[R-sig-Geo] Floods' Prediction Approach
jecogeo at gmail.com
Thu Feb 4 16:14:21 CET 2016
Hello everybody. I don't know exactly if this post could be considered an
off-topic, and sorry if so. I would like to exchange some ideas with you
all regarding a problem and, hopefully, someone is able to help.
I have a series of flood maps derived from remote sensing. Each map (13
maps) represent the flooded area in a specific water level height (with
binary values, 1=flooded / 0=not flooded) . I would like "predict" the
flooded areas in intermediate water level heights where we haven't maps,
thus trying to simulate the evolution of the flood inside the floodplain.
I used a pixel-by-pixel logistic regression approach, with the dependent
variable as the flood status (0 or 1) in each flood map and the independent
variable as the correspondent water stage height. It predicted quite well,
considering It's a very simple approach, giving the probabilities of a
given pixel to be flooded at all water stage values from the drought to the
maximum seasonal water level. But I'm getting significative errors yet
regarding my field observations (sensors monitoring the floods scattered in
my study area). I'm calculating the number of days per year some pixels
remains flooded (as I know the water levels on each day in a given year)
and I'm getting up to 120 days of difference between the predicted and
observed days of flood in a given pixel.
As I said before, It's a quite simple approach. I'm realizing that more
sofisticated techniques could perform better. I also have some other
variables to include, like a digital terrain model and rasters of daily
Has anyone some idea, some different approach (e.g. cokriging neural
networks) ? As I'm dealing with a categoric response variable I'm not sure
yet what is possible and what isn't. I'll continue my readings here and
trying to find some idea. But if someone has some time to think a little
bit and write a comment, it'll be very appreciated.
Thank you all!
Geógrafo – GEOPROCESSAMENTO IDSM | Coordenadoria de TI
Jefferson.ferreira at mamiraua.org.br
*Instituto de Desenvolvimento Sustentável Mamirauá*
Ministério da Ciência, Tecnologia e Inovação
Telefone: +55 97 3343-9710
*Google Maps* - Mapas deste e-mail:
Exibir mapa ampliado
[[alternative HTML version deleted]]
More information about the R-sig-Geo