Written by Tom Hengl

GeoEnv is a series of conferences that brings together scientists using geostatistics as a data analysis tool in various environmental fields from ecology, to geological mapping, climatology and geomorphology. Since 1996, GeoEnv conferences have been held biennially at venues across Europe. From the first conference in Lisbon, the event has been staged in Valencia (1998), Avignon (2000), Barcelona (2002), Neuchâtel (2004), and Rhodes (2006). The 2008 geoenv conference was held at the University of Southampton in the period from 6-10 September. From Saturday morning (06.09) to Sunday evening I followed a pre-conference workshop “Workshop on applied spatial data analysis with R“, which was run by Edzer Pebesma, prof. from the University of Munster. The course basically followed his new book just published by Springer:

• Bivand, R., Pebesma, E., Rubio, V., 2008. Applied Spatial Data Analysis with R. Use R Series, p. 400. Springer, Heidelberg, pp. 378.

Although the course was relatively short, I learned a lot and it was worth “every penny”. Or as the workshop moderator put it: “the price to pay to R is the time you spend on learning and improving it”. Unlike other commercial software that offers much support but then limits to users in their requests to improve it (R users develop things at a faster pace than is the case for commercial software). “R became equal to S already in 1998, but it was not until 2003 when Roger Bivand animated a large group of users to develop a support for spatial data and basically put R as a programming environment that can effectively replace commercial GIS packages”. It is useful to know that 70% of the development of R is done by Brian Ripley. His package MASS is still the most comprehensive R library among many. The free exchange of geographic data is mainly possible due to the GDAL library (originally developed by Frank Warmerdam), which allows the import of various spatial data in R.

R is today seriously competing with commercial GIS software such as ArcInfo. In fact, one of the key developers of the code for ArcInfo, Konstantine Kryvoruchko, recently admitted that the spatial analysis in R has reached a significant level (although Kryvoruchko finds ESRI packages more suitable for operational projects). Kryvoruchko has been recently active with writing a book with a working title “Introduction to Spatial Statistical Data Analysis for GIS Users”.

On Monday 8th the conference started. There were three keynote lectures at GeoEnv:

• “Geostatistical Analysis under Preferential Sampling” P. Diggle
• “Perspectives on Geostatistical Downscaling” P. Kyriakidis
• “Components of a real-time nowcasting and forecasting system for distributed hydrological models” M. Bierkens

I also gave two talks – one for the whole audience of GeoEnv and the second within the session biological applications of geostatistics:

1. “Special presentation: Geostatistics viewed through bibliometric indices: most cited publications, temporal trends and geographical distribution of scientific excellence.” T. Hengl, B. Minasny and M. Gould (see the published article)
2. “Mapping bird nest densities using kernel smoothing and regression-kriging.” T. Hengl, A. Radovic and E. van Loon (see the published article)

During my second talk, Richard Webster and Marc Naura gave me useful feedback – they were both enthusiastic about this approach, which is a good sign. Dick Webster remembered that a similar approach has been considered by a group of colleagues already in 1991, but from then, everybody has used only binomial GLMs, often without considering the geographical position of the points.

• McNeill, L. 1991. Interpolation and smoothing of binomial data for the Southern African bird atlas project. South African Statistical Journal. 25, 129–146.

In summary, I enjoyed the geoenv2008 conference very much. Plus the pre-conference workshop was excellent (in fact, this was my first R training ever). Edzer gave me several tips to improve my data analysis in R. These include:

• more compact ways to implement loops in R via lists;
• script to export map legends from R to Google Earth and similar geographical browsers;
• tips to fit space-time variograms in R by considering space-time anisotropy (see also the paper by Pebesma and Brus published in JABIS);
• approaches to fit non-linear regression models via GLMs;\par 5. tips to save geo-metadata – choice of format (we considered Idrisi format and ArcGIS metadata XML tags);

I will slowly start entering all this new knowledge into our EcoGRID user’s guide, so that anybody can actively follow the code developments.

The next GeoEnv conference will be held in Gent in September 2010.