Change Modelling

The cluster proposal is in a google doc here, ask Edzer for access.

Due to current software limitations and the complexity of spatio-temporal data, analysis of space-time phenomena often proceeds conditionally by first treating space as static maps or snapshots and then showing their evolution over time, or by first collecting time series at individual locations and then analyzing their variation over space.

In this work package we will develop methods and tools for unconditional analysis of spatio-temporal processes. We will try to find the necessary representations and operations for integral analysis of spatio-temporal phenomena, based on the spatio-temporal data currently observed that include earth observation imagery, tracking data, meteorological data and space-time point patterns.

Secondly, we will explicitly address the human role in changing environmental systems. Humans and institutions are the major drivers of environmental change today, but models that take into account their motivations and the way they make decisions or reason about their environment are still in a very early stage. To improve on this, we will develop the necessary methods and tools for taking these drivers into account in modeling land use change.

Thirdly, we will deal with calibrating and validating models. To optimize the correspondence between modeled representations and observed processes, the most influential model parameters need to be calibrated; calibration in dynamic, agent-based and cell-based models is a new research frontier. Validation methods will be extended to assess the degree to which the model approximates reality, for state as well as change in the modeled system.

So, this group is interested in three main questions:

1. Spatio-Temporal Data: Representation and Analysis
2. Modeling Human and Institutional Aspects on Multi-Scale LUCC Models
3. Assessement of Lange Change Models

1. Survey of web service specifications for spatio-temporal data
2. Spatio-temporal analysis of remote sensing imagery time series

1. PhD thesis topic: Spatio-temporal patterns mining in image time series
2. Master thesis topic: Selection and filtering in space-time fields
3. PhD thesis topic: Atomic functions and graphics primitives for analyzing spatio-temporal fields
4. PhD thesis topic: An agent architecture for simulating human behavior in environmental change models
5. PhD + master topic: Tools for Assessment of Multiple Scale Land Change Models

• several proposals for mobility measures are included in combination to the relevant PhD and MSc proposals
• Jan Dürrfeld may request a mobility measure, not as part of this proposal
• We consider a mobility measure request for Lubia Vinhas, to visit Muenster in 2010 (or later) for some months (needs confirmation).
• If all these proposals get funded, it makes sense to request a mobility measure for Edzer Pebesma to visit INPE.
• it would be useful for Karine Ferreira to visit ifgi in the final year of her PhD (2011), max. 1 month

At INPE, scale is the resolution and extend of space-time-behavior resolution in a model. Contrary, the area / volume which is represented in a measurement is its support which again is not similar to the uncertainty of the mesurements.

Application domain: environmental integrated modelling (in the wide sense), modelling deforestation in Brazil (in the narrow sense)

Motivation for modelling:

1. supporting decision on a political / society level
2. supporting tools for decision making at the indivual level, or middle levels of actors

{- modelling as a means to convince people}

What are the steps in building models?

1. decide what kind of actor or process do we need to represent in our models? (actor: something that acts; process: result of the action, or sequence of changes)
2. a conceptual theory/theoretical foundation (e.g. system theory/cellualar automata/differtial equation/random field models/…), or some coctail of them / new method
3. how do we design the model? Which tools do we use to design spatio-temporal models?
4. implementation: what tools do we need to implement the concepts?
5. implementation: decisions about extent/resolution/scale
6. which data can and should we use, and how can we access?
7. do we need calibration?
8. how can we evaluate the model results?

How do we organize all these steps? (sequentially/cyclical/spiral–increasing complexity/random)

After lunch:

What do we have right now, what can we do?

What are the missing aspects?

Deforestation monitoring: collecting, combining, classifying all available imagery without clouds; look into the past. Methodology is fixed. Deforestation estimates take cloud cover into account.

(classes for monitoring: water, forest, deforested area, no forest, clouds)

Once a pixel is deforested, it is masked out for the next year(s).

In case of false positives, there will be feedback from local sources.

some changes are detected years after they happened (clouds)

modelling: for INPE means looking into the future (-goal), including all possible explanatory factors.

1. identify factors that cause change
2. look at their relations by statistical methods
3. try to understand process
4. formulate theoretical models
5. implement
6. model past, in case of success: predict future.

1. effect of parameter uncertainty/error propagation
2. methods and tools for the calibration of spatially explicit models
3. methods and tools for the validation
4. generation of many layers for Monte Carlo results
5. analysis of Monte Carlo results
6. human behaviour modelling

Space-time considerations:

1. what data types do we need to represent space-time phenomena?
2. which operation do we need on these data, or on combinations of these data types?

Other things missing in TerraME

1. complex networks (?)
2. graphical interface
3. integration of TerraME with R
4. HPA

Focus?

1. Jan: spatio-temporal data types and modelling
2. Lubia: ST data types and operators in general
3. Tiago: foundations and tools for multi-scale modelling
4. Karine: ST data types and operations
5. Pedro: calibration & multi-scale modelling in ST modelling
6. Sergio: agent-based modelling for land use change
7. Kristina: calibration in ST modelling
8. Edzer: validation and the stochastic dimension of ST models

This group decide to split in two sub-groups:

1. one dealing with modelling aspects;
2. one dealing with spatio-temporal data types, operators and the tools to implement it Sub-group: ST-Data