Hi Martin, Sorry to briefly hijack Christian's topic but I just wanted to ask if you'd thought about combining a Bayesian approach with the single image reconstruction algorithm described in the Clint Sprott's paper ? One naive suggestion would be to use the Sprott method to construct something akin to a Bayesian prior which feeds in to a Gibbs sampling regression analysis using your other variables.
Michael 2008/11/8 Martin Desruisseaux <[EMAIL PROTECTED]>: > Hello Christian > > There is of course nice work that could be done, but it depends on which area > you would prefer to work. Referencing? Coverage? Geometry? > > In this email I will assume coverage based on your coverage-jdbc plugin, but I > could develop a bit about Referencing if it can be useful. However in order to > give more detailed suggestions, it would help if we had some idea about when > the > work would start (because the proposal may depends on ungoing work) and how > long > you can work on it. > > I would also like to know which kind of scientific theory you are looking for. > Is is computer science, mathematic or some application field (oceanography, > meteorology). > > Below is a proposal applicable to oceanography which would require a good > background in mathematic. If you choose those kind of proposal, we would be > glad > on our side to try to help you to achieve them. > > > > Proposal Number #1 > ------------------------------------------------------------------- > > In oceanography we have GridCoverage2D of different parameters > calculated from Remote Sensing data. Some of the most commons > parameters are: > > - Sea Surface Temperature (°C) > - Chlorophyl-a concentration (mg/m³) > - Sea Level Anomaly (cm) > > Unfortunatly some of those data may be missing because of weater > conditions. Sea Surface Temperature are not available if the sky > is cloudy, which is very common in tropical area. Sea Level Anomaly > can be available despite cloud cover, except if it is raining hard. > > In some cases we really need some estimation of a missing parameter > even if it is just a very approximative idea. If a Sea Surface > Temperature value is missing because of a cloud cover, we can still > get some idea using other parameters because they usually have a > strong correlation. For example cold water is often associated with > low value of Sea Level Anomaly, and conversely (hot water is often > associated with high value of Sea Level Anomaly). > > There is what we could do, most simplist approach first, more > elaborated approach later: > > 1) Compute the correlation between two arbitrary parameters > (in our example Sea Surface Temperature with Sea Level > Anomaly) using some historical data. Then when a Sea > Surface Temperature is missing, use the correlation for > computing an estimation of "probable" value using the > Sea Level Anomaly. > > 2) Above approach is very naive (real nature is much more > complex than the linear relationship assumed above). We > can still try the same idea, but replacing the linear > relationship by a neuronal network which has learn from > many parameters: Sea Level Anomaly, but also geographic > area, time of the year, wind speed, etc. > > 3) Above approach 2 is better than 1 but still not yet quite > satisfying. If give just one number (the temperature in our > example) while we would like to have some estimation of its > uncertanties. A value inferred in such indirect way from other > parameters is less "certain" than a direct measurement of Sea > Surface Temperature. Bayesian network may be a solution (but > I'm probably out of scope of a master thesis here). > > I used "Sea Surface Temperature" vs "Sea Level Anomaly" above as > a real-world example (with real applications on our side), but > such a project would actually be against any arbitrary set of > geophysics parameters. > > > > > Proposal Number #2 > ------------------------------------------------------------------- > > Same goals than above, but working on a single image without any attempt to > leverage the correlation between geophysics parameters: > > http://sprott.physics.wisc.edu/pubs/paper276.htm > > > Is it the kind of suggestions you were looking for? > > Martin > > ------------------------------------------------------------------------- > This SF.Net email is sponsored by the Moblin Your Move Developer's challenge > Build the coolest Linux based applications with Moblin SDK & win great prizes > Grand prize is a trip for two to an Open Source event anywhere in the world > http://moblin-contest.org/redirect.php?banner_id=100&url=/ > _______________________________________________ > Geotools-devel mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/geotools-devel > ------------------------------------------------------------------------- This SF.Net email is sponsored by the Moblin Your Move Developer's challenge Build the coolest Linux based applications with Moblin SDK & win great prizes Grand prize is a trip for two to an Open Source event anywhere in the world http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Geotools-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/geotools-devel
