...
Section | |||||||
---|---|---|---|---|---|---|---|
|
Table of Contents
Table of Contents | ||
---|---|---|
|
Input definition
|
Table of Contents
Table of Contents | ||||
---|---|---|---|---|
|
Input definition
...
Input data may be specified either by giving a path in the Input data may be specified either by giving a path in the Source parameter or by giving a GRIB-based data object in the Data parameter. Note that you should specify either Source or Data, not both.
...
- Grid: supply a valid string or list of numbers in the Grid parameter (see below)
- Lambert Conformal or Lambert Azimuthal Equal Area: supply details of the output grid in the set of Lambert grid definition parameters
- Template: supply a template GRIB file, whose grid structure will be used to generate the output GRIB;
- Note 1: use either Template Source or Template Data to specify the template (not both)
- Note
- 2: only GRIB fields on regular lat/lon or regular/reduced Gaussian grids may currently be used as templates
- Lambert Conformal: supply details of Filter: in this mode, the output grid in the set of Lambert grid definition parameters
- Lambert Azimuthal Equal Area: supply details of the output grid in the set of Lambert grid definition parameters
- Filter: in this mode, the output grid will be the same as the input grid, with k-nearest neighbours interpolation supporting the filtering
Grid Definition Mode: Grid
Grid
- will be the same as the input grid, with interpolation acting as a filter
- Note 1: this is equivalent to Template, with the input serving as the template (same input format limitations apply)
Note 2: Interpolation as k-nearest neighbours is the only method able to support filtering
Grid
Supply a grid definition as described here: grid - keyword in MARS/Dissemination request.
...
This parameter can be left empty to preserve the grid properties (regular/reduced lat/lon or Gaussian) while performing other kinds of post-processing (changing bits per value, calculation of gradients, etc.).
Template Source
If Grid Definition Mode
...
This mode configures regridding to output the same grid as the provided template, using one of the following parameters:
...
is Template, set path to a GRIB file to be used as template.
Template Data
...
If Grid Definition Mode
...
is Template, set a GRIB-based data object to be used as template.
Lambert Conformal or Lambert Azimuthal Equal Area parameters
These projections require setting several parameters,
...
These projections require setting several parameters, named following the convention in their descriptions:
...
| |||||||
|
Grid Definition Mode: Filter
This mode is similar to Template (see above), except the template is set to the input grid (the resulting output grid is the same as the input grid); In addition, interpolation k-nearest neighbours acts as a filter to the input values.
Wind processing
Wind processing
...
Wind Wind Processing
Activates processing that is particular to wind fields. Winds are represented by its vector Cartesian components u/v (gridded) or U/V (spectral) and, typically, they are archived as (spectral) vorticity/divergence (vo/d.) The relation between the spectral and gridded wind components is u = U / cos(latitude) and v = V / cos(latitude).
It is up to the user to specify if the input consists of wind fields. Set this appropriatelly in order to perform the correct processing.
Possible options are:
|
...
Converts pairs of Cartesian components vector fields (spectral) U/V to (gridded) u/v. This option is required if regridding wind fields on/to a rotated grid. Note: assumes that the input come in pairs of alternating U/V. | |
|
...
Converts pairs of (spectral) vo/d |
...
fields into (spectral) U/V or (gridded) u/v |
...
Wind Processing: U/V to u/v
...
. In case of gridded output, scaling by the cosine of their latitudes is applied (as above.) Note: assumes that the input come in pairs of alternating |
...
vo/ |
...
d. | |
| Each processed field is treated individually. |
Spectral to grid inverse transform
...
If the input files are spectral, the following parameters are used to fine-tune the conversion to grid points. The general workflow is:
- spectral data (input)
- if Truncation is not None (default Automatic), spectral data is truncated (intermediate spectral field, controlled by Truncation)
- if Intgrid is not None,
- inverse transform produces an (intermediate gridded field, controlled by Intgrid),
- interpolation to (final) grid
- if Intgrid is None, inverse transform produces the (final) grid
Note: if the intended (final) grid is rotated, or a given projection (eg. Lambert Conformal, LAEA, etc.), is very expensive computationally - final gridded data (see Grid Definition Mode)
Truncation
Spherical harmonics truncation, as described here:
Wind Processing: vo/d to u/v
Converts pairs of vorticity and divergence fields into U/V (spectral) or u/v (gridded.) Assumes that the input come in pairs of alternating vo/d. The resulting values will be scaled by the cosine of their latitudes (as mentioned above).
Wind Processing: Off
No special processing is performed (default). It is assumed there are no wind fields in the input, each processed field is treated individually.
Spectral to grid inverse transform
If the input files are spectral, the following parameters are used to fine-tune the conversion to grid points.
Truncation
Spherical harmonics truncation, as described here: truncation - keyword in MARS/Dissemination request.
...
- Finite Element-based interpolation with linear base functions
- Linear: FEM with supporting triangular mesh
- Bilinear: FEM with supporting quadrilateral mesh (for reduced grids, possibly containing triangles instead of highly-distorted quadrilaterals)
- Grid box method (based on Model grid box and time step)
- Grid Box Average: input/output grid box intersections interpolation preserving input value integrals (conservative interpolation)
- Grid Box MaximumStatistics: input/output grid box intersections value maximumstatistics - see parameter Interpolation Statistics for possible computations
- K-nearest neighbours based:
- K-Nearest Neighbours: general method combining nearest method (choice of neighbours) and distance weighting (choice of interpolating neighbour values)
- Nearest Neighbour: parametrised version of K-Nearest Neighbours to chose a nearest neighbouring input point to define output point value
- Nearest LSM: interpolated output point takes input only from input points of the same type (land or sea — requires setting land/sea masks)
- Structured methods, exploiting grid structure and configurable stencil for fast interpolations (non cacheable, so do not benefit from speedups on subsequent runs)
- Structured Bilinear: bilinear interpolation
- Structured Bicubic: bicubic interpolation
- Structured Biquasicubic: computationally economic bicubic interpolation
- Automatic: see above.
...
Available for any of the 'nearest' interpolation methods; Supports Interpolation K-Nearest Neighbours or Nearest LSM. Possible values are:
| input points with radius (option Distance) of output point |
| n-closest input points (option Nclosest) to output point (default 4) |
| input points respecting Distance ∩ Nclosest |
| input points respecting Distance U Nclosest |
| n-closest input points (option Nclosest), if all are at the same distance (within option Distance Tolerance) return all points within that distance (robust interpolation of pole values) |
| nearest input point, if at the same distance to other points (option Nclosest) chosen by lowest index |
| Sample of n-closest points (option Nclosest) out of input points with radius (option Distance) of output point, not sorted by distance |
| as above, sorted by distance |
Associated options supporting Nearest Method Grid Box Statistics (described above):
- Distance: in [m] choice of closest points by disntace to input pointNclosest: choice of n-closest input points distance to input point
- Distance Tolerance: tolerance in [m] tolerance checking the farthes farthest from nearest points (when Nearest Method is Nclosest or nearest)
Distance Weighting
Only available if Interpolation is K Nearest Neighbours. General way on how to interpolate input neighbouring point values to output points, including the Inverse Distance Weighting (IDW) class methods (see Wikipedia), which operates over input points returned by Nearest Method. Possible values are:
...
- Climate Filter
...
- Inverse Distance Weighting
...
- Inverse Distance Weighting Squared
...
- Shepard
...
- Gaussian
...
- Nearest Neighbour
...
- No
...
On multiple input points, weights are normalised linearly to unity. Associated options supporting Distance Weighting (described above):
- Distance Weighting Shepard Power: specify Shepard's method power parameter (see Wikipedia specific section)
- Distance Weighting Gaussian Stddev: specify Gaussian standard deviation (see Wikipedia)
Non Linear
This treatment is applied after calculation of the interpolation weights, and before they are applied to input values to generate output values. This allows modifications of these weights based on input data, such as the presence of missing values — In any case, no missing values are ever used for interpolation.
Most of the options avaiable concern modyfing the set of input points weights pertaining to a specific output point. When removing interpolation weights (pe. because they point to a missing value) all the remaining interpolation weights are re-normalised (linearly) to sum(wi) = 1.
Possible values are:
...
- Missing If All Missing
...
- Missing If Any Missing
...
- Missing If Heaviest Missing (default)
...
- Simulated Missing Value*
...
- Heaviest
...
- No*
...
- Nclosest: choice of n-closest input points to input point
Interpolation Statistics
Associated options supporting Nearest Method (described above). Possible values are:
- count
- count_above_upper_limit
- count_below_lower_limit
- maximum
- minimum
- mode_real
- mode_integral
- mode_boolean
- median_integral
- median_boolean
- mean
- variance
- skewness
- kurtosis
- stddev
- automatic
Distance Weighting
Only available if Interpolation is K Nearest Neighbours. General way on how to interpolate input neighbouring point values to output points, including the Inverse Distance Weighting (IDW) class methods (see Wikipedia), which operates over input points returned by Nearest Method. Possible values are:
| filter for processing topographic data (see IFS documentation, Part IV: Physical Processes, 11.3.1 Smoothing operator) |
| IDW of the form distance-1; If input points match output point, only that point's value is used for output |
| IDW of the form (1 + distance2)-1 |
| IDW of the form distance-p (option Distance Weighting Shepard Power, default 2.) |
| IDW of the form exp(- distance2 / (2 σ)2) (option Distance Weighting Gaussian Stddev, default 1.) |
| emulate Interpolation as Nearest Neighbour by picking first point (note that, when Nearest Method is Sample, a random near point is picked) |
| no distance weighting, average input values (irrespective of distance) |
On multiple input points, weights are normalised linearly to unity. Associated options supporting Distance Weighting (described above):
- Distance Weighting Shepard Power: specify Shepard's method power parameter (see Wikipedia specific section)
- Distance Weighting Gaussian Stddev: specify Gaussian standard deviation (see Wikipedia)
Non Linear
This treatment is applied after calculation of the interpolation weights, and before they are applied to input values to generate output values. This allows modifications of these weights based on input data, such as the presence of missing values — In any case, no missing values are ever used for interpolation.
Most of the options avaiable concern modyfing the set of input points weights pertaining to a specific output point. When removing interpolation weights (pe. because they point to a missing value) all the remaining interpolation weights are re-normalised (linearly) to sum(wi) = 1.
Possible values are:
| if all input point values (contributing to an output point) are missing, set output value to missing (it requires all input point values to be missing) |
| if any input point values (contributing to an output point) are missing, set output value to missing (it suffices one input missing value) |
| if the most significant point for interpolation (largest interpolation weight) is missing, set output value to missing (typically, not generally, this corresponds to the nearest input point) |
| allows a user-specified value (option Simulated Missing Value) with a tolerance (option Simulated Missing Value Epsilon) |
| emulate Interpolation as Nearest Neighbour by selecting the most significant point for interpolation to each output point (discarding the other contributions) |
| no non-linear corrections are applied |
* In the presence of missing values this can can create wrong results.
Associated options supporting Non Linear (described above):
- Simulated Missing Value: if Non Linear is Simulated Missing Value, set which value should not be used for interpolation irrespective of how data is described
- Simulated Missing Value Epsilon: if Non Linear is Simulated Missing Value, set tolerance when checking for value not be used for interpolation
Land-sea mask parameters
...
Land-sea masks (LSMs) can be configured for two different purposes:
- when Interpolation is Nearest LSM: use input/output LSMs to distinguish between points used for interpolation; If output point is land, only input points on land are used for interpolation (respectivelly for sea points)
- when Interpolation is not Nearest LSM and LSM is On (as described here: lsm - keyword in MARS/Dissemination request): interpolation weights for input points are rebalanced (by factor LSM Weight Adjustment) if output point is of different type to input points; Results depend strongly on interpolation method (in addition to the LSM parameters)
Distance Weighting With LSM
Only available if Interpolation is Nearest LSM. Possible values are:
- Nearest LSM: chose the closest input point (no disambiguation if there is more than one closest point at the same distance)
- Nearest LSM With Lowest Index: cross-platform compatible version (of the above Nearest LSM) with disambiguation of closest input points at the same distance of output points
- Off: use internal defaults (currently set to Nearest LSM With Lowest Index)
LSM Weight Adjustment
Only available if LSM is On, this is the factor adjusting input point weights if they are not of the same type (land/sea) as related output point; On application, all contributing input point weights are re-normalised (linearly) to sum(wi) = 1.
LSM Selection Input/Output
Specifies whether the input/output LSM file will come from LSM Named Input/Output (named, default) or LSM File Input/Output (file).
LSM Named Input/Output
Select one of the predefined names from the following:
| binary-based LSM sourced from MODIS Land Water Mask MOD44W (see reference) |
| binary-based LSM at high resolution (legacy, pre-climate files version 15) |
| GRIB-based IFS supporting climate files version 15, on this specific grid |
| (as above, for this grid) |
| (as above, for this grid) |
| (as above, for this grid) |
| (as above, for this grid) |
| (as above, for this grid) |
LSM File Input/Output
Provide the path to an input/output LSM GRIB file.
LSM Interpolation Input/Output
If input/output is not on the same grid (geometry) as provided input/output LSM (respectively), interpolate with this method to a temporary LSM with required geometry.
LSM Value Threshold Input/Output
For GRIB-based LSM (so excluding '1km' and '10min'), the threshold for condition (value ≥ threshold) to distinguish land (true) from sea (false).
Nabla differential operators
...
This options allows application of differential operators to input fields. The current available approach is similar as used in the Finite Volume Module of the IFS, specifically:
- A finite-volume module for simulating global all-scale atmospheric flows
- Atlas: A library for numerical weather prediction and climate modelling
It employs an edge-based, median-dual finite-volume method, with field values interpreted as averaged quantities of the supporting "dual cells".
There is support for both scalar and vector (u/v) fields; Due to the geometrical interpretation being ill-posed at the poles (singularities) there is an additional option to force missing values at the poles.
* In the presence of missing values this can can create wrong results.
Associated options supporting Non Linear (described above):
- Simulated Missing Value: if Non Linear is Simulated Missing Value, set which value should not be used for interpolation irrespective of how data is described
- Simulated Missing Value Epsilon: if Non Linear is Simulated Missing Value, set tolerance when checking for value not be used for interpolation
Land-sea mask parameters
Land-sea masks (LSMs) can be configured for two different purposes:
- when Interpolation is Nearest LSM: use input/output LSMs to distinguish between points used for interpolation; If output point is land, only input points on land are used for interpolation (respectivelly for sea points)
- when Interpolation is not Nearest LSM and LSM is On (as described here: lsm - keyword in MARS/Dissemination request): interpolation weights for input points are rebalanced (by factor LSM Weight Adjustment) if output point is of different type to input points; Results depend strongly on interpolation method (in addition to the LSM parameters)
Distance Weighting With LSM
Only available if Interpolation is Nearest LSM. Possible values are:
- Nearest LSM: chose the closest input point (no disambiguation if there is more than one closest point at the same distance)
- Nearest LSM With Lowest Index: cross-platform compatible version (of the above Nearest LSM) with disambiguation of closest input points at the same distance of output points
- Off: use internal defaults (currently set to Nearest LSM With Lowest Index)
LSM Weight Adjustment
Only available if Distance Weighting With Lsm is not Off.
LSM Selection Input/Output
Specifies whether the input/output LSM file will come from LSM Named Input/Output (named, default) or LSM File Input/Output (file).
LSM Named Input/Output
Select one of the predefined names from the following list:
- 1km (default)
- 10min
- O1280
- O640
- O320
- N320
- N256
- N128
LSM File Input/Output
Provide the path to an input/output LSM GRIB file.
LSM Interpolation Input/Output
If input/output is not the same geometry (grid) as provided input/output LSM (respectively), interpolate with this method to a temporary LSM with required geometry.
LSM Value Threshold Input/Output
For GRIB-based LSM (this excludes '1km' and '10min'), the threshold for condition (value ≥ threshold) to distinguish land (true) from sea (false).
...
Nabla
Activates a nabla (differential) operator processing on the fields. Possible functions options are:
| Scalar |
...
field gradient (∇) | |
| Scalar field Laplacian (∇2) |
| Vector (u/v) field gradient (∇) |
|
...
Vector (u/v) field divergence (∇⋅) | |
| Vector (u/v) field vorticity or curl (∇×) |
| no differential processing |
Nabla Poles Missing Values
Due to the supporting differential operators calculation method, values aren't well defined at the poles (singularities); This option allows forcing missing value at the poles. Possible values are On and Off.
...