| Home > Newsevents > Training > Rcourse_notes > PARAMETRIZATION > SURFACE_ASSIMILATION > |
Land surface assimilation
March 2001
By Jean-François Mahfouf and Pedro Viterbo
European Centre for Medium-Range Weather
Forecasts
Table of contents
Next Section
Previous Section
7 Initialisation of other land surface quantities
7.1 Snow mass
The methods used to analyse snow mass are relatively crude when compared to analysis schemes for the atmospheric variables (Lorenc 1986) and could easily be improved. In this section, the ECMWF snow analysis is described. The methodology used in other operational centres is very similar.
Every 6 hours a snow analysis is performed in three steps:
| 1) Snowfall analysis: From SYNOP reports of temperature and precipitation rate, a snowfall rate is estimated at the observation points, and then a spatial interpolation is done by a successive correction method. |
2) Snow mass background field:
A very simple snow model evolution is used to build the background
 from the analysed snowfall  , the previous analysis of snow mass (persistence)  , a snow climatological mass  , and
an empirical melting function M based on the 2m temperature
forecast: |
|
(13) |
| or in a discretized form: |
|
(14) |
| 3) 3) Snow mass analysis: a sucessive correction method is applied to the snow mass increments (the difference between the background field in (2) and the observations of snow depth, suitably scaled assuming a fixed snow density of 250 kg m-3). |
The main shortcomings are that informations from the model (first-guess) are not taken into account in the analysis process and that the snow mass climatology, towards which the analysis is relaxed and in fact the dominant term in data void regions, is rather poor. The snow climatology (Brankovic and van Maanen 1985, BvM85) has been constructed by running to equilibrium an empirical snow mass model forced by precipitation monthly climate values and adjusted for melting using a temperature climatology; notice that no snow observations were used and that the spatial resolution is rather poor (
). Fig. 8 shows
BvM85 climatology for January and April, and compares its values with the
US Air Force (USAF) climatology (Foster and Davy 1988), the latter taking
into consideration an extensive collection of regional and global surveys
of ground based snow depth observations. Note that, since the USAF is a
climatology of snow depth, BvM85 snow mass values have to be converted using
an empirical specification of snow density: Eq (48) of Verseghy (1991), that takes into account
the packing of snow under its own weight, was used. It is clear that in
January, the BvM85 values extend too far south in Asia, overestimate the
snow depth in south Siberia, Mongolia and eastern Europe, northwestern and
eastern Canada, and underestimate the snow depth in northern Siberia, Iraq,
Iran, western Europe and Central Canada. BvM85 values over Greenland are
much larger than USAF, partly because the USAF represents a climatology
of seasonal snow depth, while BvM85 snow mass values have been assigned
the arbitrarily high value of 10 m, "to avoid unrealistic melting" (BvM85).
In spring BvM85 overestimates the snow depth virtually everywhere, but for
the european values (e.g. the Alps and Pyrenees), where a small underestimation
is detected.Mean monthly January and April snow depth values of the ECMWF reanalysis are shown in Fig. 9 , together with their difference to the USAF climatology. The overall patterns in the right panels of Figs. 8 and 9 are very similar, indicating that the BvM85 climate influences strongly the reanalysis values. Nevertheless, it is also clear that the information from observations in January is effectively used (values of top panel of Fig. 9 are smaller than the corresponding values in Fig. 8 ), especially for Europe and western Asia. Substantial errors in the BvM85 climatology in spring lead to large anomalies of the reanalysis mean values. As a further illustration of the reanalysis information brought by the use of observations, the monthly deviations of snow mass from the BvM85 values is shown on the left panels of Fig. 10 , while the right panels show the standard deviation of the monthly mean values of analysed snow mass. Since we are dealing exclusively with snow mass values, there was no need to use the density relation in this figure. The northeastern Asian values of snow in winter are similar to BvM85, and the standard deviation is small. A separate analysis (Corti et al. 2000) shows that a very large northern Asian area corresponding to the former Soviet Union has a data cover very inhomogeneous in time: the number of observations from 1979 to 1992 are 10 to 30 times smaller than in 1993. Values for the month of May (not shown) display larger differences to the climate and larger variability in northern latitudes. This is due to interannual variability in the rates of melting implictely used in the snow analysis algorithm (Eq. (14)), and controlled by the first-guess two-metre temperature.
The analysis algorithm has been further evaluated by Martin (1996), and results for snow mass are displayed in Fig. 11 . The values of the reanalysis (labelledERA in the picture) are compared with those produced by an off-line integration of a detailed, physically based snow model, CROCUS (Brun et al. 1989; 1992), forced by ERA values of screen level temperature, wind speed and humidity, precipation, and semi-empirical derived values for incoming radiation. The picture shows daily values, from 1 August 1981 to 31 July 1991, for a point in the French Alps (47 N 6 E), where the model elevation is similar to the station height (1100 m). The reanalysis snow mass values capture well the interannual variability, with maximum values for the years 81/82 and 83/84, and minima for 82/83, 88/89 and 89/90. However, the reanalysis values are systematically too high for the accumulation period, with ERA overshooting CROCUS values at times. The melting period comes too early, although the reanalysis melting rates are sometimes similar to values from CROCUS. The snow analysis (step 3 above) requires the specification of a snow density to convert the observed depth of snow in equivalent water depths (mass values) which are the quantities managed by the land surface scheme. A fixed value of 250 kg m-3 is assumed which appears to be too high for fresh snow, explaining the overestimation during the accumulation period. In the reanalysis values, the relaxation coefficient towards the BvM85 climatology is
= 0.02 which corresponds to a time scale of 
= 12.5 days. Poor spatial resolution over the Alps might result in lack
of data and a relaxation to the underestimated climatological values (see
bottom right panel of Fig. 8 over the Alps).Possible improvements of this technique could be to use an updated snow mass climatology based on direct observations and available at a resolution on 1 degree compatible with the current ECMWF model resolution (Sellers et al. 1996). The land surface scheme could be improved to provide a more reliable first-guess. With the introduction of the snow density as a prognostic variable (Douville et al. 1995) it should be possible to perform an analysis in terms of snow depths, the observed ground-based variable, instead of snow water equivalent (since snowfall can easily be converted knowing density of fresh snow). Remotely sensed information should be used, since weekly, and more recently daily, snow cover charts are prepared by NOAA/NESDIS (Bruce Ramsay, personal communication, see also Sellers et al. 1996) based on a combination of visible, infrared and microwave imagery. Microwave values could be used to obtain snow mass and snow density, using the dependency of the signal on frequency and polarization (see review in Hallikainen 1996a; 1996b); the available algorithms are more accurate for dry snow, and have problems for wet snow or when there are large variations of the roughness in the field of view. The values of ECMWF analysis of snow mass should be routinely compared with the US Air Force daily analysis of snow depth, based on ground observations and satellite imagery.
7.2 Deep soil temperatures
Since deep soil temperatures evolve over time scales much longer than short and medium range forecasts, the lack of initialisation for these variables can lead to potential drifts as the one observed for soil moisture in the ECMWF model. Experience at ECMWF during the winter 95/96 has shown that this problem can occur in winter, when the atmospheric forcing over the surface is weaker, and the thermal influence form the soil deeper layers is more important. The combination of a misrepresentation of some physical processes in the land surface scheme (soil water freezing), excessive radiative cooling, and insufficient downwards heat transfer in very stable boundary layers led to excessive cooling of the soil, and large negative biases in screen level temperature. Nevertheless, the problem is less critical than the summer soil moisture drift since in winter the stable structure of the boundary layer decouples the atmosphere from the surface below and prevents surface errors from propagating into the mid-troposphere.
Unlike soil moisture, routine measurements of profiles of temperature in the soil are performed by most meteorological offices. Unfortunately, only very few of these measurements are transmitted in the GTS, and therefore no global or even continental coverage exists in real time. A handful of european countries sends daily values to ECMWF, and those are used regularly for validation and monitoring the model values. An example in Fig. 12 , where an average profile of soil temperature for stations in northern Germany is compared with the corresponding model values, for a recent date. The model deep temperatures show a cold bias, but it is interesting to note that the model gradient is compares well with observations. This might indicate reasonably good soil thermal properties (they determine the vertical gradient), but comparatively poor atmospheric forcing in the previous weeks (determining the rate of cooling of the whole soil slab). The results are consistent with the well known systematic underestimation of longwave downward surface radiation of the ECMWF model and other models (Wild et al. 1995; Garratt and Prata 1996).
The only existing method for initialising soil temperature has been proposed by Coiffier et al. (1987) where analysed increments at 2 m are reported directly in the soil. This sub-optimal method could be improved by an optimal interpolation following the methodology of Mahfouf (1991).
7.3 Vegetation properties
Xue et al. (1996) have recently shown that the specification of the seasonal variations of vegetation properties can have a significant impact on the simulation of monthly mean temperatures near the surface over the North American continent. The sesonal evolution of the vegetation can be of importance over areas with agricultural practices (mid-latitude continents, tropical regions). Vegetation properties are currently either fixed spatially, as in the ECMWF model (Viterbo and Beljaars 1995), or can vary from month to month according to correspondence tables (Mahfouf et al. 1995). In an operational context, the use of satellite data like Global Vegetation Indexes (GVI) (see Gutman 1994) could be a better way to capture seasonallity and inter-annual variability of the vegetation. The major difficulty is to relate satellite reflectances to input parameters of the land surface scheme (leaf area index, albedo, vegetation cover,...). However, variational techniques already used for the retrieval of atmospheric vertical profiles from satellite radiances appear promising (Eyre et al. 1993).
Training Course Notes Front Page
Table of contents
Next Section
Previous Section
 |
12.06.2002 |
 |
© ECMWF |
