| Home > Newsevents > Training > Rcourse_notes > PARAMETRIZATION > SURFACE_ASSIMILATION > |
The role of the land surface in
the climate system
April 2002
European Centre for Medium-Range Weather
Forecasts
Table of contents >>
Next Section >>
Previous Section >>
5 . Examples from ECMWF recent experience
5.1 Soil moisture
July 1993 showed anomalously high precipitation over the Central USA, with exceptional flooding of the Mississippi (Changnon 1996). During this month, the new version of the ECMWF model (CY48) and the then operational version (CY47)1, were running in parallel at full resolution (spectral truncation T213, grid-point spacing ~ 60 km), including data assimilation. Beljaars et al. (1996) compared the performance of the two schemes, looking at the average of all one-, two- and three-day forecasts verifying between 9 and 25 July. While the day one precipitation of the two systems were very similar, and similar to the observed precipitation, the forecasts at day 3 were markedly different. In the new system, the location and intensity of the maximum precipitation was similar to the observations (40 N, 95 W), while the old system had less then half the precipitation amount in the area of the observed maximum and had a spurious maximum of precipitation displaced 800 km NE. In the old system, there was a gradual reduction in precipitation from day 1 to day 3, while the new system was able to better maintain the intensity. However, evaporation at the area of maximum precipitation was similar for the old and new system, and in both systems there was no evidence of forecast spin down, strongly suggesting that the local evaporation was not responsible for the differences in precipitation. It turns out that the maximum of the evaporation difference was located over the Mexican Plateau, 1000 km SW of the precipitation maximum, two to three days upstream as suggested by backwards trajectories ending up at 750 hPa, 40 N, 95 W. The mean thermodynamic profiles, similar for day 1 forecasts, were very different for day 3 forecasts. The old model showed a too strong capping inversion above the BL, with air too warm and too dry and much lower values of CAPE. It is clear that the differential advection mechanism characteristic of the US Monsoon was responsible for the differences in precipitation. When compared to the new model, the soil on the Mexican Plateau had much lower values of soil moisture in CY47, giving a much reduced evaporation, which in turn produced a warm and dry air mass that capped the BL downstream, inhibiting convection. In CY47, the soil model values were strongly forced to an erroneous, too dry, climatology: In such a data dense area, atmospheric profiles were initiated to correct values, but during the forecast they slowly felt the influence of the erroneous soil moisture values. In CY48 the soil moisture values were initialised to field capacity at the beginning of July, consistent with values of June precipitation in the area much above normal. There was no forcing to climatology in CY48 (Viterbo and Beljaars 1995) and the model was capable of maintaining high values of moisture throughout July. Monthly integrations performed with CY47 and CY48 suggested the importance of the memory associated to idealised soil moisture anomalies in the initial conditions (Beljaars et al. 1996). The monthly precipitation fields with CY48 compared much better to observations than those of CY47.
CY48 surface model ran with predicted soil moisture throughout the first half of 1994. It was clear that a very large near-surface warm and dry bias developed over the NH continental areas at the end of spring and beginning of summer. A scheme to initialise soil water based on the short-term forecast errors of near-surface atmospheric humidity was developed to overcome that problem (Viterbo 1996). In order to test the new scheme, three complete data assimilation-forecast experiments were ran at T213 for the month of June 1994: (a) Control (CY48, no assimilation of soil moisture); (b) As control, but using the initialised soil moisture values, and; (c) As in (b), but using a prognostic cloud scheme with much more realistic cloud cover over land (Tiedtke 1993). The near-surface warm and dry bias, reduced from Control to the initialised soil water experiment, in response to a wetter soil. A lower tropospheric warm bias developed in the Control model and was greatly reduced when initialisation of soil water is used. Both experiments had too little cloud cover over land with too large surface shortwave radiative fluxes, but the wetter soil conditions of experiment (b) managed to maintain evaporation in the face of excessive net radiation at the surface. The third simulation displayed even smaller biases, associated with a larger, more realistic cloud cover and smaller radiative biases.
The algorithm to initialise soil water is successful in controlling model drifts but dampens the seasonal cycle and interannual variations of evaporation and soil moisture. Viterbo and Betts (1999a) revisited the July 1993 simulation, using the initial soil water algorithm and the prognostic cloud scheme. The new system gives poorer results for precipitation than Beljaars et al. (1996). Although much better than CY47, there is a suggestion of northward displacement and reduction in the precipitation maximum in day 2 forecasts. It appears that the initialisation of soil moisture at field capacity at the beginning of July in Beljaars et al. was crucial to obtain a good simulation of the excessive rainfall events.
5.2 Boreal forests
Surface albedo is the prime regulator of the net energy available at the surface. The albedo of snow-free land surfaces ranges from values of 0.1 in forests to values of 0.35 over deserts. For areas seasonally covered with snow, that range can extend up to 0.85. Betts and Ball (1997) analysed the annual cycle of albedo in the BOREAS experiment, performed in Canada during 1994, focussing on the snow season, comparing several measurement sites located over grass, aspen and coniferous. Representative values for daily averaged albedo of snow-covered grass sites are 0.75, while corresponding values for the aspen and conifer sites are 0.21 and 0.13, respectively, with values as high as 0.4, one to two days after snowfall. The lower albedo of the boreal forests in the presence of snow corroborates data from other observational studies and the few attempts of making a hemispheric-satellite based estimate of albedo.
The ECMWF model version of 1994, at the time of the BOREAS experiment, treated the albedo of snow covered areas with no regard to the land cover: beyond a critical value for snow depth, the albedo of snow covered areas was rarely outside the 0.7-0.8 range. As a result, when compared to experimental results, net radiation was too low and near-surface air temperatures were too cold. A modification to the scheme was designed such as the albedo of snow-covered surfaces tended to the asymptotic value of 0.2 in the presence of forests and 0.7 otherwise (Viterbo and Betts 1999b). The modified scheme has much reduced biases in temperature and radiation in the high latitudes. The cold bias in temperature in the control scheme extended in the vertical to the whole troposphere, increasing with forecast range and affecting most continental areas. The bottom panel of Fig. 7 shows the day 5 forecast error of 850 hPa temperature, averaged for March and April in 1996. The very high albedo induces cooling errors exceeding -3 K in North America and -7 K in Asia. The top panel shows the corresponding figure for 1997. In spring 1997, with the new snow albedo scheme, the cold bias over the boreal forest has been almost eliminated. The new scheme also improved the quality of the medium-range forecasts, as evidenced by better scores of 500 hPa geopotential fields.
The forecast results above corroborate the study of Thomas and Rowntree (1992) on the role of the boreal forests in conditioning the climate at high latitudes. The spring months of two five-year experiments, the first with a (realistic) snow albedo and the second with the high latitude forests removed are compared. The latter experiment is colder than the former in the continental areas north of 50 N. Pielke and Vidale (1995) suggested that the boundary between tundra and boreal forests is a region of enhanced horizontal temperature gradients, acting as a pre-conditioner for baroclinic instability and "locking" the climatological position of the polar front. In analysing further refinements to the ECMWF snow model,van den Hurk et al. (2000) show that (a) simulating the boreal forest control on evaporation in spring (reduced transpiration from frozen soils) and (b) increasing the runoff over frozen soils, improves the agreement of model results with observations.
5.3 Soil water freezing: A regulator of cold climates
The operational ECMWF forecasts for the winters of 1993-94 through to 1995-96 showed a tendency to produce a cold bias over continental areas in winter. This error was particularly severe during the winter of 1995-96 for Scandinavia, a year characterised by reduced snow amounts and, consequently, larger thermal coupling between the soil and the atmosphere above. Viterbo et al. (1999) diagnosed two main problems contributing to that error. Firstly, the energy involved in phase changes in the soil was not taken into account. When positive temperatures approach 0 C, a substantial amount of the external cooling demand (i.e., infrared cooling) is used to freeze the soil water, thereby decreasing the rate of soil cooling; a similar effect occurs in melting. Soil water phase transitions act as a thermal barrier, increasing the soil inertia at temperatures close to 0 C. Secondly, the downward sensible heat flux, prevailing in winter conditions, was too small, leading the model into a positive feedback loop where cooling reduced the heat flux, and made the soil even colder. Model changes were designed to incorporate the missing physical mechanisms. Separate seasonal integrations with both model schemes revealed a greatly alleviated soil and near-surface atmospheric cooling drift: Screen-level temperatures reduced from -10 C to close to zero. Despite the considerable warming in the model soil and near-surface winter climate in continental areas, there was negligible impact on the free atmosphere temperature and the atmospheric flow. In winter, stable situations, the atmosphere is decoupled from the surface and changes at the surface do not propagate upwards, unless they affect the momentum budget.
Training Course Notes Front Page >>
Table of contents >>
Next Section >>
Previous Section >>
1 The main differences between CY47 and CY48 rely on the surface and boundary layer processes parametrization. The soil model in CY48 has 4 layers, with no heat flux and free drainage bottom boundary condition, with soil properties (heat and water conductivities and diffusivities) dependent in a non-linear way on soil moisture (Clapp and Hornberger 1978). CY47 has 2 layers plus 1 climate layer underneath, with constant water soil properties. CY48 has a smaller roughness length for heat than for momentum, while in CY47 they are identical. For more details see Viterbo and Beljaars (1995).
Copyright © 2002, ECMWF. All rights reserved.
 |
12.06.2002 |
 |
© ECMWF |
