Written by Tom Hengl

After 150 km of cycling (Jan Janssen classic) Gerard Heuvelink and I rushed to Schiphol to catch an airplane to Shanghai, and join the Spatial Accuracy 2008 conference. Spatial Accuracy is the biannual meeting of the ISARA (International Spatial Accuracy Research Association), researchers intersecting fields such as forest, land cover and climatic mapping and (geo)statistics. The group was initiated in 1994, and ever since seven meetings were organized: three in the USA, two in Europe, one in Canada and one in Australia. To improve the geographical representation and to make Accuracy truly a global research community, the 2008 conference was put in China (Shanghai).

In Shanghai, two local PhD students were waiting for us and took us from the airport to the hotels. We soon discovered that the life of a PhD student in China is really tough: neither of the two could recommend a restaurant or a bar to eat something and relax — the last time they had a beer in a bar was two years ago (one actually never before had a drink in a bar); neither of the two ever left China; when I asked them about how the Shangai coastline looks like, they could not tell because they can not afford to go and see it (it is only 60 km away!). But if they want to graduate, they do need to publish at least one peer-review paper in international journals (just like the PhD students from western countries).

Monday and Tuesday 23-24 June I followed the workshop called “Spatial Uncertainty Propagation” that Gerard and James Brown have been running a few times already. Excellent course — well prepared, balanced combination of theory and hands-on + the case studies were relevant and illustrative (authors published several publications where the details are given). The course was definitively for advanced users (I guess I fit the profiles). Many could simply not digest using R/DUE, running geostatistical simulations and interpreting the final results, all in two days, but I admire the courage. FYI, the complete course is now available on-line at http://spatial-accuracy.org/workshopSUP. The slides are an excellent guide to someone that would like to quickly go through the theory and get an idea of what is error propagation really about. The DUE software is impressive and extensive, but from my discussion with James, there are no plans to continue its development; the same type of analysis can be run in R, but one needs to script everything from the beginning. I would have definitively designed all course in R instead of expecting students to move data forth and back from R to DUE and gstat.

The conference officially started on Wednesday. The conference organizers received about 120 papers that they printed in two volumes of the Conference proceedings entitled “Spatial Uncertainty” and “Accuracy in Geomatics“. The conference program was rich with 8 keynote lectures (truly leading scientists in the field of spatial data quality and uncertainty analysis), more than 70 oral presentations and more than 70 posters. As expected, the majority of the participants came directly from China (apparently there are 120 Universities (!!) in China with GIS study programs; the annual GIS China society meeting gathers more than 1500 researchers). The (English) language is (still) a big problem for scientists in China in general, and especially for older generations. Even the ones that lived for years abroad will have difficulties and need to be listened to with care. I kind of understand the difficulty because the Chinese and English/European languages really really differ — we had serious problems pronouncing some simple words in Chinese — even “thank you” and “hello”. Still, I was surprised by the amount of the work done in China (I am slightly less happy with the quality — PhD students tend to copy both topics and style of writing of their supervisors a lot).

The conference was opened by the keynote speaker Michael F. Goodchild (“Spatial accuracy 2.0“), a professor of geography/GIS at the University of California, Santa Barbara. Goodchild has been in the field (geoinformation science) from its beginnings, which really makes him competent to look at it from a much broader perspective. Uncertainty in spatial data is still an unclear concept to many GIS teams, which Goodchild illustrated with GE that provides location information in arcdegrees rounded to 6 decimal places (10–50 cm), although the true accuracy of the background images and scale of viewing can be way coarser (this demonstrates that GE is not really clear about the spatial accuracy issues of their products). We have experienced rapid growth of remote sensing and GIS technologies and applications in various fields, so that one might ask “do we need to still worry about the spatial data quality at all?”. Even if the GI data seems to be more precise, more detailed and richer in content then even, there is no danger for spatial accuracy. Goodchild: “There will always be spatial accuracy because there is no perfect spatial data”. Likewise, Goodchild believes that evolution of geoinformation science will continue as “spatial data today is even more pervasive”. The last few years seem to be an inflexion point in the evolution of GIS (GE registered more than 3 million downloads; MS started its own browser called “Virtual Earth”; many are likely to follow). “GIS is becoming more and more popular!”. Goodchild further introduced the reality of what is called “WEB 2.0” (new generation web that will be dominated by users; i.e. user-generated web information) and discussed how is this changing context reflect on the GI data production and use. Users (amateurs/volunteers) are increasingly becoming involved in the mapping processes. Goodchild called this “community mapping” or “participatory mapping” — typical examples are GE communities; [www.wikimapia.org] and [www.openstreetmap.org]. I immediately thought of the Dutch websites that are used to input the ecological observations by volunteers [www.waarneming.nl]. The consequence of WEB 2.0 is a need for METADATA 2.0 — Goodchild believes that users are now more engaged also in the metadata production. It appears that metadata 1.0 has never reached its goals — “I never looked at the metadata myself. Everything you need from metadata is already obvious if you know the maps/producers. Instead, I am interested in users’ experience — which data are they using, for which purposes? which not and why?”. Goodchild strongly believes that Metadata needs to be popular and ‘user-centric’. This is the “wiki approach to metadata generation”: information on users’ usage statistics, most common applications, problems and benefits will become even more relevant for the proper use and update of the data, then the traditional product lineage (metadata 1.0). We further discussed that this does not mean that surveying/mapping/GIS will disappear as a profession. The future of GIS is the hybrid approach — mapping agencies are needed to build data frameworks, provide standards and guarantee the quality, while the users need to shape outputs and possibly enrich maps with (spatio-temporal) detail.

Deren Li, Wuhan University, President of the Chinese Society of Geodesy, Photogrammetry and Cartography (a living legend in China), gave an overview of research activities in China connected with land survey, mapping and GIS. Prof Li was, for example, in charge of disaster mapping connected with the recent earthquake in Wenchuan county. We saw how satellite images (including SAR and LiDAR high-resolution data) and aerial photographs were used to rapidly inventory the damage, estimate the volumes of debris flow, detect buried buildings and estimate new water flow directions (the earthquake completely changed the river path making many new lakes). The whole mapping project took less than one week, after which a complete GIS was provided to government agencies and used as the key support for rescue operations and engineering. Li also showed several screenshots and demonstration of their GIS browsing package called GeoGlobe, i.e. ‘the Chinese version of GE’ (similarity was more than obvious!). GeoGlobe has been designed specifically for China, for LBS and the Mobile Mapping System developed in Wuhan.

Carol Gotway made a shortlist of topics she believes are ‘hot’ i.e. will/should receive attention in the near future: (1) geostatistics in non-euclidian space (i.e. space that accounts for barriers, streams, disease transmission vectors etc.); (2) assessment of spatio-temporal support (“spatial prediction methods will be increasingly compared at various spatial/temporal scales; users are increasingly doing predictions from point to area support and vice versa); (3) kriging is increasingly used with discrete data and uncertain data (this emphasized the importance of using Bayesian-based models) and (4) geostatistics as a tool of politics. Geostatistics is today finally an operational tool. Gotway: “we are trying to facilitate better decision making, better science, more dynamic research”. I specifically asked how she feels about automated mapping (‘Deus-ex-Machina’) and does she thinks that traditional geostatisticians that fit variograms visually are becoming redundant? We both agreed that automated mapping is definitively needed considering the quantity of the input data and that the computational complexity of the statistical models is gone wild. Gotway: “Automated mapping is a necessary evil. Users want to obtain maps from points in real time”.

Alfred Stein, professor of statistics at ITC, reviewed the techniques and trends connected with the use of remote sensing images for monitoring, prediction and decision-making. What Stein calls “Image mining” comprises the classification and segmentation of images in space and time. Image segmentation in simple term is a conversion from raster to vector. Except today we aim at producing clean polygon/line maps directly from RS images (for CGE this would be an increasingly interesting theory e.g. to delineate and monitor individual birds from the radar images). The main objective of building image processing frameworks is to optimize the quality of the outputs and provide outputs that fit decision-making purposes. Stein emphasized that (spatial data) quality control becomes crucial for such projects. Remote sensing is still far from fully automated detection and measurement of objects from RS images. Objects are most commonly fuzzy, dynamic, poorly defined, vague. Stein: “Does an object or boundary really exit? Do we really have something that we call a boundary?”. The team at ITC/TUD has been increasingly interested in the fuzzy GIS object structures. This is a completely new area because no fuzzy GIS formats yet exist and the GIS algebra gets more complicated: instead of binary decisions, each query needs to consider a range of possibilities (the computations increase exponentially). In fact, a new fuzzy arithmetic is needed: distance between the vague objects gives a range of values; if two cities are defined as fuzzy fields, then there is a range of distances that follow a certain curve. There is a lot of room for further research in GI science. Stein: “scale is still an important issue!”. Scale in space, scale in time, scale and object complexity, automated scaling and merging of multiscale data. Much about these topics can be found in the book Stein recently edited with the colleagues “Quality Aspects in Spatial Data Mining” (CRC Press).

At the end of the conference, Gerard Heuvelink read some of his fresh notes he wrote during the conference. Gerard’s impression is that there is a need to continue with research in the field of spatial accuracy. The group needs, however, to get more formally organized and get more ‘visible’ (among the mainstream disciplines). Maybe the GIS curricula should consider including “spatial uncertainty modelling” as separate modules? Gerard further listed topics that will need to be tackled in the near future (in order of importance): (1) new metadata standards and guides for data quality assessment; (2) sensitivity analysis of the error propagation (“what is the uncertainty of our uncertainty analysis?”); (3) new fast and robust data mining tools; (4) analysis of space-time data and development of tools for such analysis “we need to take a look at what is hot in the world of GIS (e.g. LiDAR data, meteorological images, Google Earth applets), and then develop uncertainty analysis frameworks for such applications”; (5) development of generic computational frameworks (generalization of error propagation); and (6) powerful computational tools that allow near-real time accuracy assessment.

In summary, I enjoyed meeting many new faces, especially Carol Gotway from the Centers for Disease Control and Prevention (USA) and many of the Chinese PhD students. I am looking forward to the Accuracy 2010 event in the UK.