3D-Var experiments for the ALADIN/HUNGARY model: a case study (II)

Introduction

The research on the 3D-Var data assimilation scheme for a double-nested limited area model has been continued with another case study, from the 4th of August 2002, when the operational model (in dynamical adaptation) has not been able to predict an important quantity of precipitation over Hungary.

The synoptic situation of this event is presented briefly. On the 4th of August 2002, a lower pressure field dominated over the central-eastern part of Europe. In altitude, the ridge of the anticyclone from North Africa was extended over the central-southern part of Europe, and a low cut-off was located more to the West. The temperatures in the western part of Hungary were in the morning around 23°C, increasing till 30°C at noon. The air in this area was very humid, which determined an unstable atmosphere.

A cold front was moving to the north-eastern part of Europe, through Austria and Hungary. So cold and dry air was streaming south-westerly. Advancing, the cold air moved the humid air upward, determining a convective instability. The non-frontal surface line, along which convective instability occurs, is called an instability line. Here the turbulence is severe, because of the violent updrafts and downdrafts. Thus rain showers and thunderstorms occur, with significant quantities of precipitation. Such an instability line developed in the north-western part of Hungary, causing an intensive rainfall. Thus, in less than six hours, the quantity of precipitation reached a maximum of 65 mm. (Fig. 1).

3D-Var experiments

The ALADIN/Hungary model is using for the background term the standard NMC statistics, and for the observation term, only SYNOP and TEMP data are considered in the assimilation cycle. As coupling technique, the time-consistency has been chosen both in cycling and in production. Digital filter initialization (DFI) has been applied at the beginning of the integration, both in cycling and in production (Alexandru, 2002).

The 3D-Var experiments were carried out using the operational lateral boundary conditions (LBC) for ALADIN/Hungary (from ALADIN/LACE model). From SYNOP observations, temperature and relative humidity information have been considered, and from TEMP data, wind, geopotential, temperature and relative humidity. Hereafter the name of this set of experiments is " 3dvar". The reference of these experiments was the operational forecast of the model in dynamical adaptation ("oper"). The assimilation cycle was started from 01.08.2002 06 UTC, i.e. three days before the event.

Results

In this chapter the forecasts of different fields from the model runs are described, then the impact of observations over forecast has been analysed. The maps with the cumulated precipitation have a zoom between 44°-50° N in latitude and 14°-25° E in longitude, in order to point out the location of the rainfall.

For the 04.08 00 UTC model run, both models, with and without data assimilation, did not predict the significant rainfall, either at the right moment, nor later. The quantity of precipitation do not exceed 6 mm, which is a strong underestimation, the real maximum being 65 mm. So the operational model, in dynamical adaptation, failed to predict the rainfall. The 3D-Var experiment did not have a better forecast, which means that the information coming from the observations did not give any indication about the future development of an instability line.

Also the results of the 3dvar set were checked from the 04.08 06 UTC run, but still no indication about heavy precipitation. One reason of this misforecast could be the lack of any TEMP observation near to this region, so only information from SYNOP data were taken into account. Probably because the event has happened between 06 UTC and 12 UTC, especially to the end of the period, the surface observation at 06 UTC did not have information about the event.

04.08 12 UTC run

The main difference in the forecasts of the two models, with and without 3D-Var scheme, appeared in this run. The operational analysis shows the air streaming south-westerly. The pressure gradients are rather small, the wind is not too strong, over Hungary. The difference between the models, in the forecast of the geopotential, appeared since the beginning of the integration. The 3dvar set predicted a lowest pressure over the western part of Hungary. After 6 h integration, the difference between the models for the forecast of the geopotential is almost 3 damgpm.

The operational model predicted the sky partly cloudy over Hungary and no vertical motion of the air masses. So, no precipitation is expected to develop. The model using 3D-Var scheme (3dvar), forecasted an area in the western part of Hungary with high humidity. Also the air masses have an ascending vertical velocity, reaching values as 5 hPa/s.

In the north-western part of Hungary a significant rainfall has been forecasted by the experiments using 3D-Var scheme, the quantity reaching 173 mm in 6 h. However the operational model did not predict any precipitation. One can see in Fig. 2 the great difference between the operational forecast and that when data assimilation has been used.

Seeing the difference between the two models, another set of experiments has been proposed. It is similar to the 3dvar set, but the lateral boundary conditions were provided by the ARPEGE model. The forecasts of same fields have been analysed, but they are rather similar. Only the quantity of precipitation is smaller than in the 3dvar experiments, the maximum being around 126 mm in 6 h.

So the experiments using 3D-Var scheme succeeded to forecast a huge quantity of precipitation, but with six hours delay. On the one hand, this represents a failure of the model, but on the other hand, the rainfall was still predicted. We tried to see how it was possible.

It was investigated if there are some imbalances between the models, with and without data assimilation. But the time evolution of the mean-sea-level pressure checked during 6 h integration in production, for the model in dynamical adaptation, and using 3D-Var, showed that the fields are in balance.

The impact of observations on forecast

Trying to understand how the models using data assimilation forecasted the important quantity of precipitation, some experiments were carried out, similar to the 3dvar set, except that different combinations of variables and data have been used, all along the assimilation cycle. In the first two sets of experiments only TEMP, respectively only SYNOP, observations have been taken into account. The names are 3dvar TEMP and 3dvar SYNOP(T,RH). In the last one, only the relative humidity and temperature variables have been assimilated.

For all experiments, only the precipitation cumulated in 6 h, from the 04.08 12 UTC run, between 12 UTC and 18 UTC, has been plotted. As can be seen in Fig. 3, the 3dvar TEMP set did not predict any rainfall, which means that all the information about it came from the surface observations. Indeed, using only SYNOP data, the quantity of precipitation is significant. The shape of the area of rainfall differs as it was predicted by the 3dvar set, but still in the north-western part of Hungary more than 50 mm precipitation has been forecasted.

The next experiments have been performed using different observed variables only from SYNOP data. So in turn, relative humidity (" SYNOP(RH)") has been assimilated from the surface observations, then geopotential, then geopotential and temperature, and finally geopotential, temperature and relative humidity. Other experiments were carried out, being similar with the previous one's, only that TEMP observations have been added also in the assimilation cycle. The observed variables from TEMP data are wind, geopotential, temperature and relative humidity. The conclusions from these experiments were that the geopotential and temperature measurements did not influence the rainfall forecast (not shown). The main information is coming from the relative humidity. As one can see in Fig. 4, when only relative humidity from surface observations has been assimilated, the quantity of precipitation is more than 260 mm in 6 h.

Few precipitation (around 29 mm) has been predicted also by the 3D-Var experiments when TEMP data together with the temperature and geopotential from the surface observations were assimilated. This can be an effect of the inefficient coupling between the planetary boundary layer and the troposphere, made by the observation operator. Thus the 2m temperature observation increment can influence the temperature increments from the high troposphere.

In conclusion, the measurements of the 2 meter relative humidity have been responsible for the information introduced in the model, which caused the forecast of this significant rainfall. Also the temperature measurements, from the surface and upperair observations, determined the prediction for some precipitation.

Conclusions

A case from the 4th of August 2002 has been presented in this paper. An instability line has been developed very fast in the morning, causing an intensive rainstorm. The operational model (in dynamical adaptation) was not able to predict this phenomenon. So new experiments have been performed in order to see whether the model using the 3D-Var scheme can have a better forecast.

For these experiments, different lateral boundary conditions (from the ARPEGE or ALADIN/LACE model) and SYNOP and TEMP observations available at the Hungarian Meteorological Service have been used. The reference of these experiments was the operational forecast of the model in dynamical adaptation.

The forecasts from the 04.08 00 UTC and 06 UTC runs showed that both models, with and without data assimilation, did not predict the intensive rainfall, neither at the right moment, nor later. It means that the information coming from the observations did not bring any indication about the future development of this instability line.

The 04.08 12 UTC run brought the main difference in the forecasts of the two models, in dynamical adaptation and with 3D-Var scheme. The operational model predicted fair weather, without any sign of rainfall. Comparing to reality, one can say that for this particular period of time, the operational model has a good forecast, the rainfall happening earlier than this model run. But in the same time, it is also a failure of the operational model, which did not predict the precipitation at the right moment, but neither later. Being a local phenomenon which has developed very fast, it was not forecasted by the coupling model. So no information about it came through the lateral boundary conditions from the ALADIN/LACE to ALADIN/Hungary model.

Instead, the experiments using 3D-Var scheme forecasted a huge quantity of precipitation (more than 100 mm), but with six hours delay. On the one hand, this represents a failure of the model, but on the other hand, the rainfall was still predicted.

Other experiments were carried out using only SYNOP or TEMP data, in order to find out what observations could have influenced so much the precipitation forecast. Thus it was shown that the model using the 3D-Var scheme and only TEMP data, did not predict any rainfall, which means that all the information about it came from the surface observations. Indeed, assimilating only SYNOP data, the quantity of precipitation is significant.

The last experiments have been performed using different combinations of the observed variables from the SYNOP observations, with and without TEMP data in the assimilation cycle. The observed variables from SYNOP are the relative humidity, geopotential and temperature. From the TEMP data, the wind, temperature, relative humidity and geopotential have been assimilated. The precipitation forecasts from these experiments showed that the measurements of the 2 meter relative humidity have been responsible for the information introduced in the model, which caused the forecast of this significant rainfall. Also the temperature measurements, from the surface and upperair observations, determined the prediction for some precipitation, but not so important.

So the information from the observations (from 04.08 at 12 UTC) reproduced the state of the atmosphere, with high humidity. Thus the analysis of the model caught the end of the storm. Because any other information from the lateral boundary conditions or from the first-guess did not give any indication about an existing storm, the model "said" that it is the beginning. So, one can say that the model using 3D-Var scheme had good initial conditions, but it was beyond its capacity to predict such a rainstorm. (Alexandru, 2003).

References

Alexandru, S., 2003: 3D-Var experiments for the ALADIN/Hungary model: a case study (II). ALATNET Internal Note.

Alexandru, S., 2002: 3D-VAR data assimilation experiments for the double-nested limited area model ALADIN/Hungary. ALATNET Internal Note .