home

The discussion page

Contents

The grid definition

The Grid Definition

At and following the first ELDAS progress meeting there was a discussion about the grids at which forcings are prepared and at which soil moisture is interpolated. Following this discussion I have posted the matter to the Advisory Board (see the mail message) For the LAM systems, there are the following options, REG denoting the regular 0.2deg lat/lon grid, ROT a common rotated grid).
  • 1. bring forcing into the database as REG, provide common interpolation routine to go to ROT, produce soil mosture at ROT, interpolate soil moisture to REG, store REG
  • 2. bring forcing into the database as ROT, produce soil moisture at ROT, interpolate soil moisture to REG, store REG
  • 3. same as 1. but store ROT soil moisture without interpolation
  • 4. same as 2. but store ROT soil moisture without interpolation
For the SC system at ECMWF (operating on a REG grid) this corresponds to
  • 1. use REG forcings, store REG soil moisture
  • 2. interpolate ROT forcings to REG, store REG soil moisture
  • 3. same as 1., but rotate REG soil moisture to ROT
  • 4. same as 2., but rotate REG soil moisture to ROT
If we want to store soil moisture on a common grid for all systems (which we definitely want) either the LAM or the SC system have to do interpolation of soil moisture. Given the comments I got back from the advisory board (see replies) this needs to be done with caution and depending on the reason to have this common soil moisture grid. If it deals with a model comparison of water budgets, Gianpaolo's suggestion to conserve water should be followed. If the comparison deals with effects of soil moisture on evaporation and runoff, a weighing with the evaporation or runoff response function should be included in the interpolation. This multiple interpolation demand requests that in any case the soil moisture fields should be stored both at the original grid at which they are produced, and at a common REG grid. For the SC system at ECMWF this is the same, for the DWD and CNRM system this implies two storage archives.

Given the  comments of the advisory board I would be in favor of storing the databases of the forcings at least in the original REG grid. This ensures compatibility with global datasets and easier use by third parties, and allows a direct link between the stored soil moisture and stored forcings. We can also store the forcings on a ROT grid, but it is probably better (and room saving) to produce a common interpolation procedure to scale these observations to the LAM grids. A distinction should be made between interpolation of precipitation (conserving water) and other meteorological fields (bilinear interpolation). I don't mind making this general interpolation utility available.

This implies the implementation of option 1.

An interpolation software package has been developed and can be found here.

Mail to the User Advisory Board (6 January 2003)

At the progress meeting one issue emerged about which I would like to have your opinion. As you know, a demonstration forcing database of precipitation, radiation and atmospheric properties will be constructed covering the European area for the year 2000, and from this database a number of soil moisture assimilation systems will be operated to produce soil moisture estimates. The data assimilation systems participating in ELDAS are different, not only with respect to the land surface parameterization schemes embedded, but also with respect to the general model infrastructure used. Some schemes will carry out the data assimilation using Limited Area Models (LAM's) on a rotated lat-lon grid, another group (ECMWF) will run single column models located at the grid locations specified by the ELDAS domain.

At the kickoff meeting in Januari 2002, we defined a common ELDAS grid spaced at a 0.2 degree regular lat-lon separation. Unlike this regular 0.2 degree lat-lon grid, the rotated lat-lon grids operated by the LAM's have a grid-point spacing that is fairly equal across the domain in terms of physical distance, but irregular in terms of angular lat-lon coordinates, in particular at high latitudes. The LAM's are unable to run on the regular 0.2 degree grid owing to computer cost limitations. They therefore would have to interpolate the ELDAS forcings to their rotated grid, and interpolate back their resulting soil moisture fields to the ELDAS regular 0.2 degree grid. Groups operating these LAM's now proposed to redefine the common ELDAS grid in terms of a common rotated grid, and prepare the forcings and soil moisture data sets on this common grid for all ELDAS products. In the discussion following this
proposal a number of pro's and con's to this proposal emerged:

PRO's

  • The LAM's need no further interpolation, either from forcing observations to the model domain, or from the model soil moisture product to a general ELDAS domain
  • Interpolation software would have to be standard for all groups, requiring additional coding
  • a regular grid in terms of physical length coordinates is the only way a consistent spatial detail across the domain can be obtained
  • a common grid is defined for all ELDAS products, both the demonstration forcing fields and the soil moisture products
  • a rotated grid requires less grid points to cover an identical area, thereby saving computer cost and storage capacity.
CON's
  • a first version of the precipitation database has already been prepared and needs to be reprocessed
  • the larger grid spacing in high latitudes may cause throwing away observation detail in the precipitation data (although this is not considerable given the approximately 20km grid distance in the rotated grid)
  • a common regular lat-lon grid is easier to understand and use by possible external users of the ELDAS data
  • a common regular lat-lon grid corresponds to the grid definitions of for instance GLDAS, ISLSCP/GSWP and the World Hydrological Organisation
  • a rotated grid requires adaptations to the software with which the data is archived in the ECMWF MARS archive
  • the group running the single-column setup relies on atmospheric fields which are only available at a regular lat-lon grid (produced by the global ECMWF model), and therefore will generate soil moisture fields at a regular lat-lon grid anyway. Therefore, a straightforward intercomparison between ELDAS data assimilation systems will always require interpolation of soil moisture fields, either from the LAM's to a regular lat-lon grid, or from the single-column system to a rotated grid.
We feel that in particular the arguments related to compliance to externally defined grids and easy access by external non-ELDAS users should play a role in deciding what grid formulation we should choose to store the ELDAS forcings and soil moisture products. Your opinion about this rather pragmatic matter would be highly appreciated. Please don't hesitate to contact me again in case you have additional questions. Partly based upon your opinions we will make a grid decision very soon. 
 

Replies

I am hardly the  expert on this one, but for both compatibility with ECMWF and the global community, i would recommend you maintain archive fields/products on a lat-long grid and let the LAM's interpolate to their specific requirement (Alan K. Betts)

I am not an expert but it looks that a lat/long grid has more advantages than a rotated lat/long grid as required by LAM community (and likely only this community?) (Michael Berger, ESA)

In general I would say that the user should interpolate the atmospheric forcing to his grid. The atmospheric forcing are fluxes or some quantities with very large correlation radii. Thus there is only limited loss of information or perturbation of the data. When it comes to interpolation of soil moisture I need to ask what for ? The LAM's should spin up their soil moisture on their grid using the common forcing. In any way there is little point of using the soil moisture of another land-surface scheme as the quantities are not comparable. A lesson we leaned the hard way in PILPS, see the papers by Koster and Milly. I would like to very strongly discourage the interpolation of soil moisture. It can not give any reasonable results. The correlation radius is too small to assume that the values in two points at the 0.2 deg distance have any information on the points between. Just take a valley and some small mountains 0.2 deg away, what is the relation between the two values which would allow to build an interpolation scheme ? (Jan Polcher, LMD)

We use grids similar to your rotated lat/lon grids for our crop yield applications, but will probably have to apply an interpolation from whatever grid ELDAS will produce. We therefore have no strong preference for a rotated or a regular lat/lon grid (Allard de Wit, Alterra)

On the interpolation of soil moisture I would like to share a doubt : how we may do it properly when interpolating to 0.2° means upscaling from a higher resolution (specially for LAM)? I would tend to say that is right to interpolate bi-linearly the SWI, because it conserves the evapotranspiration impact (meteo point of view?), but it would probably not conserve the water (if we have pixels in LAM of different physiography, they will loose the precipitation input with rather different behaviours, probably different from the behaviour of the physiography of the 0.2°), so maybe better (in ELDAS point of view) to interpolate the water content? (Gianpaolo Balsamo, CNRM)