(more details : samuel.westrelin @ meteo.fr)
1. Soil water freezing parameterization in ISBA
(Interaction between Soil, Biosphere and Atmosphere)
The land surface scheme ISBA has been modified by Giard and Bazile and implemented in ARPEGE and ALADIN on March 1998. A one layer soil water freezing had been implemented to solve the cold bias over frozen areas and to parameterize the impact of the freeze-thaw cycles on the thermal and hydrological characteristics of the soil. It successfully solved these problems but two deficiencies appeared : it is time step dependent and it prevents the increase of the surface temperature during the thawing period (figure 1, dotted line compared to full line).
Figure 1 : Col de Porte (1D model). Surface temperature during the thawing period (1999/04/13 to 1996/04/26). Full line: measured infrared temperature ; dotted line : ISBA with one layer freezing ; dashed line : ISBA with two layers.
As a solution a characteristic time and a superficial reservoir for frozen water (figure 1) have been introduced. The objective scores are presented on figure 2 over North America, where freezing is still effective on May, date at which the parameterization became operational. This modification was introduced quite late to allow the reservoirs to be in equilibrium for the next winter over Europe. New background error statistics for variational analysis and some modifications in soil moisture assimilation were introduced at the same time.
Figure 2 : Bias and root mean square error of corrected 2m temperature (°K, left panel) and 2m humidity (%, right panel) of ARPEGE with one frozen layer (PA, full line) and ARPEGE with two layers and new guess errors statistics - changes which had no impact on temperature in altitude - (PAD, dashed line) against SYNOP over North America from 2nd to 26th of may 1999 as a function of forecast range in hours. The number of observations is on the right axis.
The bias in temperature has been successfully reduced as well as the bias in humidity.
Reference :
Bazile, E. (1999). The Soil Water Freezing in ISBA. In HIRLAM Newsletter, Number 33, pp. 92-95. HIRLAM 4 Project, c/o Met Eireann, Glasnevin Hill, Dublin 9, Ireland.
2. Soil water content analysis
After one year of operational use, some problems have been identified in ISBA. The soil water content was oscillating too much especially during summer 1998. Moreover it showed an unexpected drying in winter because the surface analysis tried to correct the cold bias of the model by drying the soil.
To correct these oscillations, some changes in ARPEGE soil moisture assimilation have been undertaken by Eric Bazile and François Bouyssel.
A first modset, on may 1999, consisted in the following points :
- soil moisture analysis is switched off in case of frozen soil;
- the constraint on evaporation is stricter;
- the interaction between analysis conditions and smoothing of the diurnal cycle is modified to further limit soil moisture corrections.
This first set of modifications only slightly corrected the oscillations (figure 3, dashed line compared to full line).
Some stronger changes have then been implemented on October 1999 :
- the analysis conditions were replaced by a continuous weighting of soil moisture increments, p(ten-meter wind, evaporation rate, precipitation, snow cover, ...) in the range [0,1] and equal to 1 when all conditions are satisfied;
- the threshold for precipitations was reduced from .6mm/6h to .3mm/6h;
- a damping by the mean low-level cloudiness along the previous 6h forecast was added;
- the basic analysis coefficients a1, a2 relating corrections of the mean soil moisture to analysis increments of 2 meters temperature and relative humidity :
DWp = ( a1..DT2 m + a2.DH2m ). p(..., cloudiness)
were divided by 3 (a1, a2 depend on soil and vegetation characteristics and on the local solar time);
- the computation of the mean bias on T2m changed to reflect longer time-scales and the interaction with smoothing was again modified.
These changes brought a significant smoothing of the evolution of the mean soil moisture (figure 3).
Figure 3 : Evolution of soil wetness index over France from ARPEGE analysis during the summer 1998 for the old ISBA surface analysis (operational on march 1998, full line), the operational analysis on june 1999 (dashed line) and the operational one on october 1999 (dot-dashed line).
As regards to objective scores (figure 4), on one hand the model appears warmer and clearly reduces the previous cold bias. On the other hand the humidity deficit has been increased by a few percents. This modset moved to operations together with the CYCORA package.
Figure 4 : Bias and root mean square error of corrected 2m temperature (left panel) and humidity (right panel) of ARPEGE against SYNOP over France from september 1999 the 9th till october 1999 the 10th as a function of forecast range in hours. ARPEGE operational on june 1999 (PA, full line), ARPEGE operational on october 1999 (PAD, dashed line). The number of observations is on the right axis.
3. A few scores with parameterization changes in cyclogenesis, convection and radiation
The operational suite at METEO-FRANCE includes some changes in the parameterization of precipitating phenomena to correct the lack of stratiform clouds at the top of the tropical convection, to give more realism to the mesoscale convection and to avoid the dependency of the part of stratiform precipitations in the total precipitations on the model resolution. Some other changes have been implemented following the cyclogenesis failure on December 1998 the 20th over Britany. They also contain different clouds optical properties and a parameterization of clouds at the top of the boundary layer. They became operational on October 1999. See Newsletter 16 page 36 for more details.
Figure 5 : Differences of scores of geopotential (m) (root mean square error, standard deviation and bias from left to right) between ARPEGE before (PA) and after (PAD) the parameterization changes against radiosondes over the tropics (20°S-20°N) for different pressure levels (y axis) from september 1999 the 9th till october 1999 the 10th as a function of forecast range in hours. Improvements are in full line, deteriorations in dot-dashed line, neutrality in dotted line.
Figure 6 : Differences of scores of temperature (K) (root mean square error, standard deviation and bias from left to right) between ARPEGE before (PA) and after (PAD) the parameterization changes against radiosondes over the tropics (20°S-20°N) for different pressure levels (y axis) from september 1999 the 9th till october 1999 the 10th as a function of forecast range in hours. Improvements are in full line, deteriorations in dot-dashed line, neutrality in dotted line.
The main improvements are located over the tropics for geopotential and temperature, especially by reducing the bias above 200 hPa (figures 5 and 6). Over other regions scores are neutral except some slight positive signals, not shown here, above 20°N and below 20°S in geopotential.
 |
Home |