Version 56 (modified by gutierjm, 9 years ago) (diff)



Different sector-specific impact activities to be undertaken in SPECS and EUPORIAS projects require a reduced number of variables (typically at surface) from different data sources (mainly seasonal forecasts, reanalysis, and observations). The SPECS-EUPORIAS Data Server has been established by the Santander Meteorology Group (UC-CSIC) as part of the data management activities in these projects to provide a unique access for these impact-relevant variables, gathered from existing datasets. The data portal is based on a THREDDS data server providing metadata and data access using OPeNDAP and other remote data access protocols. Moreover, a user-friendly R package has also been developed for exploring and remotely accessing subsets of data, thus reducing the burden of data access in these activities. This package will be also a key component for other tasks of the projects based on R, including the validation and downscaling packages to be developed within SPECS and sector-specific calibration and modeling tools to be developed in EUPORIAS.

This trac/wiki page provides an up-to-date description of the SPECS-EUPORIAS Data Server, including information of the available datasets and the documentation and code of the R data access package. This page is currently under construction, but both a first tutorial describing the basic functioning and a first version of the R package (a R function) are already available:

Dataset catalog:

R code: loadSystem4.R

Tutorial: PDF file

Contents (under development):

    Error: Page SpecsEuporias does not exist

  1. Data Server?
  2. R Package for Data Access?
  3. Other interfaces for Data Access?

Introduction and Motivation

The impact activities on seasonal timescales involved in SPECS and EUPORIAS projects require the use of different data sources (mainly seasonal forecasts, reanalysis, and observations). These activities include the calibration, downscaling, and modelling of sector-specific indices in agriculture, energy, health, etc., building on meteorological information. Typically, only a reduced subset of surface variables (precipitation, temperatures, mean sea level pressure, etc.) or in a reduced number of vertical levels (circulation and termodynamic drivers at, e.g., 850, 500, 200 hPa) is required for these activities. The SPECS-EUPORIAS Data Portal has been established by the Santander Meteorology Group (UC-CSIC) to gather the relevant information from existing datasets in order to provide a unique homogenized access to data for the SPECS and EUPORIAS partners (in particular for impact-users).

The SPECS-EUPORIAS Data Portal is based on a THREDDS Data Server (TDS) providing metadata and data access using OPeNDAP and other remote data access protocols. Moreover, since the R language ( has been adopted for some key tasks in these projects (including the development of comprehensive validation and statistical-downscaling packages) a user-friendly R package has been developed to explore and access the data portal. This package can be used in R programs to remotely access subsets of data, thus reducing the burden of data access (versions for Python and Matlab are also available under request). This package will be continuously updated (keep informed at the documentation URL above) as part of the data management activities to build a data bridge for impact users and for the R developments to be done in these projects.

This document briefly describes the current state of the data portal, which has initially focused on data from the ECMWF's System4 seasonal model, as agreed in the downscaling parallel session of the kick-off meeting.

The THREDDS Data Server

The SPECS-EUPORIAS Data Portal is based on a password-protected THREDDS data server (TDS) providing metadata and data access to a set of georeferenced atmospheric variables using OPeNDAP and other remote data access protocols. The variables names, units and additional metadata follow the CF convention. The variables are spatial grids based on multidimensional arrays of indexed values, following Unidata's _Coordinate convention12.

Typically the data portal will include information at a daily resolution, but monthly-aggregated values could be also provided in some cases due to data limitations (in particular, Mètèo-France and Met Office have agreeed to provide monthly mean hindcasts for their use by the SPECS and EUPORIAS partners). In general, the data available will be typical surface variables (e.g. precipitation and near-surface temperature), although several variables (e.g. geopotential and temperature) on pressure levels will also be stored for the statistical downscaling activities.

The data gathering activities have initially focused on the ECMWF System4 seasonal model. The Meteorological Archival and Retrieval System (MARS) is the main repository of meteorological data at the ECMWF (European Centre for Medium-Range Weather Forecasts). It contains terabytes of operational and research data as well as data from special projects3. The large amount of information stored and the inherent complexities of data access, download and post-processing is a first shortcoming for a flexible use of these datasets by a large number of partners. To overcome this issue, a reduced subset of surface variables4 (precipitation, temperatures and mean sea level pressure) have been downloaded from MARS (a colection of GRIB-1 files) at 0.75º spatial resolution and made available throught the SPECS-EUPORIAS data portal. The downloaded data has been exposed as three different virtual datasets using TDS:

  • System4 seasonal range (15 members): There are twelve initializations (hereafter called runtimes) per year (the first of January, February, ...) running for 7 months (hereafter called simply times). An ensemble of 15 members is available for the whole 1981-2010 period.
  • System4 seasonal range (51 members): There are only four runtimes per year (the first of February, May, August and November) and the forecasts run for 7 months. An ensemble of 51 members is available for the whole 1981-2010 period.
  • System4 annual range (15 members): As in the previous case, there are four runtimes per year, but the forecasts run for 13 months. An ensemble of 15 members is available for the whole 1981-2010 period.

Data gathering activities will next move to the CFS ( version 2 hindcast, developed at the Environmental Modeling Center at NCEP and also to reanalysis and observational datasets.

Accesing the Data Portal via R

Accesing the Data Portal via Web

The SPECS-EUPORIAS Data Portal can be accessed through the Data Portal URL provided in the abstract. First of all, an authentication dialog will request a valid user name and password.

No image "loginTHREDDS.png" attached to udg/ecoms

Afterwards, the different datasets described in TDS section are listed as links in the web browser window.

No image "fig01.png" attached to udg/ecoms

By clicking in any of the datasets, a new window will appear providing information on the variables and geospatial and time coverages, and offering different options for data access and/or visualization.

No image "fig02.png" attached to udg/ecoms

Currently, only the OPeNDAP access service is fully operative in the portal. Therefore, in this example, we will illustrate the use of this service, which allows selecting time/spatial data slices from the OPeNDAP data access form shown in figure and downloading the resulting data in both ASCII and Binary formats.

No image "openDAPwindow.png" attached to udg/ecoms

Note that, as explained before, the variables provided by the data portal (e.g. minimum temperature) are stored as gridsets. Thus, in addition to these variables, also auxiliary coordinate variables (lat, lon, run, time, member) should be handled for geo-temporal data referencing (see Figure). Moreover, three time coordinates are included as referece for different grid variables because they are defined for different forecast times (one extra time for precipitation and different temporal resolution for mean sea level pressure). Note that this highly complicates the direct analysis of the data and, hence, this options is only recommend for data exploration. In the following we show how to use this service to explore the structure of the datasets and to obtain simple pieces of information in ASCII format.

By default, if no specifications are given in the different subsetting boxes of the OpenDAP form, the whole data on the whole spatio/temporal and member ranges of the dataset would be accessed. However, this option will raise an error due to the large size of the request (the maximum size of a single request has been set to 100 Mbytes in the SPECS-EUPORIAS data portal for the sake of multi-connection efficiency). The basic steps to retrieve subsets of data are the following:

  1. To select a variable click on the checkbox to its left.
  2. To constrain the variable, edit the information that appears in the text boxes below the variable. This is a vector of integers indicating index positions of length three, with the following order: [start:stride:end].
  3. To get ASCII or binary values for the selected variables, click on the Get ASCII or Get Binary buttons of the Action field. Note that the URL displayed in the Data URL field is updated as you select and/or constrain variables. The URL in this field can be cut and pasted in various OPeNDAP clients.

The main disadvantage of the OPeNDAP service from the end-user point of view is that the specifications for subsetting dimensions are not given in their original magnitudes (i.e., latitudes and longitudes are not given in decimal degrees), but by the indexes of their position along their respective axes (note that first index value is always 0). Thus, to find out the indexes for the desired selection, we need to dump and analyze the particular values defined in the coordinate variable. For instance, this figure shows the 241 values defined for the lat (latitude) coordinate, as provided by the Get ASCII option (selecting the corresponding check-box).

No image "latlonDump.png" attached to udg/ecoms

Using these facilities it can be obtained after some calculations that the closest lat and lon coordinates for a particular location of interest (e.g. Madrid) are 66 and 475, respectively. Thus, the time series for Madrid corresponding to the example described in the previous section (minimum temperature forecasts for January with one-month lead time, i.e. from the simulations started the first of December) could be requested as shown in Figure

No image "opendapquery.png" attached to udg/ecoms

Note that the indices selected for the run coordinate correspond to the December initilizations (index positions 11, 23,...; note that indexes start in 0) and for the time coordinate correspond to January (positions, 31 to 62, in days after the run time). Note that the proper use of this service requires a full understanding of the data structure and, therefore, it is only advised for data exploration.

Accessing to the Data portal using Python (Pydap version)

This section needs revision

[user@host ~]$ pip install Pydap
[user@host ~]$ python
Python 2.7.2 (default, Mar  3 2012, 10:45:44)
[GCC 4.1.2 20080704 (Red Hat 4.1.2-48)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from pydap.client import open_url
>>> dataset = open_url('')
>>> print type(dataset)
<class 'pydap.model.DatasetType'>
>>> print dataset.keys()
['lat', 'lon', 'run', 'time', 'time1', 'time2', 'member', 'Maximum_temperature_at_2_metres_since_last_24_hours_surface', 'Minimum_temperature_at_2_metres_since_last_24_hours_surface', 'Mean_temperature_at_2_metres_since_last_24_hours_surface', 'Total_precipitation_surface', 'Mean_sea_level_pressure_surface']
>>> MN2T24 = dataset['Minimum_temperature_at_2_metres_since_last_24_hours_surface']
>>> print MN2T24.dimensions
('member', 'run', 'time', 'lat', 'lon')
>>> print MN2T24.shape
(15, 360, 215, 241, 480)
>>> arr = MN2T24[0,11:360:12,31:62,66,475]
>>> print numpy.squeeze(numpy.mean(arr,2))
[ 270.79171753  273.29437256  271.56661987  271.03707886  271.82745361
  272.49279785  271.48086548  268.59121704  271.53125     273.82156372
  270.99401855  274.23626709  270.99328613  271.56115723  273.98986816
  270.50756836  272.45046997  270.65560913  271.31182861  272.77200317
  273.4359436   271.85021973  273.39648438  274.16384888  269.98248291
  271.30166626  273.11950684  271.27301025  272.29147339  270.46688843]

Accessing to the Data portal using Octave

This section needs revision and integrtion with the Matlab example section

>> ver
GNU Octave Version 3.6.1
GNU Octave License: GNU General Public License
Operating System: unknown
>> urlwrite('','netcdfAll-4.3.jar')
>> javaaddpath('./netcdfAll-4.3.jar');
>> javaMethod('setGlobalCredentialsProvider','',javaObject('','username','password'));
>> ncfile = javaMethod('openDataset','ucar.nc2.dataset.NetcdfDataset','');
>> v = ncfile.findVariable('Minimum_temperature_at_2_metres_since_last_24_hours_surface');
>> disp(v.getDimensions.toString)
[   member = 15;,    run = 360;,    time = 215;,    lat = 241;,    lon = 480;]
>> d ='0,11:359:12,31:61,66,475');
>> tmp = javaObject('org.octave.Matrix',d.reduce.copyToNDJavaArray);
>> oldFlag = java_convert_matrix (1);
>> octaveMatrix = tmp.ident(tmp);
[ (30 by 31) array of double ]
>> disp(squeeze(mean(octaveMatrix,2))')
 Columns 1 through 13:

   270.79   273.29   271.57   271.04   271.83   272.49   271.48   268.59   271.53   273.82   270.99   274.24   270.99

 Columns 14 through 26:

   271.56   273.99   270.51   272.45   270.66   271.31   272.77   273.44   271.85   273.40   274.16   269.98   271.30

 Columns 27 through 30:

   273.12   271.27   272.29   270.47

Accessing to the Data portal using Matlab

This section needs revision]

>> ver
MATLAB Version (R2009a)
MATLAB License Number: 161051
Operating System: Microsoft Windows Vista Version 6.1 (Build 7601: Service Pack 1)
Java VM Version: Java 1.6.0_04-b12 with Sun Microsystems Inc. Java HotSpot(TM) 64-Bit Server VM mixed mode
>> javaaddpath('');
>> %javaaddpath('');
>> import* %this will download the netcdfAll-4.3.jar
>> HTTPSession.setGlobalCredentialsProvider(HTTPBasicProvider('username','password'));
>> import ucar.nc2.*;
>> import ucar.nc2.dataset.*;
>> ncfile = NetcdfDataset.openDataset('');
>> v = ncfile.findVariable('Minimum_temperature_at_2_metres_since_last_24_hours_surface');
>> disp(v.getDimensions)
[   member = 15;,    run = 360;,    time = 215;,    lat = 241;,    lon = 480;]
>> data ='0,11:359:12,31:61,66,475').copyToNDJavaArray();
>> disp(squeeze(mean(data,3)))
  Columns 1 through 13

  270.7917  273.2944  271.5666  271.0371  271.8275  272.4928  271.4809  268.5912  271.5313  273.8216  270.9940  274.2363  270.9933

  Columns 14 through 26

  271.5612  273.9899  270.5076  272.4505  270.6556  271.3118  272.7720  273.4359  271.8502  273.3965  274.1638  269.9825  271.3017

  Columns 27 through 30

  273.1195  271.2730  272.2915  270.4669

Example of Data Analysis with R


Attachments (18)