7 H8 I9 J10 K11 L12 M13 N14 O15 P16 Q17 R18 S19 T20 U21 V22 W23 X24 Y25 Avg Dij Cell i The positional effect increases potential in red cells relative to blue cells.

Boundary Effects on Potential Potential 0.18 0.16 0.14 0.12 V(i) 0.1 0.08 0.06 0.04 0.

41 Boundary Effects on Potential Potential V(i) A1 B2 C3 D4 E5 F6 G7 H8 I9 J10 K11 L12 M13 N14 O15 P16 Q17 R18 S19 T20 U21 V22 W23 X24 Y25 Potential Cell i

42 NCAR Projections: New York (or other), A2 and B2 Scenarios NCAR A2 Scenario, 2100 NCAR B2 Scenario, 2100

Constructing Global-Scale Spatial Population Scenarios. Bryan Jones, CGD-IAM, NCAR Boulder, CO August 19, 2013

Constructing Global-Scale Spatial Population Scenarios Bryan Jones, CGD-IAM, NCAR Boulder, CO August 19, 2013 NCAR Community Demographic Model (CDM) Urbanization Projection (National) Household Projection