The purpose of this study was to establish the role played by the initial conditions (IC) and lateral boundary conditions (LBC) in short range mesoscale forecast. The adjoint method has been used to compute the gradients of the forecast errors with respect to the ICs and LBCs. These were introduced as small perturbations of the type dx=- a*Ñx in the initial data, x being the model state fields and a a scaling factor. The sensitivity tests were carried out for different forecast range and integration area. We have tried to distinguish between the errors coming from a lack of information in LBCs and inaccuracies in the initial state.
"Irish Christmas Storm" or the 24 December 1997 case
The domain used for the experiments has a grid size of 30km, elliptic truncation T49 and 150x144x19 points. The integration time chosen in this study was mainly 36 hours starting from 23 December 1997, 00UTC. The results have demonstrated that the ICs play a leading role for a good forecast. On the contrary, the forecast is less sensitive to the modifications in the LBCs. In Fig.1a the gradients with respect to the IC are shown. Large gradients values are concentrated in the southwestern part of the domain. This is the region where small perturbations introduced in ICs have a positive impact on the forecast. For example, 2.5 hPa have been added in the sea level pressure field taken the difference between operational ICs and the modified one. Starting from these modified initial data, the model arrives to a solution that gives birth to a cyclonic core over the British Islands as it is illustrated in Fig. 1b. The pressure of the cyclone center is 983 hPa, close to the verifying analysis available 24 December 1997, 12UTC whose core is 988 hPa but is very far from the reality since at Valentia meteorological station, the mean sea level pressure was 975 hPa. Even if the forecast cyclone is not as deep as in reality, we can say that the errors coming from ICs are responsible for the failure of the operational forecast. These experiments are in accord with those undertaken for the same case but with the global ARPEGE model (G. Hello at. al, 2000).
The second French storm or the 27 December 1999 case
The domain used in our studies was the operational ALADIN-France (T95, 288x288x31, dx=9.9km). For this high-resolution integration area a first problem regarding the gradients computation has been revealed. The strong punctual gradients over the Alps have been found (Fig. 2a). These structures are present in adiabatic model and they do not vanish whenever the Eulerian time step is reduced from 40 to 10s. It was found that even for 10s timestep there are high values over the orography for certain levels. For diminishing these unrealistic values, the boundary layer height (ho) and the scale factor for the wind speed (u*) in the adjoint linear simplified physics (Buizza, 1993) have been modified. Another way to get rid of these structures was to change the trajectory truncation in adjoint model (ex. from T95 to T35). Again like for Irish storm the influence of the LBCs is in positive sense but is rather weak. An improved forecast is obtained if the ICs are modified. The sensitivity forecast was able to move the center of the cyclone closer to the verifying analysis but with a very high price. For example, more than 5hPa have been added into a region where a rest of useful gradient was present over Biscaye Bay. It is very likely that an important part of the gradient has already gone out from the domain crossing the western boundaries and it should be found over the Atlantic Ocean.
Discussion
During this first ALATNET stay, the study has been concentrated on the effect of both the perturbations in lateral boundary data and in initial conditions in the prediction of a rapid evolving cyclogenesis. In all the experiments in which only lateral boundary perturbation was injected, the impact in the forecast was very marginal. On the contrary, we have shown that the errors in ICs are more responsible for the forecast failure. This is not because the information coming from the coupling area is perfect. The reason could be that the utilized technique for deriving LBCs gradients is not proper for ALADIN. We think that this failure is due to the coupling method (Davies) used and we could say that for a future 4D-Var assimilation in ALADIN a new coupling method should be imagined which, eventually, should be able to distinguish between inflow and outflow region of the domain.
In the future, other case studies will be carried out mainly to study the impact of simplified physical parameterization in ALADIN adjoint model. Then we will concentrate on the sensibility of mesoscale phenomenon inside the forecast area.
References
Buizza, R., 1993. Impact of a Simple Vertical Diffusion Scheme and of the Optimisation Timer Interval on Optimal Unstable Structures. RD Tech.Memo. No. 192, ECMWF.
Errico, R., T. Vukicevic and K. Raeder, 1993. Comparison of the initial and lateral boundary condition sensitivity for a limited-area model. Tellus, 45A, 539-557.
Gustafsson, N., E. Kallen and S. Thorsteinsson, 1998. Sensitivity of forecast errors to the initial and lateral boundary conditions. Tellus, 50A, 167-185.
Hello, G., F. Lalaurette and J-N. Thépaut, 2000. Combined use of sensitivity information and observations to improve meteorological forecast: A feasibility study applied to the 'Christmas storm' case. Quart. J. Roy. Meteor. Soc. 126, 621-647.
Janiskova, M., J-N. Thépaut and J-F. Geleyn, 1999. Simplified and regular physical parametrizations for incremental four-dimensional variational assimilation. Mon. Wea. Rev., 127, 26-45.
Rabier, F., E. Klinker, P. Courtier and A. Hollingsworth, 1996. Sensitivity of forecast errors to the initial conditions. Quart. J. Roy. Meteor. Soc. 122, 121-150.
 |
Home |