Each year extreme rainfall events cause significant damage, as a result of flooding, in the highly urbanised regions along Australia’s eastern coastline. This region is also subject to high population growth, with one-quarter of Australia's total increase in population between 1991 and 1996 accommodated within three kilometres of the coast. This means that the community’s exposure to extreme rainfall events is growing rapidly.

The Third Assessment Report of the IPCC states, "Precipitation extremes are projected to increase more than the mean and the intensity of precipitation events are(sic) projected to increase. The frequency of extreme precipitation events is projected to increase almost everywhere. …………… There is little consistent evidence that shows changes in the projected frequency of tropical cyclones and areas of formation. However, some measures of intensities show projected increases, and some theoretical modelling studies suggest that the upper limit of these intensities could increase. Mean and peak precipitation intensities from tropical cyclones are likely to increase appreciably."

However, these conclusions are, in general, based on the results from coarse resolution climate model simulations with grid cells measuring about 200 ´ 200 km. Extreme rainfall averaged over such large areas is much less than that found over small areas. For regional planning, there is a need to provide extreme rainfall scenarios with fine resolution if projected climate change is to be factored into major infrastructure projects that are being designed to last for decades to come.

This presentation will discuss the results from such a study in which the Colorado State University Regional Atmospheric Modelling System (RAMS) is being used to downscale the results of extreme rainfall days predicted by the CSIRO global climate model. RAMS is being run over the coastal region of southern Queensland and northern New South Wales with a grid spacing of 2.5 km. A statistically representative sample of days is being modelled for present and doubled carbon dioxide levels. The predictions from these simulations will be analysed to provide estimates of the possible changes in extreme rainfall intensity that may occur as a result of climate change.

Casey station, (66.29S, 110.52E) in East Antarctica is prone to extreme wind events where it is not uncommon for the wind speed to average at or near hurricane strength for periods up to 3 or 4 days. in such events the wind appears to be significantly super-geostrophic. Forecasting the onset and duration of such events has been problematical as has been defining the dynamics associated with the flow. The Bureau of Meteorology's local area predictive scheme (LAPS) has been adapted to run over East Antarctica and used to study these storms. Results of the model performance will be discussed along with what has been learned about the structure of the model storms.

Numerical atmospheric prediction using an isentropic vertical coordinate was first introduced in the late 1960s. Isentropic modeling has received a modest yet steady stream of research attention in the 35 years since then. Much of this work has centered on recasting atmospheric representation using this coordinate to better handle the lower boundary condition, resulting in the development of a variety of hybrid isentropic/terrain-following coordinate models (e.g., Bleck, Arakawa, Uccellini, Gall, Johnson, Shapiro). These efforts were all driven, at least in part, by a vision to translate the simplicity of the isentropic perspective on three-dimensional baroclinic structures as shown by Rossby and others into atmospheric modeling. While most applications of isentropic models up to this time have been in research, an exception is in the Rapid Update Cycle (RUC), a mesoscale high-frequency data assimilation and short-range numerical prediction system running operationally at the National Centers for Environmental Prediction (NCEP).

This operational numerical forecast niche at NCEP currently occupied by the RUC forecasts and analyses is designed for applications requiring frequently updated guidance for forecast problems in the next 12 h, including aviation and severe storm forecasting. A first version of the RUC was implemented at 60-km horizontal resolution in 1994, and major revisions were made in 1998 at 40-km resolution and 2002 at 20- km resolution. Each of these revisions included substantial changes to the RUC model in physical parameterizations and lesser changes in numerics. This paper describes the forecast model component of the RUC as of the 20-km version. In particular, the effects of mesoscale diabatic effects on quasi-isentropic representation will be addressed. It has been demonstrated by Johnson and his colleagues at the Univ. of Wisconsin that numerical diffusive effects from vertical diffusion are substantially reduced in a quasi-isentropic global model with resolution of over 100 km.

Here, a comparison is made for integrated cross-coordinate vertical transport between fixed- coordinate and adaptive quasi-isentropic-coordinate versions of the RUC model at 10-20 km resolution for real-data cases over the US for warm and cold seasons. These tests demonstrate the viability of the quasi-isentropic coordinate and substantial reduction in cross- coordinate vertical transport even at 10-km resolution in a case with strong diabatic effects.

Previously, Mills and Colquhoun (1998) linked a decision tree thunderstorm and severe thunderstorm forecasting method to the Australian Limited Area Prediction System (LAPS) model that had a 75 km resolution. Some key decisions in the tree will be explained.

Subsequently the decision tree was linked with the 25 km resolution version of LAPS and the output assessed in an operational trial that began in August 1998. The results of the trial will be described.

Modifications were made to the linked system following the introduction in 1999 of the 12.5 km resolution version of the new LAPS model. Comparison was made between the thunderstorm forecasts it produced and an operational product issued each morning by New South Wales Region Forecasting Centre severe weather forecasters. These forecasts were verified in each of the NSW forecast districts over 42 days from 5 November 2001 to 1 February 2002 based on lightning occurrence. For all districts combined the probability of detection of the operational product was .89, compared with .82 for the model forecasts. The false alarm ratios were respectively .39 and .38.

There will be discussion of the output of the linked system in two recent supercell thunderstorm outbreaks in New South Wales, and of the way operational forecasters utilize the decision tree output and other NWP data for forecasting thunderstorms and severe thunderstorms.

Reference

Mills, G. A. and J. R. Colquhoun 1998: Objective prediction of severe thunderstorm environments: Preliminary results linking a decision tree with an operational NWP model. Wea. Forecasting, 13, 1078-1092.

¹ New South Wales Regional Office, Bureau of Meteorology, Sydney, Australia
² Bureau of Meteorology Research Centre, Melbourne, Australia.

A bulk explicit microphysics scheme has been constructed at BMRC to operate within the various LAPS models. The flexible and modular nature of the code provides the potential to use the scheme in a variety of situations, anywhere from very high-resolution modelling of convective features with multiple classes of ice, to relatively simple representations of the ice-phase and warm cloud/rain. This framework also allows incorporation and testing of new features, avoiding the need to necessarily replace the entire scheme when further development is required. Use of the scheme within the 0.375° LAPS model over the Australian region during two month-long trials has yielded forecasts of precipitation objectively-assessed to be superior to those generated by the present operational scheme. Apart from using the scheme to produce precipitation, diagnostics based on the microphysics have been developed. Using the 0.05° LAPS model over the Victorian region, these diagnostics allow prediction of short-lived convective features that are often difficult to forecast.

A major problem for all forecasters, particularly in the tropics, is prediction of infrequent, heavy rain events. Key characteristics of the tropical atmosphere that make these events so devastating are the potential moisture availability and the inherent conditional instability. These allow (i) very large rainfall totals to occur, (ii) rapid communication between upper and lower levels, and (iii) the response to any large scale forcing to usually develop on the mesoscale. Thus extreme rain events often develop rapidly, occur over relatively small scales, and are difficult to forecast.

One long-term aim of this study is to determine the requirements for improved prediction. There are potential benefits from (i) enhanced observations of the large scale flow, (ii) better analysis at the mesoscale, (iii) improved physical parametrisatons, (iv) increased resolutions, and (v) improved understanding. It is somewhat unclear where research resources should be focussed. The study presented here will review the state of knowledge on extreme rain events and evaluate its relevance to a number of situations studied by the author.

To generalise the results, diagnostic and prediction aspects of events that occurred over Australia, Japan, the Philippines, Malaysia, Vietnam and Indonesia will be described. Encouraging aspects are the important role of the large-to-synoptic scale flow, and the skill – relative to its own climatology - of advanced, high-resolution, operational numerical forecast systems. However deficiencies in mesoscale initialisation and parameterisations still limit the accuracy of details in the forecasts.

Further information on the circulation changes, dynamical aspects, forecast climatology, and mesoscale analysis and prediction will be presented at the workshop.

From the southern summer of 1999-2000, the Australian Bureau of Meteorology has produced operational, high-resolution tropical cyclone forecasts with the BMRC Tropical Cyclone Limited Area Prediction System, TC-LAPS, over the western Pacific and eastern Indian Oceans. The system has 5 basic components: 1. Data assimilation to define the large-scale environment (LSE) and outer structure of the storm; 2. Vortex specification to construct a circulation consistent with the past motion, intensity and size of the storm; 3. High-resolution objective analysis to merge the vortex at the observed location into the LSE; 4. Initialisation using diabatic, dynamical nudging to balance the vortex and insert satellite-observed convective asymmetries; and 5. Prediction with the BMRC Limited Area Prediction Model, which contains advanced numerics and physical parameterisations.

Track verification statistics are rather encouraging. Mean track error at 48-hours is ~ 240 km. There are also promising signs for intensity forecasting. Over the northwest Pacific in 2001, RMS central pressure error at 48-hours for 89 forecasts was ~ 17 hPa, with a forecast-minus-observed bias of ~ 10 hPa. The system however displays some systematic errors, which will be described. Some interesting case studies of significant events, including good and bad forecasts, will also be shown. Planned improvements to each component of the system, to reduce errors will be summarised.

Heavy rainfall associated with tropical cyclones, mesoscale convective systems, and slow moving low pressure systems, is a major source of flooding in Australia. To help predict heavy rain events forecasters use gridded quantitative precipitation forecasts (QPFs) from the Bureau's own global and regional NWP models, as well as QPFs from several other operational centers. In any given case the models usually agree on the existence of a large rain system, but they rarely agree on its precise location and rainfall intensity.

Results from a poor man’s (multi-model) ensemble of QPFs from seven operational NWP models indicate that the ensemble mean gives a more accurate forecast of the location of heavy rain events than do the individual models. The RMS errors and spatial correlation with the observations are also improved. A disadvantage of averaging is that it "smears" the rainfall forecast, resulting in an artificially large rain area and reduced rain intensity. This is of major concern when predicting heavy rain events, but can be largely ameliorated by using histogram matching to transform the rain rates to have the same probability distribution function as the QPF ensemble.

The ensemble approach also allows the probability of precipitation to be estimated. The poor man’s ensemble probabilistic skill is better in winter than in summer, and exceeds that of the ECMWF 51-member ensemble prediction system for at least two days into the forecast period.

1- and 2-day quantitative precipitation forecasts (QPFs) from several operational numerical weather prediction models have been verified against rain gauge observations over the United States since 1995, and over Germany and Australia since 1997 to assess their skill in predicting the occurrence and amount of daily precipitation.

Using traditional verification statistics, model QPFs showed greater skill in winter than in summer, and greater skill in mid-latitudes than in the tropics, where they performed only marginally better than "persis-tence". It was much more difficult to predict the precise occurrence of heavy rain than of light rain. However, when forecast rain systems were verified against the observed rain systems using an entity-based pattern matching technique, the location errors for the heaviest rain systems were smaller than for lighter rain systems.

It does not appear that QPFs from operational NWP models have improved significantly during the last four to five years. As new model versions were introduced their performance changed, but not always for the better. The process of improving model numerics and physics is a complicated juggling act, and unless the accurate prediction of rainfall is made a top priority then improvements in model QPF will continue to come only slowly.

Tropical cyclones are maintained by the thermodynamic disequilibrium that exists between the tropical oceans and atmosphere. This disequilibrium is necessary to effect a turbulent enthalpy flux from the oceans to the atmosphere which, together with the net infrared flux, is necessary to balance the absorption of insolation. It follows that any decrease in the IR flux must be made up for by an increase in disequilibrium. This suggests that anthropogenic global warming should increase the thermodynamic potential for tropical cyclones.

When the intensities of actual storms in the best track data set are normalized by climatological potential intensities, the resulting cumulative distributions appear to have a universal, linear form. This suggests that increasing potential intensity will increase the intensity of all storms by the same factor. Thus the effect of climate change on tropical cyclones can be described by 1) changes in the overall frequency of events, and 2) changes in the potential intensity. I will speculate about how these two factors might vary with climate.

While an increasing body of research is directed toward understanding how tropical cyclone activity responds to climate change, there is very little understanding about how changing tropical cyclone activity might feedback on climate itself. I will argue that tropical cyclones are an integral part of the climate system owing to their control of the thermohaline circulation of the world oceans.

Severe Tropical Cyclone (TC) Chris crossed the Western Australian (WA) coastline 160 kilometres eastnortheast of Port Hedland in the early hours of 6 February 2002. It formed off the northwest Kimberley coast of WA on 3 February 2002 and rapidly intensified to a severe Category 5 TC in about 60 hours, twice the standard rate of development according to the Dvorak model. The track of Chris was slow and erratic on a generally southerly course towards the Australian coastline. It continued its southward movement over land and weakened to below tropical cyclone intensity by the early hours of 7 February 2002.

Cyclogenesis of Chris occurred in a region of cyclonic shear between a strong surge in the northwest monsoon and fresh southeasterly winds to the south. It developed in a favourable environment of low wind shear that remained present throughout its lifetime over water. The movement of Chris was influenced by an anti-cyclone over the Northern Territory and a weak ridge on its southern side.

The paper will use examples of NWP output to illustrate how TC forecasters at the Perth Tropical Cyclone Warning Centre used model guidance to predict intensity change (cyclogenesis/intensification and decay) and track direction, and how this information was incorporated into the warning strategy for northwest coastal communities.

On the afternoon and evening of 11 January 2001 a cold front transported smoke from the Winchelsea and King Island bushfires to Melbourne. The meteorology was particularly complex owing to the delicate balance between the bay/sea breeze and the synoptic northerlies, and the effect these small-scale systems can have on the timing and spatial structure of the arriving cold front. Despite these difficulties the LAPS (5-km) model produced an excellent forecast of the meteorology. The performance of the Australian Air Quality Forecasting System (AAQFS) chemical transport (Eulerian) model and the operational environmental emergency-response (hybrid Lagrangian-Eulerian) model, HYSPLIT, in forecasting this event will be compared.

The high resolution 5km Bureau of Meteorology Research Center Limited Area Prediction System (LAPS05) model is used to predict fog at Sydney and Perth airports. Model predicted results are compared with satellite images and routine observations. The preliminary results indicate that LAPS05 model has potential for providing useful guidance on forecasting conditions favorable for fog . However as the model has limitations impose by the initial conditions, cloud and PBL scheme, it is difficult to predict exactly the fog formation time and its duration.

The prediction of thermally forced atmospheric circulations over the Gulf of Carpentaria region of northeastern Australia was investigated using an operational version of MesoLAPS. The region is renowned for the common occurrence there of long westward-moving convective- and wave-cloud lines, including the celebrated 'Morning Glory' phenomenon. In a series of case studies capability of LAPS to predict the low-level convergence lines that appear to initiate and maintain the cloud lines was investigated. In most of the cases examined, the model showed considerable skill in forecasting the convergence line, even when deep convection formed along it.

Accurate quantitative precipitation forecasts at various spatial and temporal scales are highly desirable forecast products for a large number of applications (e.g., severe weather prediction, hydrological modeling, etc.). The main tools to generate such forecasts are Numerical Weather Prediction (NWP) models, ranging from high resolution (5-10 km) Limited Area Models (LAM) to lower resolution (50-100 km) global models (GM). It has long been recognized that precipitation forecasts using such NWP systems are highly uncertain due to inaccuracies in the model initial and boundary conditions and in the model formulation. The most common approach to overcome this uncertainty is the use of Ensemble Prediction Systems (EPS). While probably the most desirable approach, an EPS is also the most expensive. This talk will show that individual "deterministic" model forecasts can be used to provide probabilistic precipitation forecasts by i) exploiting available but currently unused model information on the subgrid-scale distribution of rainfall in case of lower resolution models and ii) using information from expanded temporal and spatial domains when deriving local rainfall forecasts using high-resolution LAMs.

Subgrid-scale precipitation distributions in global models

In the first part of the talk a simple conceptual model of predicting a distribution of precipitation in individual GM grid boxes will be introduced. This approach relies on knowledge of separate convective and stratiform precipitation rates (available in all GMs today) and the stratiform precipitation fraction (when available). It will be shown that simple statistical and dynamical considerations can be made to derive probabilistic forecasts of the time-averaged precipitation rates within a model grid box instead of the common grid-mean precipitation rates only. A simple toy model will be used to demonstrate both the impact and problems of this approach. Future possibilities for improvements will be highlighted.

Local precipitation distribution estimated from the neighbourhood in space and time

The second part of the talk demonstrates a method for estimating the probability distribution function of rain in a LAM grid box using its near neighbours in space and time. The premise is that heavy rain centres are very difficult to predict in space and time, particularly in convective situations, yet the detail given by high-resolution model quantitative precipitation forecasts makes them very useful to the forecaster because the observed spatial variability and range of intensities are well reproduced. The forecaster does not take the high resolution QPF at face value, but rather does a mental conversion to probability of heavy rain within the overall region. This process can be objectified by explicitly deriving the distribution of predicted rainfall in the "neighbourhood" of a grid box, defined by a spatial radius R and a temporal interval Δt, from which rain probability can be computed. We will show results that compare QPF skill using the deterministic and probabilistic approaches, as well as the sensitivity of probabilistic skill to the size of the neighbourhood.

Landfall is one of the major causes of the demise of tropical cyclones. It is often the moment at which the cyclone presents the major hazard to life and property. Observational studies have shown there may be important changes in the distribution of the low level winds at landfall, while modeling studies suggest larger scale changes in the cyclone can begin to occur when it is several hundred kilometres off-shore. These changes are all due to the changed surface characteristics at landfall, yet little has been done to examine the details of the boundary layer response to such surface contrasts. Indeed, where these have been considered, the approach has been largely to apply the well-developed theory of internal boundary layers and ignore the cyclone-scale changes in the dynamics.

Here, the initial results of some preliminary studies of the changes in boundary layer wind field produced by landfall are presented. It is shown how previous work describing the motion-induced asymmetry in the cyclone wind field may be applied to this case. At landfall, the theory predicts marked asymmetries in both the surface winds, in the height and strength of the low level jet, and in the slope of the radius of maximum winds with height. The results will be supported by comparison with Doppler radar observations of Hurricane Danny, and GPS dropsonde observations of Hurricane Floyd.

The design of ensemble methods for medium range forecasting ten years ago has been aiming mainly at adressing the problem of limited predictability of supra-synoptic weather regimes in the 6-10 days range. From this point of view, it has been shown that the ensembles have delivered improved forecasts compared to a purely deterministic approach, both by improving single-value estimates (ensemble mean) and by providing both reliable and sharp estimates of the probability distributions of large scale flow patterns such as given by 500 hPa height fields. Recent results supporting this view will be presented in the introduction to this presentation together with a short summary of the current state of development of ECMWF forecasting system.

There is no reason why the provision of probability distributions for parameters more directly related to the weather such as wind or precipitation could not help decision-making processes in the early medium-range (3-5 days) as well. Of particular interest for the users is to know if the ensembles are able to detect severe weather. These are usually not seen by the models as the most likely scenarios at these ranges, but forecasters have expressed an interest in using even small probabilities of severe weather occurrence as a useful early warnings that will help them focus their monitoring of the situation when the severe weather eventually comes closer. Because the model resolution is a serious limitation when addressing severe weather forecasting, a new method for identifying model proxys for extreme events has been designed at ECMWF. It involves first an estimate of monthly distributions of weather parameters at the model resolution. A new index was then designed (the Extreme Forecast Index) that scales the differences between EPS forecast distributions and the model climate ones. Both the global climate and case studies showing how the new index can help detecting severe weather conditions 3 to 5 days in advance will be presented. Preliminary results showing objective verification in terms of false alarms and hit rates will also be shown.

Finally, one of the most recent developments of ECMWF EPS has aimed at improving the sampling of uncertainties in tropical environments. One of the main objectives is to provide useful probabilistic guidance for tropical cyclone forecasts up to 5 days in advance. New products such as Strike Probability Maps derived from the ensemble forecasts will be presented, together with some verification results.

In recent years, the accuracy of tropical cyclone prediction has improved considerably. In the Australian Region, for example, average operational forecast errors were reduced to below 200 km at 48 hours for the first time in the 1999 - 2000 tropical cyclone season. There still remain, however, cases where large errors occur. Here, a number of tropical cyclones in different oceanic basins have been modelled using the generalised inverse of a high resolution limited area primitive equation forecast model and near continuous satellite data. The tropical cyclones examined were chosen because of difficulties with real time prediction of their tracks. Use of the high resolution model and its inverse, with the near-continuous high resolution data, was able to reduce forecast track errors considerably, resulting in errors below the seasonal mean.

A dilemma currently facing operational forecasters is how to make use of the range of products available in realtime from the Ensemble Prediction System (EPS). These products are based on a burgeoning and well-grounded science. They include postage stamp presentations of rainfall, wind, surface pressure from each member of the ensemble, as well as maps of the probability of occurrence of wind, rainfall, temperature and other elements above set cut-off values, and spaghetti diagrams for particular contours of 500 hPa height. In the case of the Australian (GASP) system, the products are available in easily accessible and user friendly format on internal web-servers, as developed by M. Naughton and colleagues (Naughton, 2002.. this Workshop volume, pp.157ff).

While day-to-day prediction of the weather is a vital role for the Bureau of Meteorology, it is the extreme events that capture the public’s imagination. It is also these events that have the potential to enhance or harm the Bureau’s reputation. Numerical modeling has a number of benefits in forecasting these extreme events, but a particular attribute is the ability to predict an atmospheric state which does not conform to "standard" conceptual models of a weather system, and thus to warn of such events.

In this talk I will concentrate on the short-range prediction of mid-latitude phenomena, as the tropics and the medium range and seasonal scales are being addressed elsewhere at this meeting. One issue that needs to be addressed is the definition of "extreme", as this is not subject to the criteria that define, for example, Severe Thunderstorms. I will present examples of several events, ranging from high temperatures, low humidities, through to extreme extra-tropical cyclogenesis, and tornadic thunderstorms. In the context of these events, I will discuss aspects of how to interpret the NWP model guidance during the forecast process, what might be needed in order to reliably predict such systems, the issues of false alarms and misses, and how numerical model fields can be used to increase our understanding of extreme weather events.

There is a range of needs in predicting fire weather that NWP guidance can assist, ranging from seasonal predictions of regional-scale conditions, through to very short-range predictions of mesoscale circulations at the locations of active fires. In this presentation I’ll pose the question "is extreme fire weather also extreme weather?", and show some examples from major fire events, discuss the issue of identifying wind-changes in nwp model guidance, and verifying model skill in this regard, and touch on the issue of extreme drying, as was observed during the 2001/2002 NSW bushfires.

Observed changes in the extremes of temperature and rainfall may result from natural climate variability, lack of long term data or global warming. However, changes in land cover caused by Human activity also affects the surface energy and water balance. Specifically, land cover changes affect the partitioning of available energy between sensible and latent heating. This must affect near-surface temperatures and the nature of the boundary layer. Using the Community Climate Model, the relative impacts on the extreme and frequency distribution of maximum temperature and convective precipitation resulting from a change in land use is compared to the impact of an increase in carbon dioxide. Simulations were performed using estimates of natural and current land use at 280, 355, 430 and 505 ppmv. Results confirm earlier analyses showing that increasing CO2 leads to increases in extreme temperatures and generally an increase in rainfall intensity. However, results also show that land cover change can cause equivalent impacts, and that the nature of those impacts depend on the exact nature of the land cover changes. Over Europe we find an increase in rainfall intensity while over China we find a decrease. In terms of maximum temperatures, we find a decrease over Europe and an increase over China by amounts which are similar to those resulting from increased CO2. Overall, we are increasingly suspicious that the changes in extremes found at local to regional scales in well instrumented landscapes which are undergoing large scale Human-modification are as likely to be caused by land cover change as by increasing CO2.

Despite dramatic increases in the resolution of global weather prediction models over the last 20 years, according to local forecasters at MetService New Zealand, these models still show particular unreliability in predicting the development and path of subtropical depressions in the Australasian region more than 2 or 3 days ahead. This could be due to many things, e.g. poor initial data, high sensitivity to initial conditions, a particular process that is not well represented in the model or a combination of all of these. In order to determine the error source and thus improve our local forecasting of the major source of heavy rain over northern New Zealand, we have begun a systematic study of these types of storm. We are particularly interested in moist processes and the relative importance of baroclinic instability and diabatic effects in the development of these storms.

In this paper, we describe a subtropical depression that began as an insignificant feature near Darwin in the monsoon trough over the north west of Australia on about January 16, 1995. This system deepened by nearly 20 hPa over the next 3 days as it migrated over the centre of Australia, causing heavy rain and flooding in the region between Alice Springs and Sydney. Although located over central Australia, during this period of its development the storm derived over 80% of its energy from latent heating. The required supply of moisture was not generated locally but advected over 1000km in a low level north easterly flow from the Gulf of Carpentaria and the Queensland coast creating inflow conditions that were conditionally unstable. The amplitude of the storm was concentrated in the lowest few km, typical of a warm cored system. The background baroclinicity appeared to be a catalyst for storm growth, but not the major energy source. We suggest that this diabatic mechanism may operate in many subtropical depressions and may explain the tendency of models to over predict them.

In spring each year the Plains states of the U.S. experience atmospheric conditions where existing deep moist convection would quickly become severe (e.g. the "loaded gun'' sounding). The major challenge to a severe weather forecaster on those days is the question of whether, where and when convection will iniate. There is ample observational evidence that convective initation focuses along ``boundaries,'' i.e. sheets of air that separate two different types of boundary layer air masses. One prominent boundary is the dryline of the Great Southern Plains.

An issue of particular interest with long (~1000-3000 km) boundaries is the preferred along-line location where initiation will take place. A high-resolution (2 km) numerical model simulation using the Penn State/NCAR Mesoscale Modelling System (MM5) successfully captured the dryline initiation event of 15 June 1991. The MM5 simulations suggest potential and testable candidates that created such preferred along-line initiation locations.

The uncertainty in quantitative precipitation forecasts can be managed by generating a number of forecast sequences which are all equally likely, rather than presenting only the expected values as a single forecast sequence. This talk presents a statistical framework for generating stochastic forecasts that are conditioned on the current situation and then provides an example based on a case study of a significant rainfall event in Melbourne. The conceptual model used for the stochastic forecasts is based on the spectral decomposition of a rainfield into components, and then using a standard linear model widely used in time series analysis to model the temporal development of each component in Lagrangian coordinates. Correlated noise is added as the information content in each component perishes so as to produce stochastic forecasts that are conditioned on the current scene. This work is a joint venture with the Met Office and is in an early stage of development.

Recently, the Korea Meteorological Administration (KMA) deployed a high-density network of Automatic Weather Stations (AWS) over South Korea to collect real-time observations of surface meteorological parameters including temperature, wind speed and direction, relative humidity, pressure, and rainfall at high temporal resolution. How to use the AWS data effectively to improve forecasting of heavy rain events over the Korean Peninsula is a challenging task. Within the context of numerical weather prediction, it would be very difficult to make use of the surface observations from such an AWS network, to improve the three-dimensional analysis of the atmosphere, with the traditional static objective analysis technique.

Obtaining an accurate model initial state is recognized as one of the biggest challenges in accurate model prediction of weather events. The variational data assimilation approach is one of the most promising techniques available to directly assimilate heterogeneous mesoscale observations, in order to improve the estimate of the model initial state. During the past three years, a research version of the MM5 3DVAR (3-Dimensional Variational Data Assimilation) system has been developed to assimilate a variety of conventional and remote sensing measurements such as upper-air sounding (TEMP); NWS surface observation (SYNOP, SHIPS, METAR); aircraft data (AIREP, MDCRS); satellite cloud track wind (SATOB); satellite retrieved thickness (SATEM); and Ground-based GPS precipitable water (GPSPW) (Barker et al, 2002). By adding new observation operators and their adjoints, we can implement an AWS component into the MM5 3DVAR system. This will also enable the direct assimilation of AWS data within the current KMA operational environment.

In this study, the observation operators, based on the similarity theory, are developed to directly assimilate the 10-m wind, 2-m temperature and moisture data from the AWS network. With this new surface data assimilation approach, more than 90% of AWS measurements are ingested in 3DVAR. We assimilate AWS data collected for the 14 July 2001 heavy rain case, and perform MM5 forecast experiments to assess the impact of the AWS data assimilation on mesoscale analysis and forecast. In order to compare the 3DVAR results with those from the KMA operational 3-Dimensional Optimal Interpolation (3DOI) system, we also perform MM5 forecasts initialized from the KMA 3DOI analysis at 0000 and 1200 UTC 14 July 2001.

Over many parts of the world, local extreme weather events can often be seen as components of large scale patterns, rather than confined phenomena. We here investigate the hemispheric-scale characteristics of the circulation which were associated with an episode of three successive cold outbreak events which occurred over southeast Australia in 2000. The first and most intense of these occurred on 27 - 28 May, and during that period Melbourne recorded its second lowest May maximum temperature since 1958.

The regional synoptic pattern and its temporal evolution leading to this event were dominated by a persistent anticyclone - cyclone dipole, located to the south of Australia. The 500 hPa height field showed that the regional dipole was part of a large-amplitude hemispheric wave train with a remarkable degree of structural organization. Much of this organization was due to the amplification of wavenumber 4 prior to the onset of the cold event. Initially, the wave energy was concentrated mainly in the eastern hemisphere. However, concurrent with the decay of the first cold event, the energy propagated downstream very quickly, resulting in an amplified wave train in the western hemisphere. This downstream wave development was found to continue for an unusually long time, making two revolutions around the hemisphere and, also, triggering the second and third outbreaks of the episode.

One of the major problems with assimilating data during extreme events is that the most appropriate spread of information from observations to grid-points is unlikely to conform to any statistics based on long-term averages. Ideally fully flow dependent background error statistics would be used to spread the information, however a fully evolving four-dimensional data assimilation system is beyond expected capabilities. This problem is being addressed in BMRC by the development of a new data assimilation system that is capable of using statistics from an ensemble of forecasts as well as using some simplified flow dependence. The flow dependence comes via the use of vertical correlations based on potential temperature rather than pressure, and horizontal correlations that allow for changes in wind velocity. Furthermore, being variational, all types of remotely sensed observations can in principle be used. This talk will discuss the progress of this project.

Statistical downscaling presents several advantages to assess extreme weathers under changed climatic conditions over direct Global Climate Model (GCM) outputs. While GCM grid points are more representative of a spatial average, downscaling allows to resolve the scaling issue by linking large scale predictions to local measurements without spatial averaging. Observed extremes are thus more likely to be match.

In this presentation, several techniques regarding extremes are applied to series of local predictants obtained from climate change scenarios predicted by GCMs. The same technique used for observed series are applied to the reconstructed series. Examples include reproduction of anomalous spells (e.g. number of consecutive dry/wet days, number of consecutive hot/cold days, above/below a fix threshold), reproduction of percentiles (70, 80, 90th percentiles), return periods of record values.

The Australian Air Quality Forecasting System (AAQFS) is a joint project between the Bureau of Meteorology, CSIRO Atmospheric Research, CSIRO Energy Technology, the Environment Protection Authority Victoria and the New South Wales Environment Protection Authority to develop a high-resolution air quality forecasting system. Currently 24–36 hour forecasts are produced in both Melbourne and Sydney twice daily at a horizontal resolution of 1 km in urban areas and 5 km in rural areas. Forecasts are provided for a number of species including, NO_x, O₃, SO₂, CO, C₆H₆, CH₂O, PM10, PM2.5 and passive tracers if required. A description of the complete system and its performance is given by Manins (2001).

Meteorological and air quality verification is routinely conducted within the Melbourne–Geelong and Sydney Airsheds. The present paper seeks to extend the verification by considering urban plume transport of CO from Melbourne to Geelong, 320 km southward to Cape Grim on the northwestern tip of Tasmania. The Cape Grim Baseline Air Pollution Station (CGBAPS) monitors global atmospheric composition and meteorology. CO is primarily generated by motor vehicles and is quasi-conservative; the path between Melbourne and Cape Grim is (nearly) free of sources. A similar study of inter-regional transport of CO to Cape Grim has been carried out by Cox et al. (1999), using analysed winds and a different model, TAPM.

The study period began on 18 September 2001 and ran for 13 days. During this time elevated CO concentrations were recorded during three distinct periods; 18–19, 23–24, and 28–30 September. Time-series of observed and modeled CO were constructed to assess the performance the AAQFS. Maximum modelled CO concentrations within a 0.1 degree radius of Cape Grim were taken as the modelled value. This is a severe test of the system. The accuracy of long-range transport predictions depends on meteorological conditions. Typical errors for trajectories, based on forecast winds, are of the order of 30% or more of the travel distance (Stohl, 1998), i.e., approx. 1° for this study.

The AAQFS demonstrated considerable success in identifying the "pollution" events, at a qualitative level, during this 13-day period . If the detecting radius was extended to 0.5 degrees (half the recognized typical trajectory error) the success rate climbed to 100%. This result is complimented by the ability of the AAQFS to provide considerable spatial and temporal detail of the development and track of the CO plume, although the verification of such detailed inter-regional transport is limited to the single point observations at Cape Grim. These results are encouraging when considering the potential to apply the system to other areas of the globe where inter-regional transport is of considerable importance.

References

Cox, M., Hurley, P., Fraser, P. and Physick, B., 1999: Investigation of Melbourne region pollution events using Cape Grim data, a regional transport model (TAPM) and the EPA Victoria carbon monoxide inventory. Clean Air, 33, 35-40.

Manins, P.C. (Chair of Committee), 2001: Air Quality Forecasting for Australia’s Major Cities– Final Report. Project Management Committee: CSIRO Atmospheric Research, Aspendale, Australia: http://www.dar.csiro.au/information/aaqfs/

Stohl, A., 1998: Computation, accuracy and application of trajectories¾ A review and bibliography. Atmos. Environ., 32, 947-966.

Current knowledge of this topic is reviewed. In the current climate, intense tropical cyclones cause high wind speeds, storm surges and extreme precipitation. All of these are likely to change in a warmer world.

Unresolved issues are identified. The credibility of simulations of tropical cyclone intensity increase in a warmer world would be improved by the use of genuinely fine horizontal resolution models (5 km or better), including an interactive ocean and up-to-date cloud physics. This also applies to simulations suggesting tropical cyclone precipitation increases. Simulations of the effect of climate change on tropical cyclone numbers give inconsistent results and physical understanding of tropical cyclone genesis in the current climate is poor. Current climate models tend to simulate crucial variables, such as vertical wind shear, inadequately. The effect of climate change on ENSO is still not resolved, yet in many parts of the world tropical cyclone numbers are strongly influenced by the phase of ENSO. The impact of climate change on the characteristics of extratropical transition, an important issue in many parts of the northern mid-latitudes, has barely been examined. Palaeoclimate studies that have suggested large changes in tropical cyclone intensities in previous centuries need to be re-examined and put on a firmer footing.

Reference

IPCC, 2002: IPCC Workshop on Changes in Extreme Weather and Climate Events, Beijing, 11-13 June, 2002.

In terms of convection and rainfall, the region around the southern islands of Indonesia is likely the most predictable on the intraseasonal time scale by way of the Madden Julian Oscillation (MJO). This has been indicated by three different empirical prediction schemes for the MJO. But does this have any use for giving advanced warning of the possibility of extreme events in the region? From a case of an MJO in January/February 2002, it appears that it sometimes does: continual rains over several weeks, corresponding to the enhanced convective phase of the MJO, resulted in devastating floods on the island of Java. In this talk I plan to discuss these floods, their association with the MJO, and the forecasts of the MJO that were produced at the time. For this case, there was the potential for advanced warning of the increased rains up to 3 weeks beforehand!