The analysis of surface parameters has an important role for short-
and medium- range weather forecast. The lack of direct observations for
soil moisture and soil temperature, and the substantial physical and technical
problems involved in the use of remote sensing, has leaded to develop indirect
schemes based on the relation with 2 meters variables (and temperature
and relative humidity), using different techniques (Optimal Interpolation,
Nudging, 1D-var).
All these techniques are based on the assumption that the near surface atmosphere is informative about the soil state, and therefore are not reliable in every meteorological situation (for instance in case of strong horizontal advection, where the screen-level parameters are mostly related to the upper model levels).
The NWP mesoscale model ALADIN is used to perform the 48h forecast with 10 km grid-length. The ISBA scheme, implemented in ALADIN (as in ARPEGE), threats the exchanges of water and energy between soil, vegetation and atmosphere. It counts a 2 soil layers parameterization for soil temperature and soil wetness with a forcing restore constraint, as described in Blackadar (1976) and Deardorff (1977).
The skin layer is of about 1 cm and the total soil layer is of order of 100 cm. Parameterization for frozen water content and snow is considered. Since the total soil layer controls mainly the evaporation from soil towards the atmosphere, the initialization for this layer and correction to the initial field has to be carefully applied.
The application of sequential analysis for water content based on relation with near surface atmosphere without strong physical constraints could lead to oscillation within the assimilation cycle or in subsequent forecasts.
Study of diurnal biases for Screen-Level parameters in ALADIN-France
The quality of the current configuration of ALADIN France forecast for Screen-Level Variable has been tested: 48 Hours forecast runs with hourly output from ALADIN France are evaluated on a high resolution dataset for hourly observations of 2m temperature and relative humidity.
A catalogue of the critical bias/rms for these variables is built over a selection of cases likely to be highly sensitive to soil surface initialization. The quality of 2 meters forecasting is estimated over a set of significant cases and an attempt to identify forecast errors due to initialization and physical parameterization is made. This step is also required to state the reliability of assimilation technique from 2 meters variables over the guess provided by the model.
In anticyclonic circulation, for instance, a weaker dependency from the upper level variables allows to state a clearer correlation between screen level variables and soil conditions.
The dataset consistency is obtained by application of the operational quality control. Removal of station gathering less than 80 percent valid observation during the case period is also applied. Station at elevation higher than 800 m are not considered. The check of OMG (observation - guess) is performed over the subdomain [LAT( 42 , 52 );LON( -5 , 10 )].
A subjective selection of Meteorological Cases is done to obtain a period of several days of anticyclonic circulation interesting the considered subdomain according to indications provided by sensitivity study (Bouttier et al., 1993) where near-surface atmosphere is mostly informative about soil state (strong radiation forcing, weak winds, no precipitation). Cyclonic and trough synoptic pattern are also considered for completeness.
It is performed the mean BIAS and RMS over the period for each selected case:
Figure 2 - Observation - forecast on 16/06/2000 - 42h forecast for
17/06/2000 at 18:00
Figure 3 - Initail Soil Wetness Index for 16/06/2000 at 00:00
Figure 4 - Maximum absolute value of the SWI correction for the whole period 13-18/06/2000
Note that what was considered here was the skill of the entire system analysis + forecast while to test only the forecast skill of the scheme, the analysis should be disabled using for instance a climatological initialization.
The initialization for SWI in the June Case
The soil wetness impact on 2 meters variables forecast is investigated for the 13-18/06/2000 case where BIAS and RMS are larger. The Soil Wetness Index is chosen as it encompasses the water content in the total soil in a fractional form. Correlation of errors is unreliable for SWI, due to the lack of observations for soil variable and no fine resolution climatology for SWI. A correlation between Strong biases of T_2m and unrealistic values of SWI can be shown in well developed AC days where the impact of other processes is reduced. The 17/06/2000 is selected for the comparison as showing the best AC pattern over the Target domain and previously 4 days of AC circulation (mainly clear sky).
The Analysis Correction applied in the initialization of SWI is investigated in the June Case. The entity of readjustments applied to initial SWI is evaluated by looking at the mean absolute value of the SWI correction provided by the analysis in subsequent days are computed for the whole period 13-18/06/2000. Maximum absolute value of the SWI correction is similarly calculated. The huge correction applied for the total soil layer water content (~ 1m depth) during the considered period are unrealistic and follow mainly the readjustments of the OI analysis of surface parameters (the ISBA parameterization scheme is not providing such big variations during forecasts).
The Sequential Analysis by OI technique for Surface Fields
A sequential analysis by application of Optimum Interpolation on a mesoscale domain allows a tuning of coefficients on ALADIN France domain with a lower number of constraints to the coefficients in operational use for LAM than for GCM. Furthermore a high resolution network is available for validation and the application of the method is a preliminary step before going to a variational approach. In order to compute the OI coefficient, it is necessary to provide the errors statistics for screen-level and surface variables.
The Errors statistics (Stat. NMC method)
The statistics are performed in winter and summer meteorological seasons separately, using a NMC type formulation for the forecast errors. Some constraints for the statistics are applied. For soil temperature Delta T_2m < 8 degrees and Delta H_2m < 60 percent. For soil moisture, in addition, are not considered grid points characterized by glaciers, snow cover, precipitation in the past 3 hours of forecast, and points with frozen soil. The assumption for observational errors are of 1 degree for 2 meters temperature and 0.10 (10 percent) for 2 meters relative humidity. The formulation of the OI coefficients is then obtained by solving a system of linear equations derived by minimizing the root mean square error of the estimation.
Conclusions and perspectives
The RMS and BIAS study for the current operational configuration show relatively small errors, but correction applied to soil fields (SWI) in order to minimize 2m errors are unphysically large and not constrained by any climatological value. The test on a specific case shows that not the entire information for the analysis is extracted for 2m correction and there is still ground for improvements. The contribution of a surface analysis (sequential or variational) based on ALADIN France guess is considered. A first set of OI coefficients is calculated but fine tuning of OI coefficients has to be done. Application of sequential analysis on surface analysis, and also the 1D-VAR analysis for surface variable with multiple time windows (24 or 48 hours) will be considered in later work.
Reference articles
Giard D. and Bazile E., 2000. - "Implementation of a new assimilation scheme for soil and surface variables in a global NWP model". M.W.R., VOL.128, 997-1015.
Douville H., Viterbo P., Mahfouf J.F. and Beljaars A., 2000. - "Evaluation of Optimum Interpolation and Nudging Techniques for Soil Moisture Analysis using FIFE Data". M.W.R., VOL.128, 1733-1756.
Bouyssel F., Cassé V. and Pailleux J., 1999. - "Variational surface
analysis from screen level atmospheric parameters". Tellus, n.51A, 453-468.
 |
Home |