OSGeo Postprocessing

OSGeo Postprocessing

It’s time to put down OSGeo impressions before it starts to fade. Very good conference on many levels, learning loads, meeting formerly virtual people, and new people, spreading a bit of worldkit-georss gospel. worldKit made it on the tshirt :). Overall there is a great sense of community, and great sense of happiness in this work. Buzz. Geospatial technologies are at this intersection of technology, science, politics, and the natural world that just make it the right place to expend useful energy.

First off, O’Reilly had a good showing, met Simon the editor, the authors (in a new locale), and actually held the book in physical form. Met author Tyler Mitchell of the other book (will need to pick up this one too) and Nat Torkington, putting the finishing touches on a promising Where 2.0.

My appetite was primed to get more involved with projects like mapserver and GRASS; I’m just waiting for the excuse. Part of my presentation, I discussed integrating worldKit and mapserver: from discussions afterwords, it’s clear that the way to proceed is to learn mapscript to build the tie-in. It was good to learn that ECW is supported in GDAL since I have a project with this format that’s slipped by. On GDAL, we’re talking about adding a driver to support geocoded RSS (something I’ve done by parsing ogrinfo output, but this would be tighter).

GRASS engages the analysis portion of geospatial, while I’ve focused on an approach to spatial data infrastructure and visualization. Marcus Neteler keynoted on the history of GRASS (another military project dating to 1984) and its capabilities. The case studies were what engaged me. Every step was fascinating, from data acquisition, analysis, visualization, and working outcome, so I’ll go overly detailed. To mitigate collisions between vehicles and deer in Italy (wildlife is classified as state property there), they identified deer migration corridors and high density accident areas by analysing road curvature, 25 variables of land cover, terrain complexity, and traffic data from sensors, with support vector implemented in GRASS and used to derive 4 risk classes, leading to signage and underpass mitigations. Machine learning within GIS, awesome. The other case study, on unexploded WWII bombs dropped in Italy and still undetected, analysed military attack reports from the US/UK, terrain (bombers sometimes had to unload just to get over mountains), and crater detection from aerial photos. Combined with current census data, they could draw up costs and evacuation plans for likely bomb sites.

From working on integrating WMS sources into mapufacture, I’ve become interested in techniques for dealing with these huge data sets. Schuyler laid out his methods for setting up the vmap0 WMS. The key seemed to be Stephen Woodbridge’s shp2tile and ogrtindex, which pyramids vector data in a similar way to prepared raster sets. Vector generalization is also key, for simplifying unnecesary complexity in Tiger (Douglas-Peucker and Grid-Based algorithms) (and I’ve been interested in this for preprocessed flash representations of vmap0). Schuyler’s idea for a distributed WMS cache, for things like Landsat7 was well received(JPL could send astronauts to the moon, but can’t keep a WMS up and running). Any sort of caching is a huge improvement, Stephen implemented with mod_rewrite (and as Anselm implemented with squid for civicmaps). The developer of Ka-maps talked a bit about their demo, and relevant problems with label edge buffers and repeated labels (perhaps solved by creating larger “meta” tiles).

There’s very healthy competition in the visualization space with mapbuilder, kamaps, and worldKit. AJAX is obviously hot, though I still think there are some advantages to Flash, like real vector drawing and smooth image manipulation (which I laid out during my presentation following a Flash-hatedown in the previous talk :). Interestingly, in Dirk’s closing keynote on the Apache Project and lessons for Geospatial, he sighted the core difference being that GIS is visual, closer to the user, and solution oriented .. less a utility like a webserver, more motivated by “visionary” led development, in an unsettled innovative arena, so many features of Apache’s processes don’t necessarily apply. Perhaps this is why mapserver is so well developed, while open source interfaces still lag behind. Think there’s also something about the data that differs between the geospatial and traditional web; interoperability is a core motivation. Still the no agenda affect and cost benefits of sharing vs hoarding are usefl ways to examine projects.

One thing I had to repeat often was that worldKit was not actually open source, but did interop with open source projects and supported the creation of open data. Surprisingly, people were not as open source fanatical as I presupposed; everyone is sympathetic to the need to eat and understand you need to find the right model for sustainability. Many companies present contribute to open source software projects as part of work for clients, or within research. I am left with a lot to think about regarding future directions.

There’s also great interest in making these tools accessible to non-experts. Google Maps, and hacks, were mentioned many times at the conference, in a its changed all expectations sort of way (makes me wonder if mapserver will be a hot topic at Where 2.0). Though there’s still some conception of “barbarians at the gates” in the People’s SDI, there were plenty of examples of efforts that I think get it. HostGIS provides a hosted service for web based GIS, with powerful data integration and display options encapsulated in useable form. They’re acquiring openly available geodata from many sources, and making available for remix by their clients (like a heavy duty mapufacture). GeoVista Studio is an interesting “codeless” programming environment, using wiremodels to construct GIS data and functional vlows. A mapserver based US Census data explorer featured many powerful features, as with a similar project in the UK.

One block to widespread geodata creation and sharing, has been the overly complex OGC standards. Even there, almost everyone rolled their eyes at WML. To get real adoption among non-experts, standards need to be at an understandable level. This is geocoded RSS and similar simple standards. There looks to be interest in making simple geoformats interoperable with more complex OGC standards. This could be just the thing to get the ball rolling. Jo is collecting prior art here. Microformats are gathering inertia, and there are many similarities with geotags .. an evolutionary, useable today format. I’m surprised a Geo microformat hasn’t come out yet, will have to remedy that.

Now, what will happen at Where?