Who are the authors?

Under which license is it distributed?

Can I contribute to its development?

• J. Saenz, J. Zubillaga and J. Fernandez (2002) Geophysical data analysis using Python, Computers and Geosciences, 28/4:457-465

Are there other mirrors of this software?

Why did we write it?

Which specific problems of the climate analyst can PyClimate solve?

That is, for instance,

• the case during the analysis of the joint spatial and temporal variability of scalar or vector fields like geopotential height, precipitation or temperature over a wide area to identify the main modes of variability embedded in a temporally and spatially variable data set, which is often performed by means of what is called the Empirical Orthogonal Function (EOF) approach, based on the Karhunen-Lo�ve decomposition of the joint temporal and spatial variability of the fields.
• Similarly, during the analysis of coupled variability of geophysical data sets, eigenvalue techniques like Canonical Correlation Analysis (CCA) or the Singular Value Decomposition (SVD) of the covariance matrices are standard tools.
• The matricial nature of the data and the solutions make very simple and easy the use of Numeric Python and the LinearAlgebra routines to code the corresponding algorithms.

On the other hand, atmospheric and oceanic data sets are usually distributed using extremely different formats, which provide some very interesting features, but which make the access to individual records and hyperslabs difficult to handle.

Just to name a few instances,
• the use of netCDF files is very common (see, for instance, the NCEP/NCAR Reanalysis Homepage). The use of globally gridded data sets usually means that the data is distributed in regular latitude/longitude grids, which makes very interesting the use of the so-called COARDS compliant netCDF files.
• However, the use of global spectral models is also very common in meteorological applications. In this case, the data is usually stored in the World Meteorological Organization (WMO) endorsed GRIB format, which allows the storage of spectral tetradimensional grids (time, vertical level, latitude and longitude) to store/retrieve the data and allowing extra features not currently supported by the netCDF interface.
• Finally, just to mention a few other data formats, satellite data is often stored using the HDF interface, instrumental surface data is very often distributed using compressed ASCII files (CRU 05 data set, GOSTA-8...) or even binary packed data, like the GTOPO30 topography data set.

This means that, routinely, a researcher on atmospheric or oceanographic sciences has to have a way to access these data sets and convert them from one format to another to be able to perform a quantitative analysis on them. Thus, a flexible tool like Python with its Numeric extensions is very helpful.

• We can directly read binary data from a binary file using file objects and convert it into its Numeric representation with fromstring().
• We can call external programs like wgrib to dump the data records on a native representation and then access the records from the interpreter, storing them under a different format. Creating netCDF files from scratch using the Scientific.IO.NetCDF interface by Konrad Hinsen is much easier than using equivalent C or FORTRAN programs.
• Similarly, the array-oriented nature of the netCDF COARDS interface makes it very easy to access whole parts or some hyperslabs of data in a COARDS compliant netCDF file. Briefly stated, Python, Numeric Python and other modules/packages accesible from the Python interpreter make very easy to work with data under very different and heterogeneous formats.

Data used in the analysis of climate variability responds to different physical phenomena. This explains that temporal variations in the fields are usually formed by a set of different frequencies, they are broad-band signals. However, it is usually interesting to separate the different scales of "motion" in a geophysical field with the aim of separating different physical effects. Such is the case, for instance, when one is interested in analysing the high-frequency (with period in the 2-10 days range) transients of the extratropical atmospheric flow, attributed to the baroclinic instability of the flow from the so-called low frequency variability (LFV) of the extratropical atmospheric circulation, associated to monthly and higher time scales. Similarly, it is of primary interest to analyse the interchange of Rossby waves from midlatitudes and tropical latitudes in the 6-25 days time-scale, which allows one to separate the extratropical-only short-wavelength baroclinic systems from the extratropical motions and to atenuate the tropical-only motions associated with the Madden-Julian Oscillation and mixed Rossby-gravity waves, which are of primary importance in the tropical latitudes, but do not exist outside the tropics.

In the ocean, it is important to distinguish between internal gravity waves and Rossby waves on the basis of their periods. Thus, it is very important to be able to perform digital filtering of multivariate datasets, without considering in detail the structure of the field, which can be unidimensional (the time series at a single measurement site), two-dimensional (the time series of a zonal average), three-dimensional (the time-varying global sea-surface temperature) or four-dimensional (time varying geopotential height field at several vertical levels and grid points in a latitude/longitude grid).

The generic nature of some of Numeric Python's array oriented operations allow an easy coding of these operations for generically shaped arrays, achieving a good performance in the computations, which is a very important requirement for this task. It is to be taken into account that, for instance, the geopotential height 12 hourly data over the whole earth for a part of the Reanalysis period (1967-1998), on a regular 2.5 degree x 2.5 degree latitude/longitude grid with 6 vertical levels (1000, 850, 700, 500, 300 and 200 hPa surfaces) takes about 2.8Gb in a netCDF file, even after packing the floating point values into Int16 accuracy through an "offset plus scaling" approach.

Which platforms are supported?