Simple features for R

The current implementation of points, lines and polygons in R does not completely map 1-to-1 with the simple features model of the OGC. This has disadvantages, e.g. that some kind of data cannot be read through he readOGR channel. The goal of this study/MSc/Diplom project is to contribute this model to R, notably to the sp package with classes and methods for spatial data. As part of this project, and perhaps having a uch greater impact, an R driver in the GDAL and OGR libraries will be developed, so that all software that reads data through these drivers e.g. Quantum GIS) can directly access spatial data from an R data base.

(Possibly Co-supervised by Barry Rowlinson and/or Roger Bivand)

Uncertainty-enabled Sensor Web

Observations are increasingly abundant in the World Wide Web. This enables several users ranging from domain experts up to non-experienced users to use these observations in their analyses and applications. However, the observation of spatial and temporal distributed phenomena is always subject to errors and uncertainties. Hence, a common mechanism for integrating uncertainty information into web-based observation systems is needed. The student will deal with questions like: How can uncertainty be integrated in common observation models of the Sensor Web? How can observations be filtered by certain quality constraints?

Please ask Prof. Pebesma or Christoph Stasch for more information.

Usability of Spatio-temporal Uncertainty Visualisation Methods

Representation of spatial and temporal distributed phenomena is always subject to uncertainties due to measurements or model errors and simplifications. It is important to provide such information together with an uncertainty estimate to allow inter-comparison and enhanced decision making. A plethora of methods has been developed to communicate uncertainties to the user. Categorisation of these methods allows automatic selection of appropriate uncertainty visualisation techniques based on data requirements to serve data to the user. However, a comprehensive categorisation and assessment of usability with regard to e.g. user background is currently missing. The work will focus on a categorisation of existing methods for automatic selection and planning and implementation of one pilot, and main user study to gain knowledge about usability of different visualisation methods for probabilistic uncertainty in spatio-temporal data.

Please ask Prof. Pebesma or Lydia Gerharz for more information.

Provenance in the Model Web

Geospatial data is the result of a process (model) chain converting raw data gathered by sensors in the field to spatial information such as thematic maps. To help users to make correct decisions metadata has to be provided such as quality information, but also information about the original source of information, and the processing that took place. In processing workflows dealing with the processing of observations (e.g. interpolation) information just about the sensors is not sufficient, as provenance information about the subset of observations which is used in the chain is crucial. One approach for a provenance model is the Open Provenance Model (OPM). While there have been attempts to extend this model for spatial data, a common approach for using the OPM for observations which vary in space and time and are processed by environmental models is currently missing. The student will deal with the following questions: Which provenance information is required in the model web? How can the Open Provenance Model be adopted to fulfil these requirements?

Please ask Prof. Pebesma or Christoph Stasch for more information.

Machine learning methods for pattern detection in human activity tracks

Low-level sensor data like GPS, RFID or WLAN positioning providing position and time, can be used to infer on high-level behaviour of the wearer. For personal exposure monitoring it is highly relevant to detect behavioural patterns like transportation mode, indoor visits and physical activity. Automated extraction of behavioural information out of simple sensors simplifies the collection of large activtiy datasets for epidemiological studies. Further it enhances the analysis of personal exposure by adding geographical information to the common diary activity records.

Selection and filtering in space-time fields (Master) in cooperation with INPE (Brazil)

Given a data structure for regular raster-based space-time field data, where attribute is continuous or categorical, a prerequisite for doing powerfull spatio-temporal analysis is to have access to the most elementary operations on these fields. The first is selection: functions and are needed to subset a full ST imagery by (i) selecting a spatial region, (ii) selecting a time period, (iii) subsampling at a lower spatial resolution, or (iv) subsampling at a lower time-resolution. The second function is filtering; one would like to be able to write filters in the spatio-temporal domain that allow the usual analytical operations of smoothing and differencing. Especially differencing in space-time is a promising new tool. The resulting space-time gradient, a 3D vector with a direction in ST and a length, is a promising tool for finding temporal changes in spatial gradients, or for finding space time interactions such as region growth. To make these instruments useful in integrated, unconditional spatio-temporal data analysis, the research requires the choice of useful, intuitive semantics as well as efficient implementation.

Supervisors: Prof.dr. Edzer Pebesma, dr. Lubia Vinhas (INPE), Karine Reis Ferreira (INPE)

Mobility measure: The ifgi MSc student should spend 2-3 months at INPE to get familiar with the space-time imagery data base structures and currently existing functions currently available in TerraLIB? .

Temporal Extension of the Java Topology Suite

The Java Topology Suite is an Open Source API for 2D spatial predicates and functions and used in a lot of projects such as Geotools or the 52°North implementations. Currently, JTS can only deal with spatial geometries. The goal of this thesis is a concept and an implementation of a temporal extension allowing to deal with spatio-temporal geometries. The student should be interested in spatio-temporal data modeling and algorithms and should have good Java programming skills.

Please ask Prof. Pebesma or Christoph Stasch for more information.

Handling massive datasets in Web Processing Services

The idea of the Model Web is to expose environmental models in the Web to find, use and easily couple models. However, one major challenge in realizing the Model Web is how massive datasets can be exchanged and processed in the Web. Therefore, the student should examine how current implementations of the OGC Web Processing Service such as the 52°North WPS, pyWPS or the deegree WPS behave when processing large datasets. Furthermore, assuming that there is space left to optimize the handling of massive datasets in processing services, the student should work on mechanism to optimize the processing regarding storage consumption and runtime performance. Therefore, common mechanisms such as persistant storage of large datasets, efficient caching strategies, or parallelization of processing tasks should be investigated. The student needs good Java programming skills and an interest in Web Service technologies and optimization.

Please ask Prof. Pebesma or Christoph Stasch for more information.

Integrating R functionality in the Sensor Web

The R project for Statistical Computing provides a free software environment for statistical computing and graphics. Integrating the huge possibilities for statistical analysis and visualisation in the Sensor Web which aims to provide all kinds of different sensor observations in the Web would be highly beneficial. While there are already approaches for integrating data from the Sensor Web in R, for example the SOS4R project, this thesis should investigate how R functionality can be provided in the Sensor Web. The INTAMAP service provides an example for a service that provides interpolation methods for observations provided in the Sensor Web. Taking this as an example, the student should try to answer the question whether a more general approach can be developed that allows for integrating R functionality in the Sensor Web, such as computing spatio-temporal aggregates or predictions. Knowledge of R and Java programming skills is of advantage for the student who wants to deal with this problem.

Please ask Prof. Pebesma or Christoph Stasch for more information.

Topics that have been taken

An SOS and interpolation WPS for air quality over Europe

The Apmosphere GMES project (http://www.apmosphere.org/) has developed an approach for interpolating air quality variables over Europe, using relevant indicators such as altitude and meteorology into account. The INTAMAP project (http://www.intamap.org/) has developed a WPS for automatic interpolation of environmental variables. Bringing the two methodologies together with air quality data from the EEA (http://air-climate.eionet.europa.eu/databases/airbase/), a system will be set up that provides air quality data over an SOS and interpolates them on demand. Contact will be sought with the EEA to get access to unvalidated air quality measurements in near-real time.

Inferring space-time correlations from tracking data

When we measure some quantity using a mobile device, and the device moves while collecting data, it is in principle impossible to tell from the data whether variability results from moving in space or in time. For the analysis of data, e.g. a spatio-temporal mapping or the estimation of space-, time- or space-time averages, it is needed to quantify variability and autocorrelation both in space and in time. This work will look at how different measurement strategies can be compared with respect to how well space-time correlation can be inferred. It is anticipated that sampling will take place in simulated realities (computers), as well as in the real world, e.g. with a hand-held particulate matter monitoring device (topic has been taken).

-- HolgerFritze - 11 Apr 2009