Skip to content

extracting oceanographic data

2 messages · LOUZAO, Michael Sumner

Original
LOUZAO
# 
Dear list,

I'm trying to extract different oceanographic data (chlorophyll, sea 
surface temperature, etc.) for doing habitat modelling. The input data 
are locations (lat, lon) of tracking data of marine top predators and I 
would like to extract oceanographic data from a given area around each 
specific location, from which compute  the median, maximum and minimum 
values of the corresponding data (sst, chl, etc.). For doing this,

1. I need to import the oceanographic data into R. Which format (ascii, 
hdf, netcdf,etc.) would you recommend? I have been working with ascii, 
importing files with import.asc().
2. For defining the area around each location, I could define a circle 
(radius of 1? in decimal degree) or a square (a total heigth of 1? in 
decimal degree) centered in each location.
In the grid package, the circleGrob() or the viewport() fucntions 
provide this approach. However, I have applied the circleGrob but I do 
not reach to obtain a circle around each location (when I plot it I only 
obtain the same configuration than when plotting the locations). I think 
that I'm missing something related with the units. In this case, x and y 
are longitud and latitud in decimal degress.

circle<-circleGrob(x=locs$Long, y=locs$Lat, r=2, default.units="native", 
name=NULL,
            gp=gpar(), vp=NULL)

I have found other two functions that I do not know if someone has used 
for doing the same.

- symbolsInPolys() in maptools

- spsample() in sp

3. The last step would be to overlay the oceanographic layer and the 
object of circles (or squares) to extract the median, maximum and 
minimum values of the corresponding data (sst, chl, etc.)

Has anybody worked on this problem? Could you please point to any useful 
documents? 

Thanks in advance,

Maite Louzao
Michael Sumner
# 
Hello,

I've worked a lot on problems of this type - here's some quick thoughts, 
not really directly addressing your questions.

HDF and netCDF can be difficult - you need to compile rgdal from against 
FWTools, or from source to have those work with readGDAL. I would just 
use gdal_translate to convert data to GeoTIFF (installing FWTools is 
trivial). Those formats can offer further challenges, just in 
interpreting the structure and content, but it depends on the actual 
data you want to use.

A question to ask here is what is the service providing the data in the 
first place? Ideally you can read directly from the server, but more 
likely you have a local cache of files. (If you have these as netCDF/HDF 
it's probably best to extract what you need to GeoTIFF using 
gdal_translate. )  The kftools package provides a direct read for SST, 
but it's a bit complicated. I had success with reading directly from the 
service behind the NASA POET site, but that wasn't useful for many 
datasets. I haven't explored the variety of services for some time, 
apparently there is a NOAA service that covers most needs.

For getting a range of pixel values around each point I would use 
overlay() in sp package, and work on creating circle polygons around 
your locations.

You didn't mention the temporal aspect that you need when overlaying 
track data with time series of oceanographic data, but this can be very 
simple when using rgdal since all your time series layers can be stored 
in one SpatialGridDataFrame dataset as separate attributes (depending on 
the size) and you simply find the interval in which your locations fall 
as well as the pixel for each location. Determining a spatial range 
around these is more complicated but not greatly so. The power of 
general indexing in R, along with the spatial extensions provided in sp 
is very helpful here.

How are you determining an appropriate range to query around your track 
locations? The tripEstimation package is a suite of functions written 
for estimating position from raw archival tags or a remote service, and 
I mention that as the outputs there can be very flexible for 
spatial/temporal overlays when location precision is important (but may 
be overkill here, and it's not user friendly).

The trip package extends the foundation provided by sp for track data, 
so that also may be useful for dealing with sets of tracks.

I hope some of that is helpful - sorry I'm not addressing your questions 
specifically. If you post more detail about the actual datasets you are 
trying to deal with I may be able to help more.

Cheers, Mike.
LOUZAO wrote: