1. Gilberto Câmara, Pedro Andrade
Land change modelling
Gilberto Câmara, Pedro Andrade
Licence: Creative Commons ̶̶̶̶ By Attribution ̶̶̶̶ Non Commercial ̶̶̶̶ Share Alike

2. Slides from LANDSAT Modelling Human-Environment Interactions
images: USGS
Slides from LANDSAT
Modelling Human-Environment Interactions
How do we decide on the use of natural resources?
What are the conditions favoring success in resource mgnt?
Can we anticipate changes resulting from human decisions?
Aral Sea
1973
1987
2000

3. TerraME: Computational environment for developing human-environment models
Cell Spaces
Tiago Garcia de Senna Carneiro, “Nested-CA: A Foundation for Multiscale Modelling of Land Use and Land Cover Change”. PhD Thesis in Computer Science, INPE, 2006.
T. Carneiro, P. Andrade, G. Câmara, A. Monteiro, R. Pereira, “TerraME: an extensible toolbox for modeling nature-society interactions” (under review).

4. Modelling human-environment interactions
What models are needed to describe human actions?

5. Modelling and Public Policy
External
Influences
System
Ecology
Economy
Politics
Desired
System State
Scenarios
Decision
Maker
Policy
Options

6. Models need to be Calibrated and “Validated”
tp - 20
tp - 10 tp
Calibration
Validation
tp + 10
Scenario
Source: Cláudia Almeida

7. Clocks, clouds or ants? Clouds: statistical distributions
Clocks: deterministic equations
Ants: emerging behaviour

8. Statistics: Humans as clouds
y=a0 + a1x1 + a2x aixi +E
Establishes statistical relationship with variables that are related to the phenomena under study
Basic hypothesis: stationary processes
Fonte: Verburg et al, Env. Man., Vol. 30, No. 3, pp. 391–405

9. Statistical-based land use models

10. Statistics: Assessment of land use drivers
Land use models are good at allocating change in space. Their problem is: how much change will happen?
A. Aguiar, G. Câmara, I. Escada, “Spatial statistical analysis of land-use determinants in the Brazilian Amazon”. Ecological Modelling, 209(1-2):169–188, G. Espíndola, A. Aguiar, E. Pebesma, G. Câmara, L. Fonseca, “Agricultural land use dynamics in the Brazilian Amazon based on remote sensing and census data”, Applied Geography, 32(2): , 2012.

11. Driving factors of change (deforestation)
source: Aguiar (2006)

12. Amazônia in 2007 x All Variables
Transportation (11)
Distance Markets(7)
Demography (3)
Tecnology (2)
Environmental (20)
Public Policy(8)
Market (8)
Agrarian Structure(6)

13. Statistics: Humans as clouds
source: Aguiar (2006)
Statistical analysis of deforestation

14. Amazônia in 2007 x All Variables
transformações e análises de correlação
de 65 para 31 variáveis
seleção do melhor modelo
de 31 para 10 variáveis
regressão linear (AIC = )
regressão espacial (AIC = )
R-squared
0.5918
Beta
p-level
log(N_TRATOR_ )
<2e-16
sqrt(PSI_ASSENTAMENTOS_CLASSICOS)
FERTIL_ALTA
log(SOJA_ )
sqrt(PSI_GPM_SEDE_AMZ)
sqrt(PSI_GPM_CAPITAIS)
sqrt(PSI_PVM_IUMID)
log(DIST_MIN_MAD + 1)
TI_2006
UC_2006
8580 Cells

15. Amazônia x Variables of 1996/2006
26 Variáveis
R-squared
0.4501
Beta
p-level
FERTIL_ALTA
<2e-16
log(DIST_MIN_MAD_96 + 1)
log(DIST_MIN_ROD_PAV_96 + 1)
DIST_MIN_PORTOS
A_UC_1996
log(POP_RUR_ )
A_ASSENT_96_NUNFAM
log(PREC_INV + 1)
A_NU_AGR_MEDIUM_96
10 Years
25Km
R-squared
0.5148
Beta
p-level
FERTIL_ALTA
<2e-16
log(DIST_MIN_MAD + 1)
log(PREC_INV + 1)
TI_2006
ASSENT_06_NUNFAM
DIST_MIN_PORTOS
FERTIL_MUITOBAIXA
log(DIST_MIN_ROD_PAV + 1)
UC_2006

16. Statistical-based land use models

17. Statistical-based land use models
Driving factors of land use/cover change QUANTITY (at time t)
Driving factors of land use/cover change LOCATION (at time t)
Feedback on spatial
drivers
Rate and magnitude of change
for each land use at time t
(DEMAND)
Cell suitability
for each land use at time t
(POTENTIAL)
Feedbacks
Bottom-up calculation
Top-down constraint
Allocation of change combining demand and cell potential at time t
(ALLOCATION)
Land use
at time t-1
Time Loop
Land use map at time t
sources: P. Verburg, A.P. Aguiar

18. Simple Demand
Difference between years
Demand submodel

19. Potential Driving factors Potential map Transition potential submodel
Multivariate Statistics
Neural Network
Mathematics

20. Potential – CLUE like Potential map Deforestation Subtract from
Protected Areas
Subtract
from
Roads
Potential map
Transition potential
submodel
Ports
Deforestation

21. Demand Trend analysis Building Scenarios
Global economic model
Transition Matrix (Markov chain)

22. Allocation Demand Landscape map at t Landscape map at t+1
submodel
Rank-order
Stochastic
Iterative
Landscape map at t+1
Potential map at t

23. Scenario exploration: linking to process knowledge
Cellular database
construction
Manaus-
Boa Vista
Santarém
Porto Velho-
Manaus
São Felix/
Iriri
Humaitá
Apuí
BR 163
Cuiabá-Santarém
Boca do
Acre
Exploratory analysis
and
selection of
subset of variables
Aripuanã
Scenario exploration
Selection of variables – Priorities:
Policy oriented (accessibility and protection).
Importance as a land use determinant (based on knowledge and correlation to land uses).
Correlated > 0.80 excluded (some correlated > 0.50 to many too).
Definition of groups: using the 12 variables left, groups are formed allowing only < 0.50 correlated enter the same group. There are 2 or 3 groups for each space partition (see groups excel file)

24. Scenarios for deforestation in Amazonia (2020)

25. SimAmazonia Simamazonia www.csr.ufmg.br
Modeling conservation in the Amazon basin
Soares Filho et al., Nature, 2006

26. SimAmazonia Subregiões do modelo SimAmazonia.
Modeling conservation in the Amazon basin
Soares Filho et al., Nature, 2006

27. SimAmazonia: deforestation scenarios
Business-as-usual
Governance
Modeling conservation in the Amazon basin
Soares Filho et al., Nature, 2006

28. How good are statistical models?

29. Agents as basis for complex systems
An agent is any actor within an environment, any entity that can affect itself, the environment and other agents.
Agent: flexible, interacting and autonomous

30. Agents: autonomy, flexibility, interaction
Synchronization of fireflies

31. Bird Flocking No central authority: Each bird reacts to its neighbour
Not possible to model the flock in a global manner. Need to necessary to simulate the INTERACTION between the individuals

32. Modelling collective spatial actions
Space Agent
Agent Space
Benenson and Torrens, “Geographic Automata Systems”, IJGIS, 2005
(but many questions remain...)

33. Agent-Based Modelling: Computing approaches to complex systems
Representations
Environment
Goal
Communication
Communication
Perception
Action
source: Nigel Gilbert

34. Modelling collective spatial actions
Space Agent
Agent Space
source: Benenson and Torrens, “Geographic Automata Systems”, IJGIS, 2005

35. Agent-Based Modeling of Land change
Theoretical Models
Empirical models
Neste trabalho estamos interessados nos modelos empiricos, dado o interesse de utiliza-los na regiao da amazonia.
Porem existem diversos desafios, quando tentamos usar estes modelo na amazonia
Agent-based models (ABM) range from theoretical to empirical. Theoretical models use simple generalizable ideas, whereas empirical models require more complexity and case-specific data..

36. Different farm sizes, different actors
61,000 ha
30,000 ha ha
60,000 ha
50 ha
20,000 ha
200 ha
é possivel observar a heterogenidade das fazendas, onde podemos encontrar fazendas de mais de 60,000 hectares, como é possivel encontra milhares de fazendas com menos de 50 ha. Este é um dado provisorio do cadastro rural, aqui ainda está sendo realizado.

37. Frontier areas
Frontier: an area of changing resource use whose boundaries are continually evolving

38. Census data for São Félix do Xingu

39. How to place agents in a frontier?
São Felix do Xingu in 1985: farms are there, deforestation not yet started

40. Division of frontier in large and small farms
Use the history of São Felix do Xingu to make assumption about farms

41. Evolution of cattle/deforestation in São Félix do Xingu
Agora olhando a densidade de cabeças de gado sobre ha desflorestado. Podemos tambem observar o comportamento coletivo variando no tempo. Até 1995 existia muito desflorestamento para pouco gado. Dado que era uma fase de ocupação, existia outros atores, como madereiros, mineradoras e garimpo. Depois de 1995 começou a crescer a pecuaria, uma nova fase. E depois, principalmente devido a criação de areas especiais, houve uma estabilização.

42. Different actors and strategies in SFX
Nesta area de estudo, se darmos um zoom nesta regiao,.

43. Model initialization (estimated farm distribution)
(a) Deforested areas in 1985
(b) Estimated distribution of farms in 1985
Deforestation
Com estas informações, geramos este mapa de inicialização, onde as cores diferentes mostram os limites das propriedades.

44. Deforestation maps (PRODES)
The frontier “exploded” between 1985 and 1997

45. Land concentration in Tucumã

46. Evolution of land tenure in São Felix

47. How to place agents in a frontier?
São Felix do Xingu in 1985: farms are there, deforestation not yet started

48. Division of frontier in large and small farms
Use the history of São Felix do Xingu to make assumption about farms

49. Extracting patterns from sequences of images
M. Silva, G.Câmara, M.I. Escada, R.C.M. Souza, “Remote Sensing Image Mining: Detecting Agents of Land Use Change in Tropical Forest Areas”. International Journal of Remote Sensing, vol 29 (16): 4803 – 4822, 2008.

50. Hypothetical land tenure in 1985
(a) Deforested areas in 1985
(b) Estimated distribution of farms in 1985
Deforestation

51. Patterns of tropical deforestation

52. Agent states in São Felix do Xingu

53. Pasture management model

54. Deforestation: simulated x observed

55. Frontier evolution in Sao Felix
Consolidated (dark red), pre-frontier (light red), frontier (light green) and post-frontier (dark green).

56. Model results
Both the spatial values and the deforestation totals emerge as a result of the agent’s decisions
Model updates the support capacity of the region changes in response to agents’ decision.
Agents then sense how geographical space has changed and use this information in their decision-making.

57. Some caution necessary...
“Agent-based modeling meets an intuitive desire to explicitly represent human decision making. (…)
However, by doing so, the well-known problems of modeling a highly complex, dynamic spatial environment are compounded by the problems of modeling highly complex, dynamic decision-making. (…)
The question is whether the benefits of that approach to spatial modeling exceed the considerable costs of the added dimensions of complexity introduced into the modeling effort. The answer is far from clear and in, my mind, it is in the negative. But then I am open to being persuaded otherwise ”.
(from “Why I no longer work with agents”, 2001 LUCC ABM Workshop)
Helen Couclelis

59. GLOBIOM: a global model for projecting how much land change could occur
source: A. Mosnier (IIASA)

60. GLOBIOM: land use types and products
source: A. Mosnier (IIASA)

61. REDD-PAC: land use policy assessment
GLOBIOM, G4M, EPIC, TerraME
TerraLib
Land use data and drivers for Brazil
Model cluster - realistic assumptions
Globally consistent policy impact assessment
Information infrastructure