Scientific Cooperation between INPE and IFGI, University of Münster

The Brazilian National Institute for Space Research (INPE) and the Institute for GeoInformatics (ifgi) have cooperated for several years, e.g., in two ALFA projects, continuous research contacts, conference organization, exchange of teaching, and student mobility measures.

As there is a strong overlap in research goals, activities, and interests, this meeting aims at facilitating a more substantial and topically focused cooperation. Successful cooperation requires personal acquaintance and trust, which is why will be improved at a 10 day joint research seminar of the two groups in Brazil. The event main goal is to create a shared research agenda and outline a program to pursue it. The bilateral event is scheduled for March 10-19, 2009, on the premises and in the neighbourhood of INPE, Brazil.

The field of Geoinformatics is undergoing an important transformation. During the 1980s and 1990s, the scientific results produced by researchers in this area helped to set up the current billion dollar industry of Geographical Information Systems (GIS). GIS is now regularly being used as a corporate tool to manage large geospatial databases, and as a research tool for understanding our environment.

However, almost all of current GIS applications use static data, which represent temporal and change information too simply, if at all. The new generation of GIS, called GIS-21 (or “GIS for the 21st century”) will be different from GIS-20 (or “GIS for the 20st century”), thanks to scientific and technological advances. These advances include the distributed spatial processing on the Web and a new generation of mobile devices and remote sensors. The two institutes, ifgi and INPE, have been instrumental in pushing ahead the state of the art in these two directions: INPE as a world leader in remote sensing technology and spatial analysis methods, ifgi as a world leader in research and development for distributed spatial processing standards.

Currently, the main challenge for GIS-21 is to adequately represent information produced from the new generation of sensors. Sensor networks consist of small nodes (sensors) deployed in some geographical area. They can obtain different types of environmental data, from simple measures of temperature or pressure to more sophisticated applications for measuring earthquakes and biophysical parameters (Akyildiz et al., 2002). Recent developments in geosensor technology allow novel ways of obtaining environmental data. Geosensors can now provide new data for environmental science (e.g. climate models) as well as hazard warnings (for example flood alerts). These new sources of information were not available earlier due to high cost of measurement or to inaccessibility for analysts. Environmental sensor networks provide a ‘virtual’ connection with the environment, and allow new approaches to the study of environmental processes. There are many kinds of environmental sensor networks already used, including applications such as habitat monitoring, seismology, soil and water contaminants, air quality, and oceanography (Hart and Martinez, 2006). The ARGOS data collection system uses the METOP and NOAA satellites to access more than 20,000 data collection platforms worldwide.

Early sensor network research focused on the technological challenges of building and deploying the networks. Issues such as fault tolerance, reliability and scalability were given priority. Fault tolerance requires that even if some nodes fail, enough are alive (or redeployed) to produce useful data. Scalability involves placing thousands of nodes and protocols that ensure that these nodes communicate reliably.

As these challenges are being addressed, a new scientific challenge emerges: transforming sensor data into information for monitoring the environment. This transformation will require the capacity to model the processes measured by sensor networks. The challenge is to develop computer models that can bridge the gap between the low-level detail of geosensor networks and users’ high-level domain conceptualizations in terms of change, process, and event. These models should enable representating and defining different types of spatial events resulting from continuous changes in the real world. They should be computational, i.e., discrete and compatible with distributed data processing (Duckham et al., 2005; Worboys et al., 2006).

Along with geosensor networks, remote sensing images provide a second important and challenging source of new data for understanding our environment. The new generation of remote sensing satellites already launched or planned for the next decade will provide much new data, especially related to land change in the developing world. One example concerns tropical forests. It is known that tropical forests contribute to greenhouse gases emission, but there are large uncertainties on these measurements (Kintisch, 2007). Land patterns measured in remote sensing images reflect the diversity of the human actions, which occur non-uniformly in geographical space. We need new approaches that use Geoinformatics for producing better models of information in remote sensing images (Câmara et al., 2001). An environmental GIS-21 should be able to search for changes in a series of remote sensing images instead of the current search for content. The emphasis should not be placed on simple image classification procedures, but on capturing dynamics over the landscape (Silva et al., 2005; Carneiro et al., 2008).

Using multitemporal remote sensing data, a new generation of Geoinformatics-based methods is able to describe the change trajectories at local and regional scales. These methods can identify when a landscape object was created, how it evolved and what its trajectories of change were (Silva et al., 2008; Bittencourt et al., 2007; Pimenta et al., 2008).

References for Joint Research Topics

Geographic Information Science (GI Science) has evolved from research on automated mapping to a science concerned with information about processes in the human environment. Technologies like earth observation satellites, sensor networks, and mobile devices are generating vast amounts of data about the environment, but scientists and decision makers lack sound theories of change and models to represent and interpret it. This joint research seminar will target GI Science methods for Dynamic Environmental Sensors. These methods will integrate sensor data management, process models, ubiquitous computing technology, and semantic interoperability techniques. Together, they will provide modeling and reasoning capacity to support domain experts in observing, understanding, decision-making, and acting on changes in the environment. Two application areas will be addressed, namely Amazonian Land Use and Cover Change and Global Climatic and Environmental Changes Studies.

The Institute for Geoinformatics (ifgi) of the University of Muenster, Germany, founded 1994, is the first of its kind in Germany. ifgi is one of the biggest and most renowned GI institutes in Europe, with4 professorships and 60 graduate students). ifgi contributes to a wide range of GI research topics, e.g. GI design and applications, interoperability, semantics, usability, HCI and visualization. Currently, ifgi has three working groups and is currently filling the vacant junior professor position.

The Brazilian National Institute for Space Research (INPE) promotes and conducts studies, scientific researches, technological development, and capacity building in the fields of Space and Atmosphere Sciences, Space Applications, Meteorology, and Space Engineering and Technology, as well as in related domains, in accordance with the policies and guidelines set forth by the Science and Technology Ministry.

INPE and ifgi have cooperated for several years. Based on previous personal contacts, we initiated the ALFA projects eduGI.LA and, as a follow-up, eduGI.LA2, www.eduGI.net/eduGI.LA, which performed a summer school, mobility measures for students and teachers, virtual mobility by exchange of e-Learning courses. Apart from these projects, additional mobility measures for students and researchers have been performed without project funding. The INPE Director General, Dr. Gilberto Camara, has visited Münster as a guest lecturer in the ifgi Spring School 2008 (http://ifgi.uni-muenster.de/springschool/), for a 2006 block course on spatial analysis, and as a keynote speaker at the GIScience 2006 conference, the most renowned scientific conference in this field.

Wed, Mar 11

10:00 – 10:30 Welcome, introduction: Dr. Gilberto Camara (INPE), Dr. Werner Kuhn (ifgi)
10:30 – 11:15 Change we can model (slides) (Dr. Werner Kuhn)
11:15 – 12:00 "How to include human actions in Earth System Science Modelling?".(Dr. Gilberto Câmara)

12:00 – 14:00 Lunch

14:00 – 15:30 Excursion to Visitors Center and LIT at INPE

15:30 – 16:00 Coffee break

16:00 – 16:30 Metadata and information communities (slides) (Patrick Maué)
16:30 – 17:00 OGC® Sensor Web Enablement (slides) (Dr. Lúbia Vinhas)
17:00 – 17:30 Schema translation and data quality (slides) (Sven Schade)
17:30 – 18:00 Spatial modeling in urban inequalities and public health issues (Dr. Antonio Miguel V. Monteiro)

Thu, Mar 12

09:00 – 09:30 The spatio-temporal modelling lab @ ifgigeospatialmodellinglab.pdf (Dr. Edzer Pebesma)
09:30 – 10:00 TerraME slides (Dr. Tiago Carneiro)
10:00 – 10:30 INTAMAP project - An OGC Web Processing Service for automated interpolation (slides) (Jan Dürrfeld)

10.30 – 11.00 Coffee break

11:00 – 11:30 Amazonia LUCC patterns and process (slides) (Dra. Isabel Escada)
11:30 – 12:00 Sampling design for spatio-temporal processes kristina_helle_inpe_3_2009.pdf (Kristina Helle)
12:00 – 12:30 Biodiversity modeling slides(Dra. Silvana Amaral)

12:30 – 14:00 Lunch

14:00 – 14:30 Humans in the sensor food chain (slides)(Dr. Antonio Krüger)
14:30 – 15:00 Beyond SDI: integrating science and communities to create environmental policies for the sustainability of the Amazon slides(Dr. Fred Fonseca)
15:30 – 16:00 Towards a generalization of Geosensor Networks data stasch_generalization_gsn.pdf(Christoph Stasch)

16:00 – 16:30 Coffee break

16:30 – 17:00 Data suppression schemes for improving data collection using geosensor networks (slides)(Dra. Ilka Reis)
17:00 – 17:30 Closing the gap between Sensor Networks and the Sensor Web broering_slides.pdf(Arne Bröring)
17:30 – 18:00 Agent-Based modelling (presentation) (Pedro Ribeiro de Andrade)

Fri, Mar 13

08:30 – 09:00 Modelling Deforestation and its Intraregional Interactions in the Brazilian Amazon Rainforest. (Giovana Mira de Espindola)
09:30 – 10:00 Representative Agents for Regional Scale (Slides) (Sergio Costa)
10.00 – 10.30 Towards an algebra for spatio-temporal database (Presentation) (Karine Reis Ferreira)

10:30 – 11:00 Coffee break

11:00 – 11:30 Geographical data mining (slides) (Thales Sehn Korting)
11:30 – 12:00 Computational cognition of spatial objects (Slides)(Dr. Angela Schwering)
12.00 – 12.30 SISMADEN sismadeninpeifgi.ppt sismadeninpeifgi.pdf(Dr. Laercio Namikawa)

12.30 – 14.00 Lunch

14:00 – 15:30 Plenum (moderation Kuhn/Camara):
1. Identification of common research interests and complementary capacities
2. Items for a joint research agenda

15:30 – 16:00 Coffee break

16:00 – 16:30 Funding opportunities in Germany/Europe (Christoph Brox)
presentation, background materials, DFG template research proposal
16:30 – 17:00 Plenum (moderation Kuhn/Camara): Discussion of preliminary results, Plans for next week

1. Cellular automata that model change processes
2. Ontologies of change and processes
3. Processes and networks
4. Web technologies and processing of spatial data, SWE
5. Collaborative Modeling
6. Exchanging data and models between communities, integration of information
7. Reasoning on patterns (from static to dynamic patterns)
8. Integrating GI-technology into social environments: sensors of various kinds, communicate decisions
9. Correlation of different factors which influence change processes
10. Tool design: R, TerraXX, SWE (52N), Processing Services, Ontology-tools
11. Spatial temporal data modeling (e.g. combine SOS with algebra), dynamic feature generation
12. Scale and generalization, aggregation and uncertainty

• INPE: Problems, data, tools, spatial temporal algebra
• IFGI: SWE, SDI tools, Metadata repository, HCI knowledge, cognitive models, similiarity based reasoning,

1. How to move to a more collaborative Digital Earth? (fred, toni, pat, arne, davis, lúbia, sven)
2. How to make the interaction with GI-technology more usable? (angela, toni, jano, tiago)
3. How to share models? (fred, jan, lúbia)

• Brain Dump: Working Document (Google Doc)
• Abstract: Collaboration in the Model Web
• Abstract: Representing Environmental Models

Group 2: Geosensors

1. What are meaningful abstraction levels in the sensor food chain? (arne, christoph, werner, sven, jano, ilka)

Group 3: Change Modelling

1. How do humans reason on change and processes at multiple scales? (edzer, tiago, werner, sergio, pedro)
2. What are the minimal and necessary tools to model geospatial change?(karine, gilberto, lúbia, jan, kristina)
3. How to design tools to reason on process and change? (jano, tiago)
4. How do you design multi-scale models? (sergio, tiago, giovana, pedro, pat, christoph, ilka, kristina)

Group 4: Change Theory

1. How do humans reason about spatial patterns? (angela, giovana, gilberto, thales, thomas)
2. How to choose the right reasoning mechanisms and tools to analyze change? (jan, gilberto, kristina)

Template for each working group

1. title, organizations/partners/names of those involved
2. summary (finished)
3. state of the art
4. preliminary work (your own work in this direction)
5. objectives (appr. half a page)
6. work plan / schedule (most important), half the text of the entire applications; should clearly state why needed, how used.
7. staff cost, travel cost–what is needed to realize this work plan, conference related costs, visiting scientists, workshops
8. PhD and MSc topics

• Semantic Annotations & Models (PhD)
• Sharing models? Ask the community. (Master)
• R Models in the Web (Master)
• TerraME Models in the Web (Master)
• Development of suitable tools for different sensor abstraction levels (Master)
• Ontology (SWRL) of processes and data sources for retrieval and composition in the framework (PhD)
• Statistical methods for the conversion between the different abstraction levels of the framework (PhD)
• At which sensor abstraction level do we need to introduce objects (Master)
• Develop domain-specific theory of change for one domain (Master)
• Develop a generic theory of change (PhD)
• Generate a stakeholder-sensitive rain forest report (Master)
• TerraMe on a multi-touch surface (Bachelor)
• Interaction paradigms for natural interaction with time sequenced data (PhD)
• GeoTorrents (Master)
• Application of the algebraic view on sensors to the dengue fever (aedes aegypti eggs) use case (Master)
• Develop a domain-specific theory of land use changes in the Brazilian Amazon (Master)
• Develop a domain-specific theory of land use changes in urban areas in Brazil and Germany (Master)
• Develop a domain-specific theory of spatial changes in Europe (Master)
• Develop a General Theory for Spatial Patterns of Change (PhD)
• Analogical Reasoning on Spatial Temporal Patterns in the Brazilian Amazon (PhD)

Sat, Mar 14

10.00 Moving to Ilhabela

evening: midterm reception

Sun, Mar 15

9.00 - 10.15 Short plenum, then: Kick-off meetings of the 4 clusters

Deliverables for each of the 4 clusters

1. Cluster project proposal (input in cluster proposal template ) 1.1. Abstract 1.2. State-Of-The-Art, related literature 1.3. Work packages descriptions 1.4. Deliverables, time plan

2. Two publications (extended abstract at Ilhabela)

3. Identifying suitable tools for development and evaluation

4. Two topics for jointly supervised Masters- and PhD theses

5. One mobility measure proposal

6. Presentation of preliminary results (15+15 min. each) on Tue, Mar 17, 13.30-15.30

7. Presentation of final results (15+15 min. each) on Thu, Mar 19, 14.00-16.00

10.30 - 18.00 Excursion Ilhabela

Mon, Mar 16

9.00 - 19.00 Ongoing work of the 4 clusters

Tue, Mar 17

9.00 – 12.30 Ongoing work of the 4 clusters

12.30 – 13.30 Lunch

13.30 – 15.30 Plenum: Presentation of preliminary results (30 min. per cluster, 15 min. presentation + 15 min. discussion)

16.00 – 19.00 Hands-on INPE and ifgi tools

Wed, Mar 18

9.00 – 10.00 Writing the summary of the overall joint project proposal

10.00 - 12.30 Ongoing work of the 4 clusters

12.30 – 14.00 Lunch

14.00 – 19.00 Ongoing work of the 4 clusters

Thu, Mar 19

9.00 – 12.30 Ongoing work of the 4 clusters

12.30 – 14.00 Lunch

14.00 - 16.00 Plenum: Presentation of final results (30 min. per cluster, 15 min. presentation + 15 min. discussion)

16.30 – 17.30 Plenum: Summary of results

16.30 – 18.30 Plenum: Agreement on further step, close-out

19.30 Farewell party

Fri, Mar 20

11.00: Transport to Sao Paulo airport

Link to All Photos from Angela (please use the same login as for this wiki)

Jano's pictures (new pictures added!)

The following list shows the confirmed participants, including a brief abstract of their contribution to the bilateral event.

Change we can model. Our understanding of information semantics is now applicable to data discovery and integration. What is missing for an understanding of change are process semantics and a link between observations and processes. This talk will present an approach to process semantics and outline how it is expected to support change modeling.

Humans in the sensor food chain. Humans are often involved in providing sensor data to the complex mechanism of the sensor food chain (SFC). The SFC is the processing chain which starts with the raw data from the sensors, which filters and processes the data and which finally leads to an action of a decision making system. I propose to investigate closely how humans fit into the various stages of the SFC and how non-desktop technology can help them to provide support along the SFC.

Spatio-temporal modelling, image analysis, OS software. Automatic or semi-automatic satellite image analysis and classification is a task that is needed for processing large amounts of satellite images, needed e.g. for change monitoring and forecast activities. It requires development of both statistical analysis methods as well as hard- and software demands. The talk will discuss both issues and look at opportunities from connected fields of machine learning and pattern recognition.

Funding acquisition. The acquisition of funding and projects is essential for each university or institute. For example, the Institute for Geoinformatics in Münster is working with 7-8 staff members regularly paid by the university. But 30-50 additional staff members (students and research assistants) are working at the institute and are paid by funded projects. This talk provides the theoretical background for funding organizations and programs, partner acquisition, evaluation of funding programs, proposals/proposal writing, and it will provide the details of how to apply for a DFG research project, which will be applied in the second week of the bilateral event.

Computational Cognition of Spatial Objects. The research examines human cognition of spatial objects and strategies to acquire and organize knowledge about spatial environments and reason on them. Gestalt psychology postulates that we experience things as an integral, meaningful whole, which is more than just the sum of its parts. A computational model which is intended to reflect human cognition must be flexible enough to adapt a representation of primitive elements and produce a structured representation of the spatial object reflecting the perception of the meaningful whole.

Similarity and context modeling. Similarity is a key to human categorization and has been proven to be useful for information retrieval and decision support. While we have a good understanding of the similarity between entities and geographic feature types, measuring the similarity between (geo)processes is the major next step. I will present recent work on semantic similarity measurement and context modeling, focusing on its role in modeling change.

Schema translation and data quality. Integrating sensor data in decision processes often requires translating their formats and underlying conceptualizations. Accounting for accuracy and other quality parameters is a key part of this translation. I will present recent work on ontological approaches and how it can help to improve our understanding of the environment.

Metadata and information communities. The traditional metadata models have failed in practice. New ideas evolving in the social web provide better answers to information seekers and decision makers. I will present a software architecture applying these ideas, showing how it can be used for environmental information.

Towards a Generalization of Geosensor Networks data. The purpose of cartographic generalization is to improve the visualization of geographic data. In the context of sensor data, the main purpose of generalization will be to simplify the use of the sensor data for further sensor based applications such as simulations or early warning systems. I will present my first ideas towards a generalization of geosensor data.

Closing the Gap between Sensor Networks and the Sensor Web. The work of this theses focuses on bridging the gap between Geosensor Networks and the Sensor Web. The integration of Sensor Networks into spatial data infrastructures can already be realized by the means of OGC’s Sensor Web Enablement architecture which can be seen as one implementation of a Sensor Web. But there is a conceptual gap between these two layers of Sensor Networks and the Sensor Web. The contemporary way of facilitating the connection between the two tiers is often cumbersome and inefficient. Coherent concepts and methods are missing which describe and facilitate an infrastructure level to connect the two distinct layers by guaranteeing a strong performance and simultaneously ensuring a high level of adaptivity for ambiguous sensor types.

Sampling design for spatio-temporal processes. Optimal placement of sensors can improve the quality of models and predictions based on them. Monitoring of dynamical systems differs in many aspects from the static case. Directed and especially periodic structures are typical. Besides extrapolation becomes more important for predictions of the future whereas spatial modeling concentrates on interpolation. Therefore the deterministic trend is more important and sampling should also support parameter fitting.

INTAMAP project - An OGC Web Processing Service for automated interpolation. The INTAMAP project is an European project with the aim to develop an automated interpolation service. Several novel approaches to interpolation methods were implemented in “R” statistical environment and are offered through a web service based on the OGC Web Processing Service specification. The presentation will give a general overview of the project and the WPS to R interface in particular.

How to include human actions in Earth System Science Modelling?

OGC® Sensor Web Enablement. The OGC consortium provides an interoperability framework for Web-based access and control of sensors and sensor data. I will present a general view of this framework and a survey of projects using it, so the participants have a view about the technological aspects related do sensor data use. I expect to help the participants to have a critical view about this framework.

Participative GIS; Spatial Modeling in Urban Inequalities and Public Health issues.

Environmental Modeling. The Environmental modeling of the Amazonian -GEOMA is a cooperative network of researches whose objective is to develop models to evaluate and foresee sustainability scenarios for the Amazonian area. It was created to overcome the dimension of the environmental problems in the Amazon and the spread out of the scientific knowledge among different research institutes. We are focused on understand the challenges, barriers and goals facing by the GEOMA research team in the modeling processes of the biodiversity and processes of forest conversion associated to the human processes of land occupation. As examples of GEOMA first results we can present an initial database at regional scale, including high-resolution remote sensing imagery, and some valuable models and prospective scenarios of deforestation. However, to understand the biodiversity patterns and the human processes, in order to contribute to the sustainable planning for the region, it is necessary to increase the environmental and geospatial data available. Fieldwork campaigns and data exchange efforts are essential for the ability in providing reliable and consistent information for the scientific community and the police makers.

Amazonia LUCC patterns and process. The first results achieved through an interdisciplinary and multi-institutional effort conduct by the Geoma Network, aiming to advance in the comprehension of the new frontiers in the South of the Pará State, in the Amazonian region, examined the new patterns of deforestation and the underlying processes that are generating them. Our objective is to product information aiding to draw responsible public policies considering not only one dimension of the problem like roads infra-structure, as usual. Based on the analysis presented here we pointed out that only an integrated solution, that considers the main actors and the organization of economical processes in the productive chains over the region, would be capable to minimize the effects of deforestation and would drive an integrated development policy to the region bringing benefits to forest conservation and the local population.

SISMADEN. SISMADEN provides technological infrastructure to develop operational systems to manage alerts of environmental risks. The platform is an Open Source Computational System based on service oriented architecture. SISMADEN main services are: data gathering and formatting; analysis by comparison with risk layers or by executing models; risk model edition for alerts; and alert handling and management. The first version of SISMADEN targets environmental risks related to hidrometeorological information from forecasts and sensors. Sensors include Data Collection Platforms (DCPs) and Meteorological Satellite images. Main research questions are the validation of input from DCPs and images based on the the behavior of hidrometeorological processes, and the definition of aggregation functions and influence region for each available information based on the particular environmental risk.

TerraME. TerraME is a software platform towards spatially-explicit dynamical modeling which is based on the Cellular Automata (nested-CA) and Agent theories. It links cell spaces to spatial databases managed for Geographic Information Systems (SIG) allowing realistic environmental model development. I will present the TerraME models for scale, space, time and behavior representation. I will compare TerraME to other software platforms and discuss deeply the TerraME high level programming language form environmental model description. Different environmental models developed in TerraME will be shown: hydrological models, land use and cover change models, forest dynamic models and fire propagation models.

Beyond SDI:Integrating Science and Communities to Create Environmental Policies for the Sustainability of the Amazon. This presentation will explore ways to go beyond the traditional SDI (spatial data infrastructures) in the direction of the Digital Earth, with the objective of supporting environmental policies that will lead to sustainability. We use the Amazon region as a starting point for the discussion. Environmental policy making for a place such as the Amazon has to take into account that phenomena occur and are modeled in various geographic scales, ranging from microbiology to planetary climate impacts. There are also multiple and sometimes conflicting views on the same reality, including the many scientific disciplines, governmental and non-governmental views, and the view of the local populations. Currently, the combination of technologies, people, and policies that defines an SDI is probably the best approximation we have to solve these problems, but some important elements are missing. A broader SDI would be an enabler for understanding space, not only delivering general-purpose maps, but disseminating spatial data to support policies for sustainable development. We think it is necessary to go beyond SDI to integrate science and communities in the effort of creating, enforcing, assessing, and revising environmental policies. We discuss the limitations of current SDIs with regards to data and information flow, semantics, and community building. We also review the information needs and modeling challenges for SDIs when used as a support for environmental policy making.

Data suppression schemes for improving data collection using geosensor networks. Geosensor networks are instruments to sample spatio-temporal data. Although some applications require continuous monitoring of environmental characteristics, sending all sensed data to the user at a base station would spend the network’s resources very fast. Therefore, data suppression is a key strategy to get continuous monitoring without continuous reporting. To define a data suppression scheme, nodes and base station have to agree on an expected behavior for the nodes’ readings. Thus, nodes only send reports to the base station when their values do not fit to the expected behavior, which is used to estimate the suppressed data. Model-driven data suppression, for instance, defines the mean of a node’s observations as their expected behavior and models this means using temporal or spatio-temporal correlations. In this presentation, I will talk about my contributions to improve the estimates at base station considering their statistical quality and the robustness to aberrant readings of a suppression scheme.

Deforestation and its intraregional interactions in the Brazilian Amazon. The Brazilian Amazon rainforest is a largely diverse region, in which subregions with different rates of change coexist, due the diversity of accessibility, ecological, socioeconomic and political conditions. In recent decades deforestation has prevailed owing mostly to private investments in agricultural expansion, associated with large-scale cattle ranching, small-scale subsistence farming and soybeans expansion. Deforestation rates changed from 1988 to 2007, the total deforested area in the region varied from about 11,000 sq km to more than 29,000 sq km a year. In 2004, the estimated rate dropped to 27,379 sq km, and in 2007 to 11,224 sq km. The lower rates since 2005 have been associated to control actions conducted by the Brazilian government, including law enforcement actions and the creation of protected areas, and partly to lower commodities prices in the international market. However, there is an expectation that rates may start to rise again in the period of 2008-2010, due to rise in the commodities prices and the expansion of biofuel. Therefore, the objective of this work is to understand the intraregional distribution of deforestation rates, using quantitative methods to evaluate the possible impacts of alternative policies and decision making processes. The hypotheses of this work are: (A) the current hotspots of deforestation contributed differently to deforestation rates variation, as they responded to market and policy conditions according to their specific socioeconomic context; and (B) localized deforestation control policies applied to one municipality, such as the creation of protected areas or localized law enforcement actions, might stimulate the occupation of other areas in the medium and long run. This proposal adds to previous works increasing the knowledge about the last decade deforestation rates variation, exploring the intraregional multiscale interactions of distinct socioeconomic and political contexts.

Geographical Data Mining. Nowadays, several remote sensing databases are available. While they provide quick and comparatively inexpensive information about land use over large areas, information extraction in remote sensing databases requires adequate methods. Data mining tools can, in fact, increase the analysis of strategic data. However, few techniques for image data mining and information extraction in large image data sets have been developed. Although exists a large research effort in content-based image retrieval (CBIR) techniques, the specific problem of mining remote sensing image databases has received much less attention. The novel system for spatio-temporal data mining, called Geographical Data Mining Analyst – GeoDMA, is intended to run as an addon for TerraView software. TerraView deals with spatial datasets, comprising images, and regions (or shapes) resultant from segmentation process, or other GIS information, i.e. road maps. GeoDMA is being endowed with temporal analysis, allowing the user to input the same geographical region, but in different epochs, or periods. With such data (images and shapes), GeoDMA extracts several features, from spatial to statistical and spectral attributes, performing the complete data mining process, including attributes selection, training, classification, visualization and validation.

Agent-based modelling. Agent-based modelling (ABM) provides a bottom-up method for building complex systems. It has been gaining growing acceptance in various fields of science because of its capacity to show how collective social actions emerge from individual behaviour. One important challenge for ABM concerns how to represent human actions in landscapes. The agents have to be grounded to a representation of the world, on which they can sense and act. In these studies, ABM toolkits need to support different types of geospatial data. However, the demands and formalisms for creating agent-based models for geospatial problems are not well established. This work studies how geospatial data can feed entities of agent-based models and their relations for simulating complex spatial problems. We propose the four types of relations are necessary. They represent connections within and between agents and space partitions. Our hypothesis is that a Generalized Proximity Matrix (GPM) is a foundation to represent these relations. The GPM measures relations between objects in geographical space, combining data from Euclidian spaces and from topological relations such as network connections. To test our hypothesis, we are developing an ABM extension for the TerraME software, called TerraME-ABM. It will support creating relations using the GPM.

Environmental Modelling. There is a large diversity of modeling approaches in the literature. However, is possible distinguish two major approaches to design land change models: top-down and bottom-up. Top-down models originate from landscape ecology, and are based on remote sensing and census data. The Top-down approach describes a whole study area through of a statistical or mathematical formulation. On other hand, models conceived using a Bottom-up approach describe explicitly the actors of land change. This approach uses agent-based modelling theory, which consists of agents, an environment where the agents interact and rules that define the relations between agents and their environment. The selection of a given approach is very much dependent on the modeling goals. In the current State of Art of land change models, it is almost a tradeoff between heterogeneity and scale of analysis. One of the challenges of this scientific field is to combine both approaches, as purely bottom-up or top-down may be insufficient to represent biophysical and socioeconomic processes interactions at different levels of organization, from local to global. My research interests are to study these different modeling approaches, as well a possible multi-approach or hybrid solution.

Towards an algebra for spatio-temporal database. The recent technological advances in geospatial data collection, such as Earth observation and GPS satellites, wireless and mobile computing, radio-frequency identification (RFIDs) and sensor networks, have motivated new types of applications which handle spatio-temporal information. Examples include recording of animal tracking, transport systems, oil slicks on the ocean, and tracking land change objects. To satisfy this demand, there has been research on how to represent spatio-temporal information in geographical information systems (GIS). In the GIS literature, there are many proposals of spatio-temporal data models. However, according to Pelekis et al. (2004), a serious weakness of existing spatio-temporal models is that each model focuses on certain types of applications. Therefore, the models are not general enough to be a basis for a spatio-temporal GIS. To improve this, there is a need for a general-purpose spatio-temporal data model that can be used for a new generation of dynamic GIS. Thus, this work aims to define a first version of a spatio-temporal algebra for dynamic geographical processes.

Analysis and modeling of networks. In recent years, the analysis and modeling of networks have been the subject of considerable interdisciplinary interest. Researches on the dispersion of infectious diseases and land change models are among the main areas that can benefit from the study of networks. Nevertheless, its integration with the GIS field stills in an early stage. Most of the current GIS available in today's market and in the open source world lack a good support for general networks. They are heavely focused on modeling transportation networks. Our work aims to develop and integrate a general network support to the TerraLib open source library in order to offer better support for network analysis and to be able to spatially enable the network when needed.