Modification of a weather analysis and its effect on the forecast

Water vapour (WV) radiance mesasurements from satellites can give valuable information on the state of the atmosphere. To a first approximation, they indicate the topography of the tropopause. The potential vorticity (PV) on certain isentropic levels is also indicative of the tropopause topography. By comparing WV from satellites with PV of a numerical weather analysis, it is possible to identify errors in the numerical weather analysis. The fundamental idea in this project is that when such an error is detected, one should be able to modify the PV of a model field based on WV observations and subsequently let the numerical model calculate a new (better) forecast.

A method has been developed to modify the PV of a numerical weather analysis and to assimilate this modified potential vorticity in HIRLAM, KNMI's
operational weather forecasting model. From the modified analysis, a new forecast can be produced. In a previous project (Vosbeek et al. 2001) the foundations were laid; in the current project the method was implemented in the operational version of the model and further refined. Now it runs fast enough to be used operationally. The previously
developed graphical interface to compare and modify PV and WV was used to determine modifications. But a new graphical interface was developed for future use.

Two cases where investigated in which the modification method was used as a research tool. The development of a small upper-level low and a larger
upper-level cyclone were studied. There was no reason to modify the original analysis since the forecast was accurate. But by eliminating or reinforcing some PV structures the effect of these structures on the development of the upper-level
low and the cyclone could be investigated. The results gave indeed more insight in the source area of the PV structures that were essential for the development and in the effect of other PV structures on the final position and shape. The high resolution and the possibility to study realistic meteorological cases made the method
a valuable research tool.

A third case was studied which was indeed motivated by a mismatch between the WV images and the PV of the analysis. In the ensuing forecast the
development of a depression was underestimated. A displacement of PV improved the forecast in terms of mean sea level pressure of the depression. But the mechanism of development was not identical to the actual development. To test the sensitivity
to the precise modification, similar but slightly different modifications were applied. These simulations gave a stronger depression than the original run, but not all were strong enough or predicted the right position. The correspondence between PV and WV is too qualitative to constrain the modification well.

Results were compared with the ECMWF ensemble in terms of mean sea level pressure and position of the depression, showing that the original forecast corresponds to an extreme ensemble member whereas the modified run was closer to the average. The modifications have created a `poor man's ensemble'. But the important differencewith classical ensemble methods is that the ensemble is not based on maximum
growth of structures but on observations and conceptual models applied by a meteorologist. In the near future, the developed PV modification method has most potential for the operational forecasting practice when it is used to generate short-term ensembles.