I'm looking for papers investigating the sensitivity of automatic variogram
fitting functions to outliers. I'm aware of Genton's work
on robust variograms but still look for papers.
Of course I would welcome a discussion about such topic.
Unless I'm completely out of date, most recent developments in automatic
mapping systems (useful in emergency situations) have involved machine
learning algorithms (e.g. Support Vector Machine) that still can not compete
with functions involving some kind of spatial correlation analysis. Moreover,
the tuning of machine learning algorithms remains tricky. Consequently, it
seems to be that performing a variogram analysis and modelling are still
required. Therefore, I'm wondering how much an automatic fitting of the
semivariogram can be misleading when outliers (e.g. wrong measurements) are
present in the data. The case I would like to see discussed should not involve
any prior knowledge of the monitored phenomenon.
Of course other mapping methods do exist but I don't see any other function
than can compete with one based on geostatistics: these functions are robust,
known to provide good results, can be adapted in order to deal with many
different problems (IRFk, multivariate, Bayesian...), can provide some
probabilistic information to be above specific thresholds (risk maps), provide
some information about the reliability of the estimates (kriging variance in
the case of homoscedasticity) and even help decision makers to improve the
sampling strategy (e.g. by means of simulations). In summary, one would have
many good reasons to implement geostatistical functions in an automatic
mapping system if there was not a risk to derive all results from a
badly/wrongly fitted semivariogram... How high is such risk is my (main)
Thank you very much for any input.
* To post a message to the list, send it to ai-geostats@...
* As a general service to the users, please remember to post a summary of any useful responses to your questions.
* To unsubscribe, send an email to majordomo@...
with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list
* Support to the list is provided at http://www.ai-geostats.org