• Appenzeller, C., and H. C. Davies, 1992: Structure of stratospheric intrusions into the troposphere. Nature,358, 570–572.

  • Arpe, K., A. Hollingsworth, M. S. Tracton, A. C. Lorenc, S. Uppala, and P. Kallberg, 1985: The response of numerical weather prediction systems on the FGGE level IIb data. Part 2: Forecast verifications and implications for predictability. Quart. J. Roy. Meteor. Soc.,111, 67–101.

  • Bader, M. J., G. S. Forbes, J. R. Grant, R. B. E. Lilley, and A. J. Waters, 1995: Images in Weather Forecasting: A Practical Guide for Interpreting Satellite and Radar Imagery. Cambridge University Press, 499 pp.

  • Bell, R. S., B. Macpherson, D. Robinson, and P. Andrews, 1995: Technical details of the Unified Model data assimilation scheme. Unified Model Documentation Paper P3, 53 pp. [Available from United Kingdom Meteorological Office, Bracknell, Berkshire RG12 252, United Kingdom.].

  • Böttger, H., M. Eckardt, and U. Katergiannakis, 1975: Forecasting extratropical cyclones with hurricane intensity using satellite information. J. Appl. Meteor.,14, 1259–1265.

  • Browning, K. A., S. A. Clough, C. S. A. Davitt, N. M. Roberts, T. D. Hewson, and P. G. W. Healy, 1995: Observations of the mesoscale substructure in the cold air of a developing frontal cyclone. Quart. J. Roy. Meteor. Soc.,121, 1229–1254.

  • Chesters, D., L. Uccellini, and A. Mostek, 1982: VISSR Atmospheric Sounder (VAS) simulation experiment for a severe storm environment. Mon. Wea. Rev., 110, 198–216.

  • Cullen, M. J. P., 1993: The Unified Forecast/Climate Model. Meteor. Mag.,122, 81–94.

  • Davis, C. A., and K. A. Emanuel, 1991: Potential vorticity diagnostics of cyclogenesis. Mon. Wea. Rev.,119, 1929–1953.

  • Demirtas, M., 1996: Sensitivity of short-range forecasts to local potential vorticity modifications. Ph.D. thesis, University of Reading, Reading, United Kingdom, 227 pp. [Available from Department of Meteorology, University of Reading, Whiteknights, Reading, Berkshire RG6 6AH, United Kingdom.].

  • ——, and A. J. Thorpe, 1996: A new PV-based method for short-range weather forecasting quality control and intervention. Preprints, Seventh Conf. on Mesoscale Processes, Reading, United Kingdom, Amer. Meteor. Soc., 396–398.

  • Downton, R. A., and R. S. Bell, 1988: The impact of analysis differences on a medium range forecast. Meteor. Mag.,117, 279–284.

  • Evans, M. S., D. Keyser, L. F. Bosart, and G. Lackmann, 1994: A satellite-derived classification scheme for rapid maritime cyclogenesis. Mon. Wea. Rev.,122, 1382–1416.

  • Fehlemann, R., and H. C. Davies, 1997: Misforecasts of synoptic systems: Diagnosis via potential vorticity retrodiction. Mon. Wea. Rev.,125, 2247–2264.

  • Fischer, H., N. Eigenwilling, and H. Muller, 1981: Information content of Meteosat and Nimbus/THIR water vapor channel data: Altitude association of observed phenomena. J. Appl. Meteor.,20, 1344–1352.

  • Hewson, T., 1993: The FRONTS ’92 experiment: A quicklook atlas. JCMM Internal Rep. 15, University of Reading, Reading, United Kingdom, 176 pp. [Available from JCMM, University of Reading, Whiteknights, Reading, Berkshire RG6 6AH, United Kingdom.].

  • Hoskins, B. J., 1990: Theory of extratropical cyclones. Extratropical Cyclones. The Erik Palmén Memorial Volume, E. O. Holopainen, Ed., Amer. Meteor. Soc., 63–80.

  • ——, M. E. McIntyre, and A. W. Robertson, 1985: On the use of and significance of isentropic potential vorticity maps. Quart. J. Roy. Meteor. Soc.,111, 877–948.

  • Ingleby, N. B., and C. A. Parrett, 1994: Quality control of atmospheric data. Unified Model Documentation Paper P32, UK Meteorological Office, Bracknell, United Kingdom, 36 pp. [Available from United Kingdom Meteorological Office, Bracknell, Berkshire RG12 252, United Kingdom.].

  • Joly, A., and Coauthors, 1997: Definition of the Fronts and Atlantic Storm-Track Experiment (FASTEX). Bull. Amer. Meteor. Soc.,78, 1917–1940.

  • Keyser, D., and M. A. Shapiro, 1986: A review of the structure and dynamics of upper level frontal zones. Mon. Wea. Rev.,114, 452–499.

  • Manney, G. L., and J. L. Stanford, 1987: On the relation of 6.7 μm water vapour features to isentropic distributions of potential vorticity. Quart. J. Roy. Meteor. Soc.,113, 1048–1057.

  • Mansfield, D., 1996: The use of potential vorticity as an operational forecast tool. Meteor. Appl.,3, 195–210.

  • Molteni, F., R. Buizza, T. N. Palmer, and U. Petroliagis, 1996: The ECMWF ensemble prediction system: Methodology and validation. Quart. J. Roy. Meteor. Soc.,122, 73–119.

  • Petersen, R. A., L. W. Uccellini, A. Mostek, and D. A. Keyser, 1984:Delineating mid and low level water vapour patterns in pre-convective environments using VAS moisture channels. Mon. Wea. Rev.,112, 2178–2198.

  • Rabier, F., E. Klinker, P. Courtier, and A. Hollingsworth, 1996: Sensitivity of forecast errors to initial conditions. Quart. J. Roy. Meteor. Soc.,122, 121–150.

  • Reed, R. J., 1955: A study of a characteristic type of upper-level frontogenesis. J. Meteor.,12, 226–237.

  • Sanders, F., and J. R. Gyakum, 1980: Synoptic-dynamic climatology of the “bomb.” Mon. Wea. Rev.,108, 1589–1606.

  • Shapiro, M. A., 1980: Turbulent mixing within tropopause folds as a mechanism for the exchange of chemical constituents between the stratosphere and troposphere. J. Atmos. Sci.,37, 994–1004.

  • Shutts, G. J., 1990: Dynamical aspects of the October storm, 1987: A study of successful fine-mesh simulation. Quart. J. Roy. Meteor. Soc.,116, 1315–1347.

  • Stoelinga, M. T., 1996: A potential-vorticity based study of the role of diabatic heating and friction in a numerically simulated baroclinic cyclone. Mon. Wea. Rev.,124, 849–874.

  • Thorncroft, C. D., B. J. Hoskins, and M. E. McIntyre, 1993: Two paradigms of baroclinic-wave life cycle behaviour. Quart. J. Roy. Meteor. Soc.,119, 17–55.

  • Thorpe, A. J., and C. H. Bishop, 1995: Potential vorticity and the electrostatic analogy Ertel–Rossby formulation. Quart. J. Roy. Meteor. Soc.,121, 1477–1495.

  • Uccellini, L. W., D. Keyser, K. F. Brill, and C. Wash, 1985: The President’s Day Cyclone of 18–19 February 1979: Influence of upstream trough amplification and associated tropopause folding on rapid cyclogenesis. Mon. Wea. Rev.,113, 962–987.

  • Weldon, R. B., and S. J. Holmes, 1991: Water vapour imagery: Interpretation and applications to weather analysis and forecasting. NOAA Tech. Rep. 57, NESDIS, Washington, DC, 213 pp.

  • Young, M. V., G. A. Monk, and K. A. Browning, 1987: Interpretation of satellite imagery of a rapidly developing cyclone. Quart. J. Roy. Meteor. Soc.,113, 1089–1115.

  • Ziemianski, M., 1994: Potential vorticity inversion. JCMM Internal Rep. 39, University of Reading, Reading, United Kingdom, 17 pp. [Available from JCMM, University of Reading, Whiteknights, Reading, Berkshire RG6 6AH, United Kingdom.].

  • View in gallery

  • View in gallery

    The Meteosat water vapor imagery in gray shades and the 315 K isentropic surface PV in contours at initial analysis time for 0000 UTC 24 Jan 1996. Dark shades represent dry air, light shades represent moist air. Contour interval is 0.5 pvu.

  • View in gallery

    Initial conditions for FA160295 and FA160295/R1. The Meteosat water vapor imagery in gray shades and the 315 K isentropic surface PV in white contours at initial analysis time for 0000 UTC 16 Feb 1995. Dark shades represent dry air, light shades represent moist air. (a) The original forecast (FA160295), (b) the FA160295/R1 forecast. Contour interval is 0.5 pvu.

  • View in gallery

    PV modification locations of FA160295/R1 forecast for 16 Feb 1995 case.

  • View in gallery

    Plots of the surface pressure differences for 1800 UTC 16 Feb 1995. (a) FA160295 (operational forecast) − verifying analysis, (b) FA160295/R1 − verifying analysis, (c) FA160295/R2 − verifying analysis, (d) FA160295/R3 − verifying analysis. Operational forecast and reruns are at T + 18. Contour interval is 2 mb.

  • View in gallery

    Tracks of the mean sea level pressure centers of the verifying analysis (UA16), rerun forecast FA160295/R1 (RR16), and the original forecast (FA16) shown in dark lines from 0000 UTC 16 Feb to 0000 UTC 17 Feb 1995.

  • View in gallery

    The PV modification locations of FA160295/R2 forecast for 16 Feb 1995 case.

  • View in gallery

    Initial conditions for FA160295/R2 and FA160295/R3. As in Fig. 2 but for (a) FA160295/R2 and (b) FA160295/R3, respectively.

  • View in gallery

    The PV modification locations of FA160295/R3 forecast on 16 Feb 1995 case.

  • View in gallery

    The Meteosat water vapor imagery in gray shades and PV on the 315 K isentropic surface in white contours at 0300 UTC 27 Apr 1992. Dark shades represent dry air, light shades represent moist air. (a) FA270492 forecast at T + 3. (b) FC260492 forecast at T + 15. Contour interval is 0.5 pvu.

  • View in gallery

    The PV modification locations for rerun forecasts in the 27 Apr 1992 case.

  • View in gallery

    As in Fig. 9 but for (a) FA270492/R1 and (b) FA270492/R2 forecasts, respectively, for T + 3 at 0300 UTC.

  • View in gallery

    Plots of the surface pressure differences for 1800 UTC 27 Apr 1992. (a) FA270492 − FC260492, (b) FA270492/R1 − FC260492, (c) FA270492/R2 − FC260492, (d) FA270492/R3 − FC260492. Contour interval is 2 mb. Note that the FC260492 forecast is for T + 30, whereas the FA270492 forecast and all related reruns are for T + 18.

  • View in gallery

    Initial conditions for FB280995 and FB280995/R1. The Meteosat water vapor imagery in gray shades and the 330 K isentropic surface PV in black contours at initial analysis time for 0600 UTC 28 Sep 1995. Dark shades represent dry air, light shades represent moist air. (a) The original forecast (FB280995), (b) FB280995/R1 forecast. Contour interval is 0.5 pvu.

  • View in gallery

    The PV modification locations of FB280995/R1 forecast for 28 Sep 1995 case.

  • View in gallery

    The PV modification locations of FB280995/R2 forecast for 28 Sep 1995 case.

  • View in gallery

    Plots of the surface pressure differences for 1800 UTC 29 Sep 1995. (a) FB280995 forecast − verifying analysis, (b) FB280995/R2 forecast − verifying analysis. Contour interval is 2 mb. The original forecast and FB280995/R2 are at T + 36.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 28 28 7
PDF Downloads 10 10 3

Sensitivity of Short-Range Weather Forecasts to Local Potential Vorticity Modifications

View More View Less
  • 1 Joint Centre for Mesoscale Meteorology, Department of Meteorology, University of Reading, Reading, United Kingdom
© Get Permissions
Full access

Abstract

A new method is described to interpret satellite water vapor (WV) imagery in dynamical terms using potential vorticity (PV) concepts. The method involves the identification of mismatches between the WV imagery and a numerical weather prediction model description of the upper-level PV distribution at the analysis time. These mismatches are usually associated with horizontal positioning errors in the tropopause location in the oceanic storm-track region in midlatitudes. The PV distribution is locally modified to minimize this mismatch, and PV inversion is carried out to provide dynamically consistent additional initial data with which to reinitialize the numerical forecast.

One of the advantages of using this method is that it is possible to generate wind and temperature data suitable for inclusion as initial data for numerical weather forecasts. By using PV additional data can be inferred that cannot otherwise be simply derived from the WV data. In this way dynamical concepts add considerable value to the WV imagery, which by themselves would probably not have as significant a forecast impact.

Several examples of the use of this method are given here including cases of otherwise poorly forecast North Atlantic cyclones. In cases where the analysis errors occur at upper levels of the troposphere, the method leads to a significant improvement in the short-range forecast skill. In general, it is useful in highlighting where forecast problems are arising.

* Current affiliation: Turkish State Meteorological Service, Ankara, Turkey.

The Joint Centre for Mesoscale Meteorology is supported by the U.K. Meteorological Office and the Department of Meteorology, University of Reading.

Corresponding author address: Dr. Meral Demirtas, Turkish State Met. Service, P.O. Box 401, Kalaba-Ankara, Turkey.

Email: tud@ecmwf.int

Abstract

A new method is described to interpret satellite water vapor (WV) imagery in dynamical terms using potential vorticity (PV) concepts. The method involves the identification of mismatches between the WV imagery and a numerical weather prediction model description of the upper-level PV distribution at the analysis time. These mismatches are usually associated with horizontal positioning errors in the tropopause location in the oceanic storm-track region in midlatitudes. The PV distribution is locally modified to minimize this mismatch, and PV inversion is carried out to provide dynamically consistent additional initial data with which to reinitialize the numerical forecast.

One of the advantages of using this method is that it is possible to generate wind and temperature data suitable for inclusion as initial data for numerical weather forecasts. By using PV additional data can be inferred that cannot otherwise be simply derived from the WV data. In this way dynamical concepts add considerable value to the WV imagery, which by themselves would probably not have as significant a forecast impact.

Several examples of the use of this method are given here including cases of otherwise poorly forecast North Atlantic cyclones. In cases where the analysis errors occur at upper levels of the troposphere, the method leads to a significant improvement in the short-range forecast skill. In general, it is useful in highlighting where forecast problems are arising.

* Current affiliation: Turkish State Meteorological Service, Ankara, Turkey.

The Joint Centre for Mesoscale Meteorology is supported by the U.K. Meteorological Office and the Department of Meteorology, University of Reading.

Corresponding author address: Dr. Meral Demirtas, Turkish State Met. Service, P.O. Box 401, Kalaba-Ankara, Turkey.

Email: tud@ecmwf.int

1. Introduction

Forecast errors arise from either analysis or model errors or both. Attempts to estimate the relative magnitude of these two error sources invariably show that analysis error is a major contributor (Arpe et al. 1985;Downton and Bell 1988). In the extratropics, especially over oceans or other regions of sparse data, the majority of forecast errors come from deficiencies in the operational analyses used as initial conditions for the forecast model. In addition, sensitivity studies have shown that the growth of errors or other small perturbations is generally localized in relatively small regions that vary in location from day to day, depending on the character of the flow. Rabier et al. (1996) stated that in the European Centre for Medium-Range Weather Forecasts (ECMWF) model, the largest forecast errors more often arise from errors in the initial conditions than from errors in the model formulation. It may be possible to improve the forecast quality by making a small but carefully chosen change in the analysis.

The accuracy of the analysis is a major factor determining the quality of the numerical weather prediction (NWP) guidance. Errors in the analysis come from errors in the background fields and an inadequate observational dataset. Deficiencies in the analysis can create errors in the location of key upper/low-level potential vorticity (PV) features and the consequent richly structured subsynoptic-scale features of the flow. As mentioned earlier, the quality of weather forecasting relies heavily on available and reliable data. Poor observational coverage and inaccurate data can significantly reduce forecast skill. One of the main ways of reducing forecast errors is to make best use of satellite data over the oceans. A technique for combining observations and dynamical concepts would be a powerful tool for the exploitation of satellite data and application of theories for day-to-day weather forecasting. The stratospheric PV intrusion contributes substantially to determining the depth and location of the associated surface cyclone. Therefore, a good representation of the structure and amplitude of these elements is a vital requisite for a good forecast. It is encouraging to note that the existence and evolution of tropopause level flow features that are seminal to surface development are evident in water vapor (WV) images (Appenzeller and Davies 1992; Young et al. 1987). The link between PV and WV distribution comes about through the twin properties of stratospheric air—high PV and low humidity. For example, there is a close relationship between the dry regions on the WV image and high PV values at the rear of a developing cyclone.

Based on using PV and WV images together (outlined in section 2) we have developed a new method (described in section 3) to monitor incipient errors in short-range weather forecasts and to provide dynamically consistent data to rerun potentially poor forecasts. Its use in short-range forecasting is illustrated by several case studies (section 4). A general discussion of the results and a set of open questions that may guide research to be conducted in the future will also be given (section 5).

2. PV–WV view of stratospheric intrusions, tropopause folds, and their contributions to cyclogenesis

One of the central issues of meteorology concerns the understanding of and ability to forecast the behavior of extratropical weather systems. With the use of synoptic examples and sensitivity experiments, processes contributing to the rapid development of extratropical cyclones have been identified in the scientific literature. Lately, the discussions have been extended based on potential vorticity concepts. These give a vivid view of how on the one hand upper-level and low-level processes, and on the other hand physical and dynamical processes, interact with each other during cyclonic developments.

The advantages of PV concepts to describe and understand atmospheric dynamics occur due to two main properties of PV, which are its conservation and its invertibility (Hoskins et al. 1985, hereafter HMR85). The conservation property enables one to identify and follow significant features in space and time, if the airflow is both frictionless and dry adiabatic. This is a good approximation for short time intervals and for airflows away from the earth’s surface and clouds. The invertibility property allows one to quantify the importance of PV features in terms of the strength of their associated circulations and temperature patterns so that their capacity to enhance the development of other features can be assessed (Davis and Emanuel 1991; Thorpe and Bishop 1995). It could also be particularly useful in providing dynamically consistent wind and temperature data to initialize numerical weather forecasts.

Dynamically meaningful information can be illustrated by displaying PV distributions on isentropic surfaces. In practice, a suitable isentropic surface for viewing PV is one that intersects the polar front jet since the anomalies responsible for many atmospheric developments are visible there. Such an isentrope will be in the lower troposphere in the subtropics and will rise through the polar front, emerging in the lower stratosphere on the poleward side of the jet. In winter the 315 K isentropic surface and in summer the 330 K isentropic surface typically lie above the 300-mb level at high latitudes, then intersect the tropopause at midlatitudes, and slope down to the 800-mb level at low latitudes. Another useful display is the height of the dynamical tropopause or the so-called “PV = 2 surface” (1 pvu unit = 10−6 K kg−1 m2 s−1). When the PV = 2 pvu height is very low, this means that the tropopause is close to the surface, thus bringing high PV values to low altitudes.

Upper-level trough/ridge patterns and jet streaks provide the divergence aloft needed for deepening of surface cyclones. They are also associated with distributions of PV, which in turn, have significant impact on the spinup of storms through the descent and horizontal advection of stratospheric air toward the cyclogenetic region. Attempts have been made to link the intrusion of the stratospheric dry air into upper and middle troposphere to cyclogenesis by using the conservation of PV. As stratospheric air descends into the troposphere, the air mass is stretched and static stability decreases. As a result of this, the absolute vorticity increases with respect to parcel trajectories as long as the stratospheric values of PV are conserved. By using the invertibility principle, HMR85 showed that a positive PV anomaly that extends downward from the stratosphere into the middle troposphere induces a cyclonic circulation that extends throughout the entire troposphere to earth’s surface.

An intrusion of dry stratospheric air can modify the outgoing radiance. This will be observed as a “dark” dry zone (sometimes referred to as “dry slot”) in the WV imagery indicating the dry air. In comparison with the upper-tropospheric air, stratospheric intrusions are rich in PV and ozone and deficient in water vapor (Appenzeller and Davies 1992). In midlatitudes a dry intrusion is marked by a tongue of high PV extending toward the low latitudes from the PV rich stratosphere. In their mature phase stratospheric PV tongues are observed to develop further and then either curve cyclonically to form a large-scale (1000–2000 km) vortex above the surface low (Young et al. 1987) or curve anticyclonically to form an elongated streamer of stratospheric air (Thorncroft et al. 1993). Some observational studies, such as Uccellini et al. (1985) and Keyser and Shapiro (1986), showed that narrow tongues (typically 100–200 km in width) of high PV originate within tropopause folds and could penetrate as low as 700 mb. Tropopause “folding” as explained by Reed (1955) can be thought of as a process for detaching and transporting stratospheric air from the main “reservoir” toward the middle troposphere along the axis of the jet streams. The role of tropopause folding in the exchange of air between the stratosphere and troposphere is observable by the associated large values of potential vorticity, ozone, and other chemical constituents (Shapiro 1980).

a. Signatures of cyclogenesis in WV imagery

The WV satellite imagery shows radiance distributions measured in the water vapor absorption band (5.7–7.1 μm) emitted by the earth–atmosphere system. It has an absorption band centered on 6.7 μm (Weldon and Holmes 1991). The WV imagery at 6.7 μm represents conditions over a broad layer of the atmosphere (Chesters et al. 1982). This band measures WV in the middle to the upper troposphere, generally between 300 and 500 mb (Fischer et al. 1981). The interpretation of the 6.7 μm image is relatively simple for dry regions since the dry midlevel air provides a distinct and sharply contrasting signal to any surrounding midlevel moisture (Petersen et al. 1984). Moreover, for dry upper-tropospheric conditions, the level of maximum radiance for the 6.7-μm channel is located at low levels. Therefore, distinct dry (warm) signatures on the WV image represent dry conditions not only near the 400-mb level, but also through a deep layer of the atmosphere extending down to at least 700 mb (Uccellini et al. 1985).

It has been shown that dry intrusions are always present when significant surface cyclogenesis was observed within a marked baroclinic zone. Sometimes, very narrow distinct dry tongues are associated with explosive deepening, for example, on 17 October 1984 when a decrease of central pressure of 24 mb in 6 h occurred off western Ireland, giving storm force winds (Young et al. 1987). The study of the explosive cyclogenesis of the “President’s Day Storm” by Uccellini et al. (1985) showed that the eastward movement of a marked dry tongue in the WV imagery was linked with pronounced surface development. They stated that this “dry air tongue” was associated with stratospheric air, identified by high PV, which descended along isentorpic trajectories to as low as 700 mb.

Böttger et al. (1975) and Bader et al. (1995) studied the origin and significance of the “cloud head” observed in the satellite images. Their statistical survey showed that all 22 cases of cloud heads were found to be followed by cyclogenetic developments. Shutts (1990) in a study of the Great October Storm showed, from the forecast of the 400-mb RH, the existence of a cloud head and a dry slot cutting in to the rear of the system. Evans et al. (1994) examined satellite images for 50 cyclogenesis events that occurred during the 1970s and 1980s, 46 of which were classified as explosive cyclones. The first category, called the “emerging cloud head,” is characterized by the formation of a cloud head of an “S-shaped” cirroform band associated with a troposphere-spanning baroclinic zone (i.e., polar front). As rapid deepening starts to take place, the cloud head gradually forms a comma cloud shape, with a dry slot becoming prominent upstream of the position of the surface low. During rapid deepening, the surface low remains near and equatorward of the cloud head (Evans et al. 1994).

b. Complementarity of WV images and PV distributions

A number of studies have been devoted to the relationship between observed radiance and PV distributions. Although there are number of difficulties relating the WV measurements to PV, considerable insight into the structure of subsynoptic-scale features can be gained by a qualitative comparison of these two datasets. Manney and Stanford (1987) postulated that using certain assumptions there should be a correspondence between the WV images and PV on an isentropic surface in the upper troposphere. They were only able to show a qualitative relationship between the two datasets. It was noted that a deep intrusion of stratospheric air is distinguished from tropospheric air by low concentrations of WV and high values of PV. Another qualitative study was carried out by Appenzeller and Davies (1992) via juxtaposition of Meteosat WV images with ECMWF PV maps. They showed that during stratospheric intrusions, the PV and WV imagery displays are complementary.

Recently, the use of WV and PV comparisons in the short-range operational forecasting were also studied (Mansfield 1996; Demirtas 1996). In spite of the difficulties in theoretical interpretation of WV data, encouraging similarities are found in qualitative comparisons between features in WV images and PV maps. Despite, the correspondence between the 6.7-μm WV band and PV being strong, it is not a one-to-one relationship. Nevertheless, in the polar front region the shape of the WV image is closely mirrored by the PV distributions. It should be noted that the exact relationship varies according to the particular synoptic situation.

Our studies also showed that tropospheric values of PV match well with the moist/cloudy regions of the WV imagery. Frontal cloud forms mostly within the air mass of subtropical origin and would therefore be expected to be associated with air of small tropospheric PV. In addition, latent heat release concentrates PV gradients aloft, shown by the marked edges of the PV distribution on, for example, the 315 K isentropic surface where the characteristic cyclonic comma clouds are seen.

The relationship between the PV and WV is considered to be reliable enough to provide a quality control of NWP model analyses and forecasts. An example of good match between model PV and WV imagery is given early in section 3. However, depending on the dynamical situation in some cases these relationships may not be seen so clearly or may cause misinterpretations.

  1. If there is a cutoff low (such that parcels are essentially trapped and circulate around for a long time), the related WV image will show dry regions mixed with the moist air.
  2. In the polar regions in winter the lower atmosphere is so cold that there is little contrast between regions of low tropopause heights (high PV) and surrounding areas even though the latter contain cloud (Mansfield 1996).
  3. Subtropical air may be warm enough to contribute to “dark regions” of the WV image, in spite of the high tropopause, if the upper troposphere is relatively dry. Therefore, dry slots in the WV image can appear in these air masses with little or no relation to the PV fields.

3. Using PV–WV relationship in short-range weather forecasting

In a forecasting context first finding the portion(s) of the flow domain most sensitive to initial conditions, and second improving the analysis in those regions can generate an improved analysis. The deficiencies in an analysis may well relate to the position and magnitude of the errors of upper-level components of the large-scale flow. Information on the sensitivity to these initial error fields in forecasts can be acquired by modifying the model’s initial upper-level PV features. Here the concept of making PV modifications is applied to some examples of poor forecasts. This work extends the use of PV and satellite WV images in studying cyclone dynamics itself, but also attempts to describe a new method for improving the accuracy of operational forecasting.

The data used in this study were taken from the U.K. Meteorological Office (hereafter UKMO) archived operational analyses and forecasts. The UKMO Unified Model (hereafter UM) is hydrostatic and uses spherical polar coordinates [for details, see Cullen (1993)]. UKMO uses the same model for operational weather prediction in global (0.8° × 1.25° resolution, 19 levels) and regional [also known as the Limited Area Model (LAM); 0.442° × 0.442° resolution, 19 levels, 229 × 132 points, rotated pole at 30°N, 160°E] configurations.

The LAM data used in this study have been analyzed using a display system called DISP at the Joint Centre for Mesoscale Meteorology at Reading University. The original LAM data have a horizontal resolution of about 50 km, and a vertical resolution, above the boundary layer, of about 1 km. The DISP data are interpolated to lower resolution with a horizontal resolution of about 100 km, and a vertical resolution of 100 mb. The basic data extracted from the LAM are geopotential height, temperature, RH, wind components, and vertical velocity. After interpolating these data on the DISP levels, other diagnostics are calculated from these basic fields, including the PV.

Forecasts contain fields at 3-hourly intervals from T + 0 to T + 48 (h). Four forecast runs (the 0000 UTC, 0600 UTC, 1200 UTC, and 1800 UTC) are available. Analyses are performed every 6 h. The T + 3 forecast fields are used as nominal intermediate analyses.

The Meteosat satellite data are taken from AUTOSAT-2 and interpolated on the model area at a resolution of 25.4 km along the 20°W meridian and 13.5 km across the meridian. The datasets provided from the WV images of Meteosat, a geostationary satellite of the European Space Agency situated at an altitude of 36 000 km above the equator on the Greenwich meridian, have a 5 km × 5 km resolution at the subsatellite point and about 8 km × 5 km resolution in midlatitudes.

a. Approach used in the PV–WV method

It has been noted before that WV images reveal some features that are important for cyclogenesis and qualitatively agree with PV fields. In light of the foregoing discussion it is clear that on the one hand a need exists to refine conventional analysis procedures, and on the other to include dynamical concepts such as PV. The objective here is to assess the contribution of errors in the analysis of the upper-level PV distribution to poor forecasts. The method involves the following sequence of operations.

  1. Quality control: Estimation of the closeness of fit between key upper-level PV features, as described by the forecast model analyses, and the WV images at the initial analysis time.
  2. Modification: Having identified key mismatched PV features, the model PV distribution is locally modified to provide a closer fit.
  3. Intervention: Inversion of the modified PV distribution at the initial time is used to derive dynamical and thermodynamical fields to initialize the forecast.
  4. Rerun: Use conventional forward prediction including the modified initial data.

This approach will isolate the contribution to the forecast error predominantly from the upper-level PV error in the model analysis and can provide insight on the role of upper-level effects in cyclogenesis. For example, a successfully modified forecast would lend credence to the hypothesis that upper-level effects are crucial.

1) Quality control

We compared the UKMO UM operational short-range forecast products of PV charts with the WV imagery on a daily basis. Comparisons of PV and WV images have the potential to show positional errors of PV contours as well as pointing out errors in the amplitude of the PV distribution (Mansfield 1996; Demirtas and Thorpe 1996). In cases of good analyses, which resulted in good forecasts, PV gradients were large over the distinct WV contrasts between dry and moist regions. For some cases, when these relationships were not seen, forecasts were not able to predict the intensity of the observed system.

In the case studies in section 4 we show several examples of mismatches between model PV and WV imagery but here we give an example of a good match between these two elements. Figure 1 illustrates various aspects of the relation between the PV and WV imagery. Here PV on the 315 K isentropic surface is superimposed on the WV image. Dark gray shows dry places, whereas light gray or white indicates moist zones. For example, around 48°–52°N and 30°–40°W, the PV maximum (about 7 pvu) coincides with dry regions of the WV imagery. This PV center is associated with a cyclonic flow at the surface. The WV imagery shows the dry intrusion and the PV distribution on the 315 K surface points to its dynamical nature. Taken together, they both confirm the stratospheric origin of the flow. This occasion also illustrates the good fit of the PV and dry air in the rear of a surface depression.

It also shows that PV contours (gradient) curving along the cloud edge on the forward side of the dry slot, near around 50°–60°N and 25°–40°W. The PV values (about 1 pvu) correspond well with the moist regions (or the developing “comma cloud”). Another distinct feature is over the southern part of the United Kingdom. In that area, dry air has mixed with the moist air during the maturing phase of the cyclone. The high PV, of about 6–7 pvu, is over the dry region. As the PV contours turn cyclonically, they take on the shape of the comma cloud. Other apparent correlations are that the tight PV gradient between the upper troposphere and the low stratosphere is over the region of the sharp transition zone between the dry and moist/cloudy air on the image. Although it is not shown in this paper, similar relations were also found with the PV = 2 pvu surface height chart and the WV image.

In summary, then, during an intrusion event the WV radiance and relevant PV distributions are complementary. The dark zone (or the dry slot) in the WV imagery is a vertically averaged, but horizontally finer-scale, visualization of the streamer, whereas the tongue of stratospheric flow on the isentropic surface is a coarser resolution visualization of the dynamically important PV tracer (Appenzeller and Davies 1992).

2) PV modifications

Misfits between WV imagery and PV charts can indicate a locally poor analysis. To correct this error, the next step is to modify the model analyzed PV distribution over the mismatched regions to produce a better fit with WV imagery (Demirtas and Thorpe 1996). The PV–WV mismatch suggests the need for repositioning of model PV features. If (low) high values of PV contours on isentropic surfaces are displaced relative to WV imagery dry slots (moist regions), they should be modified to coincide with them. This is done by either adding or subtracting PV anomalies such that the PV gradients are moved to better coincide with the transition between dry and moist air on WV imagery. Amplitudes of anomalies were chosen to achieve this transition but without introducing any new PV gradients away from WV transition zones.

By examining a range of isentropic and pressure levels, modifications were made to the three-dimensional PV distribution to provide as coherent a modified PV distribution as possible. (Note that no attempt was made here to maintain the mass-weighted volume integral PV during these modifications.) The modifications can best be visualized as a movement of the 3D PV contours typical of the tropopause in such a way that any comparison between WV imagery and any 2D surface distribution of PV is improved. Note that in the cases to be presented here only a single isentropic surface is shown by way of illustration of the full 3D modifications made.

These modifications were done in this study by an iterative process using the DISP system to assess visually the fit between WV and PV. This is a very effective method given the high acuity of the human eye and brain for pattern recognition. However, in future work it would be useful to be able to automate this iterative process. In addition it would be instructive to verify the local changes to the PV field from additional data not included in the original analysis, such as from field experiments such as FASTEX (see Joly et al. 1997).

3) Intervention and reruns

This part of the method involves the following sequence of operations.

i) PV Inversion: By using inversion of the complete, but locally modified, PV distribution, one then obtains the related rotational wind and temperature fields. In this study we use the nonlinear balance PV inversion technique developed by Davis and Emanuel (1991), and Ziemianski (1994). In regions where the PV has been locally modified the related wind and temperature values from the inversion are then assimilated into the model analysis in the form of so-called bogus observations. The prediction model will rapidly develop the implied divergent wind component.

ii) Reruns: The extra data provided for reruns generally have 100 km resolution in the horizontal. Due to the operational coding of the UM, there is a limitation on the total number of points in the bogus dataset that can be used in a rerun. For some reruns this was 500 points, whereas for others it was increased to 1000 points. The limitation on the bogus data puts some constraints on the modification area in the horizontal and vertical.

All data assimilation processes and data quality control are also applied on these extra data [for details see Bell et al. (1995); Ingleby and Parrett (1994)]. After this assimilation, another conventional forward prediction is performed with the revised fields. A problem is that by going through the data assimilation the model does not“fully” accept the revised fields as they are presented. This is a limitation in testing the full impact of our PV modifications that needs to be overcome in future work along the same lines.

After a rerun, PV maps are compared with WV imagery to check, retrospectively, the accuracy of the PV adjustments and also to make a further forecast quality control. This comparison indicates to what extent the model has accepted PV modifications. The reruns presented in this paper are the result of the first attempt to modify the PV distribution in each case.

The new method just described forms part of the total forecasting system illustrated as a flow chart in the appendix. Most parts of this system are already automated, except for quality control and PV modifications that have been manually performed. It is anticipated that in the future these will also be automated.

4. Examples of rerun forecasts

The PV–WV method has been applied to real cases (Demirtas 1996). In the following subsections several of these cases will be given. The UKMO original operational forecasts will be denoted by symbols like FA160295 or FB280995, where F denotes forecast, A denotes initialization at 0000 UTC, B at 0600 UTC, C at 1200 UTC, and so on, and where remaining symbols denote the date of initialization. For example, FA160295 will mean the forecast initialized at 0000 UTC on 16 February 1995. Reruns with modified initial conditions will similarly be denoted as FA160295/R1, FA160295/R2, and so on.

a. First case study: 16 February 1995

1) Synoptic situation and operational forecast

We study here the rapidly deepening cyclone, which developed from a small wave, situated near 40°–45°N, 35°–45°W by 1200 UTC on 15 February 1995. From 1800 UTC on the 15th to 1800 UTC on the 16th, the central surface pressure of the low decreased from 1007 mb to 983 mb. This means that pressure deepening was 1 mb h−1 for 24 h, the threshold for a so-called bomb event (Sanders and Gyakum 1980). The associated weather observed over the United Kingdom on 16 February 1995 can be given in summary as rainy, heavy at times particularly from late afternoon onward, very strong gusty winds, reaching gale force in some places.

The forecast from 0000 UTC on 16 February 1995 (hereafter FA160295) took the system across the southern part of the United Kingdom with little deepening, whereas the actual low tracked farther north and it was about 8 mb deeper than in FA160295. It also failed to forecast significant precipitation over many parts of the United Kingdom.

2) Quality control

The first step is to identify WV imagery key patterns and related model PV features at analysis time. Figure 2a illustrates an example of an analysis where the PV on the 315 K isentropic surface appears to be misplaced to the north relative to the dry area on WV imagery. The position of stratospheric high PV (6 pvu), near 52°N, 30°W, was situated slightly north of the main dry region (48°–50°N, 20°–28°W). The dry slot and the cloud head suggest a much more intense system than is indicated by the relatively undistorted PV contours. Recall from Fig. 1 that typically the PV over the dry region has a much richer structure than in Fig. 2a. Note also that the PV distribution over the moist regions does not capture well the developing cloud head shape that is situated at 48°–52°N, 25°–15°W at 0000 UTC (Fig. 2a). Another apparent misfit was that the tight PV gradient of the tropopause was not over the region of the sharp transition zone between the dry and moist/cloudy air on the image.

3) PV modifications

Having identified mismatched PV features, we then modify the model analysis of the PV distribution to provide a closer fit. In the first experiment, PV features were modified over the dry and moist regions of the WV image. In the subsequent experiments PV modifications were partitioned in order to evaluate the impact of each PV feature separately: rerun 2 is based on the modifications only over the moist region, and rerun 3 is based on the modifications over the dry region (Tables 1 and 2 give other details of the reruns).

4) Rerun 1: PV modifications over the dry and moist regions

Local PV modifications were made over the mismatched regions shown in Fig. 3, using the relationship between the WV image and the PV distribution. This involves moving PV contours horizontally. This results in an increase of upper-tropospheric PV values over the dry slot of about 1 pvu and a decrease of about 1 pvu of lower-tropospheric PV values over the “cloud head” region.

The initial PV distribution for this rerun is presented in Fig. 2b. Comparisons of Fig. 2a with Fig. 2b show that the modified PV features are more richly structured in Fig. 2b than the PV of FA160295 in Fig. 2a. This time dynamically significant features match with the WV imagery; for example, the modified distribution better represents the PV gradient over the dry and moist regions. Another point is that PV over the cloud head region in Fig. 2b matches better with the WV imagery than the PV in Fig. 2a. These are regions where the new PV distribution is still misaligned with WV imagery. This is due to the data assimilation process acting on the new data and no attempt was made here to further correct the distributions.

By using PV inversion the modified wind and temperature fields were obtained. These dynamically consistent new data were then used to provide additional initialization data allowing the rerun of the FA160295 (hereafter FA160295/R1). A 48-h forecast starting from the modified initial state at 0000 UTC 16 February (FA160295/R1) yields a forecast with an upper PV pattern closer to the WV imagery. The surface development in this rerun bears a closer resemblance to the analysis. The surface pressure difference between the FA160295/R1 and the analysis is about 1.8 mb at the start of the forecast. We made intercomparisons of the verifying analysis, operational forecast (FA160295), and FA160295/R1. When the low-pressure system arrived over the United Kingdom at 1800 UTC, it was noted that FA160295 was predicting a central pressure minimum 7.6 mb higher than the verifying analysis (Fig. 4a), whereas FA160295/R1 was 3 mb higher than the verifying analysis (Fig. 4b). Comparing Figs. 4a and 4b implies that FA160295/RI compensated for most of the lack of surface development in FA160295. Figure 5 presents tracks of the surface cyclone in the analysis and as predicted in FA160295 and FA160295/R1 showing the improvement.

Radar images were also compared with this rerun’s precipitation prediction (not shown here), and it was found that this run captured the location of the main rainfall regions better than the original operational forecast [see Demirtas (1996) for more details].

Taking into account the fact that PV was modified over a small area and additional data provided with a coarse resolution, results of FA160295/R1 are rather encouraging. It shows that the PV–WV method made substantial impacts on the forecast in question.

5) Rerun 2: PV modifications over the cloud head region

This rerun forecast (hereafter FA160295/R2) attempts to incorporate the contribution of PV modifications made only over the cloud head region. The FA160295/R2 forecast evaluates how strongly these PV features contribute to the surface development. Figure 6 shows the area where upper-level PV modifications were made.

Figure 7a depicts the modified PV (at 48°–53°N, 12°–25°W) superimposed over the associated WV imagery. Although the modifications are not as large as in the FA160295/R1, the impact of the FA160295/R2 is still seen on the surface development. An intercomparison of the original forecast (FA160295), the analysis, and the FA160295/R2 forecast was made. The surface central pressure differences between FA160295/R2 and the verifying analysis are presented in Fig. 4c. It shows that FA160295/R2 resulted in a surface central pressure minimum 6.3 mb higher than in verifying the analysis, whereas it is 2 mb lower than in FA160295.

6) Rerun 3 forecast: PV modifications over the dry slot

This rerun forecast (hereafter FA160295/R3) elucidates how upper-tropospheric PV modifications over the WV dry region affect the cyclone development. Figure 8 presents the area where upper-level PV modifications were made. Figure 7b shows the modified PV (50°–53°N/20°–30°W) superimposed over the WV imagery.

The impact of upper-level PV modifications over the dry region was assessed by making some comparisons with respect to the verifying analysis and FA160295. Fig. 4d presents central pressure differences between FA160295/R3 and the verifying analysis. The maximum difference is about 5.3 mb with respect to the verifying analysis, whereas it is 3.7 mb with respect to FA160295. When Fig. 4d was compared with Figs. 4b and 4c, it was noted that PV modifications over the dry region make more impact than those over the moist region. Note also that this rerun has PV modifications over a much smaller area than the others. It indicates that the PV modifications over the dry region make more impact than the PV modifications over the moist region. In light of these results, it is clear that to obtain a better forecast, PV needs to be modified both over the dry and moist regions.

b. Second case study: 27 April 1992

1) Synoptic situation and operational forecast

A cold frontal wave was situated near 40°–50°N, 20°–30°W by 0000 UTC on 27 April 1992. The cold front wave was first identified on 25 April 1992, as a slow moving system about 1000 km southeast of Newfoundland (Browning et al. 1995). The wave moved in an easterly direction with slight development before deepening 12 mb in 12 h after 1200 UTC 27 April. The low tracked east-northeastward through the English Channel during the night of 27–28 April, generating many hours of rain over southern England (Hewson 1993).

Short-range forecasts valid at 0000 UTC on 28 April 1992 predicted incorrectly the position of the low near the United Kingdom by up to 300 km. Forecasts of central pressure ranged from 1000 to 1010 mb (Hewson 1993). A study of this FRONTS’92 IOP3 case revealed that earlier forecasts (0000 UTC and 1200 UTC runs from 26 April 1992) were better than the subsequent forecasts in predicting the synoptic situation at 1800 UTC on the 27th. The later forecast (0000 UTC run from 27 April 1992) was not as successful as the runs from the 26th. Comparisons of hand and model analyses showed that the model analyses were also poor in simulating what happened as described by additional aircraft dropsonde observations taken on this occasion.

In this case study, the method described in section 3 was applied in a slightly different manner. Here we transplant the PV distribution predicted after 12 h in the good model run into the analysis of the subsequent model run, which operationally gave a poor forecast. This application, in some respects, is similar to the diagnostic study of Fehlmann and Davies (1997). Their approach is also founded on a PV perspective. After identifying key upper-level PV errors at the forecast time, by using their Lagrangian retrodiction back to initial analysis time in a process also involving PV inversion, they performed a rerun based on the revised state.

2) Quality control

The quality of the analysis used for the 0000 UTC 27 April forecast (hereafter FA270492) was assessed by comparing its model analysis of PV charts with the WV imagery. It was noted that the PV distribution over the moist regions did not capture well the developing cloud head shape that was situated at 45°–50°N, 25°–15°W at 0300 UTC (Fig. 9a). Another apparent misfit is the position of stratospheric high PV (at 45°–47°N, 30°–32°W), which was situated slightly north of the main dry region (45°–50°N, 30°–35°W). The PV contours have not properly captured the shape and position of this feature.

The previous forecast from the 1200 UTC 26 April (hereafter FC260492) seemed to capture the dry slot better than the forecast from 0000 UTC 27 April. This earlier forecast FC260492 has larger PV over the dry region (Fig. 9b). Over the moist region, both of the forecasts have not fully captured the cloud head shape. This probably implies that for this case, the magnitude and position of PV contours are more important over the dry region than the moist region. Comparisonsof the PV of FC260492 and FA270492 suggest that FC260492 can be taken as the more accurate representation in order to readjust the PV distribution at analysis time of the poorer operational forecast FA270492.

3) PV modifications

The role of the upper-level PV of the FC260492 in providing a better forecast can be examined by “transplanting its PV” at T + 15 into the analysis for FA270492 and comparing this rerun with the original poor forecast FA270492 and the good forecast FC260492.

Over the moist region, FC260492 and FA270492 have similar PV structures. Therefore, the PV modifications were focused on the dry slot. The PV modification area is the same for all reruns, but there are some differences between the bogus data used. (Horizontal locations of modifications are shown in Fig. 10.) Reruns performed are the following.

  1. In the first experiment R1, the upper-level PV of the FC260492 was inverted and its wind and temperature transplanted into the FA270492.
  2. In the second experiment R2, instead of using wind and temperature fields from the PV inversion they were taken directly from the data of FC260492.
  3. In the third experiment R3, the upper-level PV of the FC260492 was increased by the order of 1 pvu (over the locations shown in Fig. 10). After the PV inversion process, the new data were transplanted into the FA270492 (Tables 3 and 4 give other details of the reruns).

4) Rerun 1

Wind and temperature data, deduced from the PV inversion of FC260492, were put into the FA270492 initial analysis. The reconfigured PV pattern on the 315 K isentropic surface derived from the initial analysis is depicted in Fig. 11a. The intention was to place PV gradients over the dry region to coincide with Fig. 9b. But the rerun gave a pattern in between those in Figs. 9a and 9b. This rerun (hereafter FA270492/R1) shows that the UM assimilation did not “fully” accept the additional initialization data.

The modified upper-level PV directly intensifies the surface development. To further illustrate this point we consider some aspects that arise directly from an intercomparison of the good forecast (FC260492), the poor forecast (FA270492), and the rerun forecast (FA270492/R1). Figure 12a depicts surface pressure differences between the poor forecast (FA270492) and the good forecast (FC260492) for 1800 UTC on 27 April 1992. The FA270492 18-h forecast predicts central surface pressure 9.6 mb higher than the 30-h FC260492. After upper-level modifications FA270492/R1 shows more deepening than in FA270492. Figure 12b presents pressure differences between this rerun and the verifying forecast (FC260492). The maximum difference is about 6 mb. When Figs. 12a and 12b were compared, it was noted that FA270492/R1 is better than FA270492 in terms of reducing the gap between earlier and later forecasts. Results of FA270492/R1 could have been better if the model had accepted fully the additional data that were based on the PV modifications.

5) Rerun 2

In this rerun (hereafter FA270492/R2), since the good forecast was known and its data were available, wind and temperature data were provided from FC260492 rather than from the PV inversion of FC260492. Hence, FA270492/R2 takes into account both the rotational and divergent parts of the wind. The aim of FA270492/R2 is to evaluate and compare the use of the total wind for making reruns, because the wind data provided from the PV inversion model includes only the rotational part of the wind.

The modified PV pattern on the 315 K isentropic surface is presented in Fig. 11b: note the marked similarity with the PV pattern for FA270492/R1 in Fig. 11a. It shows that providing the data either from PV inversion or from the model total wind does not make much difference to the PV distribution. Surface pressure differences between FA270492/R2 and the verifying forecast (FC260492) are about 7 mb (Fig. 12c). When it is compared with FA270492, this rerun makes central pressure minimum 5 mb lower than in FA270492. Similarities between Figs. 12b and 12c are noticeable. It implies that using total wind or only the rotational part of the wind does not make a remarkable difference to the forecast.

6) Rerun 3

This rerun (hereafter FA270492/R3) is similar to FA270492/R1 except that the PV of FC260492 over the dry area was increased by about 1 pvu. One of the aims of doing this is to compensate for the diminution of the PV modifications that occur due to the data assimilation process following implantation.

The surface central pressure value of this rerun is similar to the others at the initial time. This rerun was also compared with the verifying forecast (FC260492) for 1800 UTC 27 April 1992. Figure 12d shows pressure differences between the two mentioned forecasts. FA270492/R3 predicts central pressure minimum 6 mb higher than in the verifying forecast. We also compared this rerun with the poor forecast; central pressure minimum in FA270492/R3 is 5 mb deeper than in FA270492. Results implied that increasing the upper-level PV by 1 pvu contributed to the surface development more than the other reruns of this case study, even though the data assimilation scheme “smoothed out” some of the “new data.”

c. Third case study: 28 September 1995

1) Synoptic situation and operational forecast

A small depression moved northward from the subtropics. It later decayed into a trough and was steered to the southwest of the United Kingdom in the flow on the forward side of an upper ridge, downstream of a complex pre-existing low. The analysis showed that there was a high-pressure system and no rain over the United Kingdom at 1800 UTC 29 September 1995.

The forecast run from 0600 UTC 28 September (hereafter FB280995) wrongly predicted the existence of the low-pressure system and rainfall over the United Kingdom. Other forecasts from 27 and early on 28 September 1995, valid for 1200 UTC 29 September, incorrectly brought rain from the Atlantic over southern England. This poor precipitation forecast for the southern United Kingdom in the global and the LAM runs was not corrected until the forecast run from 1200 UTC 28 September.

2) Quality control

The quality of the initial analysis of the FB280995 forecast was assessed by comparing its model analysis of PV with the WV imagery. The 330 K isentropic surface1 PV distribution was compared with the WV image at the start of the forecast (Fig. 13a). It was noted that the PV contours had a very slight misfit with the dry and moist/cloudy regions of the WV image (50°N, 30°–45°W). A “trough” of high PV (compared to its background) is located over the isolated dry area. It was expected that the base of the PV trough should coincide with the location of the driest area. There was a slight eastward shift of the PV relative to the WV dry slot. Over the moist/cloudy regions of the WV image, the associated PV contours did not capture well the shape of these regions.

3) PV modifications

In this case study, the method described in section 3 was applied in a slightly different manner. The method was used to assess the role of the upper-level and low-level PV of FB280995 in producing a poor forecast. In the first experiment R1, upper-level PV was modified over the dry and moist regions. In the second experiment R2, PV was modified only at low levels (Tables 5 and 6 give other details of the reruns).

4) Rerun 1

This rerun (hereafter FB280995/R1) is based on PV modifications made using guidance from the WV imagery. The aim is to examine how the forecast will respond to the local PV modifications at upper levels. In Fig. 14 the locations of PV modifications are shown with crosses. They highlight the places where the WV image and the PV distribution do not match well.

The PV distribution of FB280995/R1 was compared with the WV imagery at 0600 UTC 28 September. There were no substantial misfits between them (Fig. 13b). FB280995/R1, which was based on this modification, did not make a noticeable impact on the FB280995. Comparing surface charts of FB280995/R1 and FB280995, it was noted that the surface pressure patterns were very similar. The difference in the surface pressure minimum is about 1 mb, FB280995/R1 forecast, leading to an increase in the central pressure by 1 mb in 36-h forecast. The influence of upper-level PV modifications on the surface development was found to be very small.

5) Rerun 2

It is of interest to investigate the role of low-level PV modifications in the surface development. The dominant feature at low levels is the zone of high PV values that is situated immediately north of the warm front, with a maximum of 1.8 pvu at 1000 mb near the cyclone center. Many studies have suggested that similar features in other midlatitude cyclones are the result of latent heating within a frontal ascent region (Davis 1992; Stoelinga 1996). The extant studies on the predictability of errors in NWP systems have several avenues. For example, singular vector analysis of idealized simulations of cyclogenesis (Rabier et al. 1992) and operational ensemble predictions (Molteni et al. 1996) both highlight that the amplitude of the dominant singular vectors are located in the lower troposphere. Ameliorating the deficiencies at these elevations can induce PV features at upper levels.

Our study of the low-level PV distributions indicated that there are local maxima of 1.8 pvu at 1000 mb, 1.6 pvu at 900 mb, and 1.2 pvu at 800 mb levels at 30°–40°W, 50°–55°N. It is not common to see such high PV maximum at these low levels, especially when there is a high-pressure system at the surface. Climatological distributions of PV also suggest that at midlatitudes in the lower troposphere values of PV are more typically 0.3–0.5 pvu (Hoskins 1990). High values of PV at low levels might have been created by diabatic processes. The FB280995 started with these large PV anomalies, and these might have contributed to the later development. Therefore, the objective of this rerun (hereafter FB280995/R2) was to assess the contribution to the system of modifying the low-level PV distribution. The related PV anomalies on the 900-mb, 800-mb, 700-mb, and 600-mb levels were decreased by approximately 1 pvu. In Fig. 15 the areas of the local PV modifications are shown by crosses.

In this experiment the surface pressure deepening was reduced noticeably. The central pressure difference between the FB280995 and the FB280995/R2 is about 8.3 mb at 1800 UTC on the 29th. (Recall that the FB280995 predicted surface pressure maxima 10 mb lower than the analysis.) Comparing Fig. 16a with Fig. 16b indicates that while the operational forecast is substantially different compared to the verifying analysis, these differences are quite small in FB280995/R2.

Results of FB280995/R2 showed that the forecast is more sensitive to the PV modifications at lower levels than at upper levels. The upper-level PV anomaly may have been of secondary importance in the synoptic-scale development of the cyclone in FB280995. Compared to FB280995 and to FB280995/R1, FB280995/R2 with low-level PV modifications gave the most beneficial impact, producing significant improvements to the surface pressure and the rainfall over Ireland. Although not shown here, the FB280995/R2 held back the trough, and rain was consequently only forecast for the southeast of Ireland.

5. Concluding remarks

This study highlights that this new method is useful for the estimation of errors in the initial stages of short-range forecasts. The application of PV concepts sheds light on the location of the initial error and provides dynamically consistent data to rerun a potentially poor forecast. It also gives dynamical insight on the nature of the space–time evolution of initial errors. The method also opens new avenues on practical strategies for remedying deficiencies in the analysis. The approach presented here can also help in the design of adaptive observational strategies. The PV errors at tropopause level can influence the flow field throughout the troposphere, and their removal may only require in situ data.

The main conclusions are as follows.

  1. The method presented in this paper can be used to improve short-range weather forecasts. In effect the PV and WV image-based approach serves to pinpoint the location and subsynoptic-scale structure of the upper-level features. In cases where the analysis errors occur in the upper troposphere the method can lead to a significant improvement in the short-range forecast skill. Upper/low-level PV modifications emphasize the importance of upper- and low-level data for NWP. Focusing on predictability studies in terms of PV features, one might better understand the nature of the NWP sensitivity to errors in the initial conditions.
  2. During dry intrusions the WV imagery and PV on an isentropic surface are complementary. The dry slot in the WV imagery visualizes the streamer, whereas the tongue of stratospheric flow on an isentropic surface depicts the dynamically important PV tracer.
  3. Comparisons of Meteosat WV images and model PV on an isentropic surface, or the height of the PV = 2 pvu dynamical surface, can be used as a “quality control” tool in operational weather forecasting to validate the model results. Mismatches between the WV image and the model PV can highlight potential errors in the initial conditions of the model.
  4. In cases where the errors were mainly at low levels the method was used as a diagnostic tool. The PV anomalies at low levels, generated by local diabatic process, are situated ahead of the upper-level PV anomaly and can noticeably influence the system development. This sometimes may cause enhanced pressure deepening at the surface and if poorly analyzed can lead to a poor forecast. In the experiment in which the magnitude of the diabatically produced low-level PV anomalies was decreased by 1 pvu, the surface deepening was reduced by around 8 mb, improving substantially the accuracy of the forecast.
  5. There are some limitations and problems in the process of putting additional data into the UM to do with data assimilation. They have some impact on the results and need to be addressed in future studies along these lines.
In a forecasting context, it is appropriate to note that aspects of these PV features are discernable in the WV images. The imagery is available quasi-continuously and this poses the challenge to devise schemes to 1) relate the WV image directly to the PV distribution, and 2) perform short range “forecast-analysis-PV modifications and PV inversion-rerun” cycles to incorporate the information in quasi–real time. One could argue that forecast improvements might also accrue by more accurately assimilating the WV image data directly into the model initialization. It is our contention that by using PV concepts we can infer additional data on the atmospheric flow and temperature that cannot be derived simply from the WV data. In this way dynamical concepts add considerable value to the WV data, which by themselves would probably not have as significant a forecast impact.

One can speculate whether such wind and temperature modifications could be automatically produced if the data assimilation system included a cost function to minimize the difference between the PV of the first guess and observations. It is at present unclear whether this will work as the link between water vapor content and the location of the tropopause is, from the prediction model viewpoint, indirect. Therefore changes to the WV content at the analysis time are likely to take a long time to filter through into consistent wind and temperature changes associated with positional alterations to the tropopause location. The power of the technique used in this paper is that these changes are introduced at analysis time. When new data assimilation systems are introduced it will hopefully be possible to assess these questions in a more systematic way.

Acknowledgments

Some parts of this paper are taken from the first author’s Ph.D. thesis at the University of Reading. The authors would like to thank Julian Heming and Humphrey Lean for assisting with rerunning the operational forecast, and also Andy Macallan, Dave Hennings, Nigel Roberts, and Tim Hewson for assistance with other data sources and programs. We would like to also thank Andy A. White, Michael E. McIntyre, Sid Clough, and Douglas Mansfield for useful discussions on this work. The first author is grateful to the Turkish Ministry of Education for a grant to support the original thesis work. Comments from the referees were very helpful in improving the exposition.

REFERENCES

  • Appenzeller, C., and H. C. Davies, 1992: Structure of stratospheric intrusions into the troposphere. Nature,358, 570–572.

  • Arpe, K., A. Hollingsworth, M. S. Tracton, A. C. Lorenc, S. Uppala, and P. Kallberg, 1985: The response of numerical weather prediction systems on the FGGE level IIb data. Part 2: Forecast verifications and implications for predictability. Quart. J. Roy. Meteor. Soc.,111, 67–101.

  • Bader, M. J., G. S. Forbes, J. R. Grant, R. B. E. Lilley, and A. J. Waters, 1995: Images in Weather Forecasting: A Practical Guide for Interpreting Satellite and Radar Imagery. Cambridge University Press, 499 pp.

  • Bell, R. S., B. Macpherson, D. Robinson, and P. Andrews, 1995: Technical details of the Unified Model data assimilation scheme. Unified Model Documentation Paper P3, 53 pp. [Available from United Kingdom Meteorological Office, Bracknell, Berkshire RG12 252, United Kingdom.].

  • Böttger, H., M. Eckardt, and U. Katergiannakis, 1975: Forecasting extratropical cyclones with hurricane intensity using satellite information. J. Appl. Meteor.,14, 1259–1265.

  • Browning, K. A., S. A. Clough, C. S. A. Davitt, N. M. Roberts, T. D. Hewson, and P. G. W. Healy, 1995: Observations of the mesoscale substructure in the cold air of a developing frontal cyclone. Quart. J. Roy. Meteor. Soc.,121, 1229–1254.

  • Chesters, D., L. Uccellini, and A. Mostek, 1982: VISSR Atmospheric Sounder (VAS) simulation experiment for a severe storm environment. Mon. Wea. Rev., 110, 198–216.

  • Cullen, M. J. P., 1993: The Unified Forecast/Climate Model. Meteor. Mag.,122, 81–94.

  • Davis, C. A., and K. A. Emanuel, 1991: Potential vorticity diagnostics of cyclogenesis. Mon. Wea. Rev.,119, 1929–1953.

  • Demirtas, M., 1996: Sensitivity of short-range forecasts to local potential vorticity modifications. Ph.D. thesis, University of Reading, Reading, United Kingdom, 227 pp. [Available from Department of Meteorology, University of Reading, Whiteknights, Reading, Berkshire RG6 6AH, United Kingdom.].

  • ——, and A. J. Thorpe, 1996: A new PV-based method for short-range weather forecasting quality control and intervention. Preprints, Seventh Conf. on Mesoscale Processes, Reading, United Kingdom, Amer. Meteor. Soc., 396–398.

  • Downton, R. A., and R. S. Bell, 1988: The impact of analysis differences on a medium range forecast. Meteor. Mag.,117, 279–284.

  • Evans, M. S., D. Keyser, L. F. Bosart, and G. Lackmann, 1994: A satellite-derived classification scheme for rapid maritime cyclogenesis. Mon. Wea. Rev.,122, 1382–1416.

  • Fehlemann, R., and H. C. Davies, 1997: Misforecasts of synoptic systems: Diagnosis via potential vorticity retrodiction. Mon. Wea. Rev.,125, 2247–2264.

  • Fischer, H., N. Eigenwilling, and H. Muller, 1981: Information content of Meteosat and Nimbus/THIR water vapor channel data: Altitude association of observed phenomena. J. Appl. Meteor.,20, 1344–1352.

  • Hewson, T., 1993: The FRONTS ’92 experiment: A quicklook atlas. JCMM Internal Rep. 15, University of Reading, Reading, United Kingdom, 176 pp. [Available from JCMM, University of Reading, Whiteknights, Reading, Berkshire RG6 6AH, United Kingdom.].

  • Hoskins, B. J., 1990: Theory of extratropical cyclones. Extratropical Cyclones. The Erik Palmén Memorial Volume, E. O. Holopainen, Ed., Amer. Meteor. Soc., 63–80.

  • ——, M. E. McIntyre, and A. W. Robertson, 1985: On the use of and significance of isentropic potential vorticity maps. Quart. J. Roy. Meteor. Soc.,111, 877–948.

  • Ingleby, N. B., and C. A. Parrett, 1994: Quality control of atmospheric data. Unified Model Documentation Paper P32, UK Meteorological Office, Bracknell, United Kingdom, 36 pp. [Available from United Kingdom Meteorological Office, Bracknell, Berkshire RG12 252, United Kingdom.].

  • Joly, A., and Coauthors, 1997: Definition of the Fronts and Atlantic Storm-Track Experiment (FASTEX). Bull. Amer. Meteor. Soc.,78, 1917–1940.

  • Keyser, D., and M. A. Shapiro, 1986: A review of the structure and dynamics of upper level frontal zones. Mon. Wea. Rev.,114, 452–499.

  • Manney, G. L., and J. L. Stanford, 1987: On the relation of 6.7 μm water vapour features to isentropic distributions of potential vorticity. Quart. J. Roy. Meteor. Soc.,113, 1048–1057.

  • Mansfield, D., 1996: The use of potential vorticity as an operational forecast tool. Meteor. Appl.,3, 195–210.

  • Molteni, F., R. Buizza, T. N. Palmer, and U. Petroliagis, 1996: The ECMWF ensemble prediction system: Methodology and validation. Quart. J. Roy. Meteor. Soc.,122, 73–119.

  • Petersen, R. A., L. W. Uccellini, A. Mostek, and D. A. Keyser, 1984:Delineating mid and low level water vapour patterns in pre-convective environments using VAS moisture channels. Mon. Wea. Rev.,112, 2178–2198.

  • Rabier, F., E. Klinker, P. Courtier, and A. Hollingsworth, 1996: Sensitivity of forecast errors to initial conditions. Quart. J. Roy. Meteor. Soc.,122, 121–150.

  • Reed, R. J., 1955: A study of a characteristic type of upper-level frontogenesis. J. Meteor.,12, 226–237.

  • Sanders, F., and J. R. Gyakum, 1980: Synoptic-dynamic climatology of the “bomb.” Mon. Wea. Rev.,108, 1589–1606.

  • Shapiro, M. A., 1980: Turbulent mixing within tropopause folds as a mechanism for the exchange of chemical constituents between the stratosphere and troposphere. J. Atmos. Sci.,37, 994–1004.

  • Shutts, G. J., 1990: Dynamical aspects of the October storm, 1987: A study of successful fine-mesh simulation. Quart. J. Roy. Meteor. Soc.,116, 1315–1347.

  • Stoelinga, M. T., 1996: A potential-vorticity based study of the role of diabatic heating and friction in a numerically simulated baroclinic cyclone. Mon. Wea. Rev.,124, 849–874.

  • Thorncroft, C. D., B. J. Hoskins, and M. E. McIntyre, 1993: Two paradigms of baroclinic-wave life cycle behaviour. Quart. J. Roy. Meteor. Soc.,119, 17–55.

  • Thorpe, A. J., and C. H. Bishop, 1995: Potential vorticity and the electrostatic analogy Ertel–Rossby formulation. Quart. J. Roy. Meteor. Soc.,121, 1477–1495.

  • Uccellini, L. W., D. Keyser, K. F. Brill, and C. Wash, 1985: The President’s Day Cyclone of 18–19 February 1979: Influence of upstream trough amplification and associated tropopause folding on rapid cyclogenesis. Mon. Wea. Rev.,113, 962–987.

  • Weldon, R. B., and S. J. Holmes, 1991: Water vapour imagery: Interpretation and applications to weather analysis and forecasting. NOAA Tech. Rep. 57, NESDIS, Washington, DC, 213 pp.

  • Young, M. V., G. A. Monk, and K. A. Browning, 1987: Interpretation of satellite imagery of a rapidly developing cyclone. Quart. J. Roy. Meteor. Soc.,113, 1089–1115.

  • Ziemianski, M., 1994: Potential vorticity inversion. JCMM Internal Rep. 39, University of Reading, Reading, United Kingdom, 17 pp. [Available from JCMM, University of Reading, Whiteknights, Reading, Berkshire RG6 6AH, United Kingdom.].

APPENDIX

A Flow Chart for the New Method

Fig. 1.
Fig. 1.

The Meteosat water vapor imagery in gray shades and the 315 K isentropic surface PV in contours at initial analysis time for 0000 UTC 24 Jan 1996. Dark shades represent dry air, light shades represent moist air. Contour interval is 0.5 pvu.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 2.
Fig. 2.

Initial conditions for FA160295 and FA160295/R1. The Meteosat water vapor imagery in gray shades and the 315 K isentropic surface PV in white contours at initial analysis time for 0000 UTC 16 Feb 1995. Dark shades represent dry air, light shades represent moist air. (a) The original forecast (FA160295), (b) the FA160295/R1 forecast. Contour interval is 0.5 pvu.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 3.
Fig. 3.

PV modification locations of FA160295/R1 forecast for 16 Feb 1995 case.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 4.
Fig. 4.

Plots of the surface pressure differences for 1800 UTC 16 Feb 1995. (a) FA160295 (operational forecast) − verifying analysis, (b) FA160295/R1 − verifying analysis, (c) FA160295/R2 − verifying analysis, (d) FA160295/R3 − verifying analysis. Operational forecast and reruns are at T + 18. Contour interval is 2 mb.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 5.
Fig. 5.

Tracks of the mean sea level pressure centers of the verifying analysis (UA16), rerun forecast FA160295/R1 (RR16), and the original forecast (FA16) shown in dark lines from 0000 UTC 16 Feb to 0000 UTC 17 Feb 1995.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 6.
Fig. 6.

The PV modification locations of FA160295/R2 forecast for 16 Feb 1995 case.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 7.
Fig. 7.

Initial conditions for FA160295/R2 and FA160295/R3. As in Fig. 2 but for (a) FA160295/R2 and (b) FA160295/R3, respectively.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 8.
Fig. 8.

The PV modification locations of FA160295/R3 forecast on 16 Feb 1995 case.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 9.
Fig. 9.

The Meteosat water vapor imagery in gray shades and PV on the 315 K isentropic surface in white contours at 0300 UTC 27 Apr 1992. Dark shades represent dry air, light shades represent moist air. (a) FA270492 forecast at T + 3. (b) FC260492 forecast at T + 15. Contour interval is 0.5 pvu.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 10.
Fig. 10.

The PV modification locations for rerun forecasts in the 27 Apr 1992 case.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 11.
Fig. 11.

As in Fig. 9 but for (a) FA270492/R1 and (b) FA270492/R2 forecasts, respectively, for T + 3 at 0300 UTC.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 12.
Fig. 12.

Plots of the surface pressure differences for 1800 UTC 27 Apr 1992. (a) FA270492 − FC260492, (b) FA270492/R1 − FC260492, (c) FA270492/R2 − FC260492, (d) FA270492/R3 − FC260492. Contour interval is 2 mb. Note that the FC260492 forecast is for T + 30, whereas the FA270492 forecast and all related reruns are for T + 18.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 13.
Fig. 13.

Initial conditions for FB280995 and FB280995/R1. The Meteosat water vapor imagery in gray shades and the 330 K isentropic surface PV in black contours at initial analysis time for 0600 UTC 28 Sep 1995. Dark shades represent dry air, light shades represent moist air. (a) The original forecast (FB280995), (b) FB280995/R1 forecast. Contour interval is 0.5 pvu.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 14.
Fig. 14.

The PV modification locations of FB280995/R1 forecast for 28 Sep 1995 case.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 15.
Fig. 15.

The PV modification locations of FB280995/R2 forecast for 28 Sep 1995 case.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Fig. 16.
Fig. 16.

Plots of the surface pressure differences for 1800 UTC 29 Sep 1995. (a) FB280995 forecast − verifying analysis, (b) FB280995/R2 forecast − verifying analysis. Contour interval is 2 mb. The original forecast and FB280995/R2 are at T + 36.

Citation: Monthly Weather Review 127, 5; 10.1175/1520-0493(1999)127<0922:SOSRWF>2.0.CO;2

Table 1.

Rerun forecasts and bogus data for case 1 on 16 Feb 1995.

Table 1.
Table 2.

Schematic representation of the verification time, the operational forecast (FA160295), and reruns for the 16 Feb 1995 case.

Table 2.
Table 3.

Rerun forecasts and bogus data for case 2 on 27 Apr 1992.

Table 3.
Table 4.

Schematic representation of the verification time, the FC260492 good forecast, the FA270492 poor forecast, and reruns for the 27 Apr 1992 case.

Table 4.
Table 5.

Rerun forecasts and bogus data for case 3 on 28 Sep 1995.

Table 5.
Table 6.

Schematic representation of the verification time, the operational forecast (FB280995), and reruns for the 28–29 Sep 1995 case.

Table 6.

1

In this case study PV was displayed on the 330 K isentropic surface. The reason is that the choice of the isentropic surface is related with seasonal changes. In midlatitudes in the Northern Hemisphere, the 315 K isentropic surface is suitable for winter and the 330 K isentropic surface is suitable for summer.

Save