The Auto-Nowcast System (ANC), a software system that produces time- and space-specific, routine (every 5 min) short-term (0-1 h) nowcasts of storm locations, is presented. A unique feature of the ANC is its ability to forecast storm initiation. A primary component is the identification and characterization of boundary layer convergence lines. Boundary layer information is used along with storm and cloud characteristics to augment extrapolation with nowcasts of storm initiation, growth and dissipation. A fuzzy logic routine is used to combine predictor fields that are based on observations (radar, satellite, sounding, mesonet and profiler), a numerical boundary layer model and its adjoint, operational model output, forecaster input and feature detection algorithms. The ANC methodology and verification statistics are presented. The ANC is able to routinely improve over extrapolation and persistence.
The requirements of the radar processing software within the Meteorological Service of Canada are demanding. The severe weather forecaster is responsible for providing warnings typically encompassing an area of six to eight radars. So the forecaster must be able to maintain a broad view of the weather (situational awareness) while at the same time able to focus on individual thunderstorms.
The concept that has been developed involves using multi-layered high resolution (1 km per pixel) multi-radar (composite) imagery as a basic tool to monitor the weather. Multi-layer imagery refers to a display where several radar products such as reflectivity based CAPPI's or Echo Tops are overlaid and displayed. Also, images created from algorithmic outputs can be overlaid. From this multi-radar image, the forecaster is able to "drill down" to detailed information by pointing and clicking with a mouse button on a thunderstorm cell to access a multi-panelled zoomed in image of the cell called the "cell view". The sub-panels of this image are various radar products that help the forecaster assess the severity of the storm cell before preparing a severe weather warning. These sub-images may include CAPPI's at various heights, VIL, EchoTop, radial velocity images, automatic cross-sections and an ensemble depiction of the outputs of the severe weather algorithms.
Several severe weather algorithms are used to identify various thunderstorm features such as the cell centroid location, the mesocyclone, the microburst, etc. The cell features are identified using a thresholding and region growing feature identification technique. The cells are prioritorized using a ranking formula based on the attributes of the radar features. They can also be classified using a rules based approach which use the relative location of various severe weather features identified by the algorithms.
Where the same cell is detected in the overlap region between radars, a simple selection rule is used to decide which radar to use to present the information. The cell information is displayable via a tabular representation called the Storm Classification Identification and Tracking (SCIT) display.
Interaction and display of the radar products is done through a Java based application and is therefore platform and hardware independent providing easy access to the information. The user is able to access the cell views from either interacting with the main image display or from the tabular SCIT display.
GRAPES (short form of Global /Regional Assimilation and PrEdiction System) is new generation of Numerical Weather Prediction - a unified model at China Meteorological Administration (CMA). A meso-scale version of GRAPES has been run in real time at NMC/CMA since last summer 2004. GRAPES system 3DVAR for data assimilation, non-hydrostatic and full compressible approximation for dynamic core and lat.-lon. grid are particularly used in.
We are actually developing super-higher resolution versions of our GRAPES model to produce short-range forecasts. We aim to incorporate a 1 km grid mesh of numerical model into the nowcasting system GRAPES_Nowcast. This would produce forecasts for about six hours ahead, updated frequently to use new data, especially from high resolution radar and satellite systems. Work is underway to address the considerable scientific challenge this represents. The current experimental version of the GRAPES_Nowcast system consists of 6 components: Data-Ingesting and Preprocessing, Quality controlling, Hot-starting, Data assimilating rapidly, High resolution Model integration and Post-processing/visualizing. The key-points of GRAPES-Nowcast are high resolution model of GRAPES (1~2km grid mesh) to generate the meso or small scales systems, rapid GRAPES_3DVAR analysis to introduce the new data, hot-start of the cloud physical parameters to reduce the spin-up influence, representation in detail of the PBL and SFC physical and dynamical processes.
An ensemble precipitation forecasting system that combines advection forecasts with downscaled NWP rainfall forecasts has been developed. The uncertainty in the advection forecasts arises from errors in the initial estimations of rain field intensity and motion, and uncertainties due to the temporal development of the field intensity and motion during the forecast period. The uncertainty in the temporal development of rainfall intensity depends on both the scale of the feature being tracked and the lead-time of the forecast. The errors in the rainfall predictions based on NWP forecasts are also dependent on scale but are not as sensitive to lead-time in the short forecast period. STEPS models the various sources of uncertainty using a number of stochastic models to generate an ensemble of perturbations of the deterministic forecast, thereby expressing the forecast uncertainty to the user.
This paper will present the techniques that are used to merge the advection and NWP forecasts into a single deterministic forecast for the period 0-6 hours, and the stochastic models that are used to generate the perturbations about this deterministic forecast. STEPS has been tested using a month of data from the United Kingdom and results from this trial will be presented.
The Hong Kong Observatory has been operating a radar-based nowcasting system SWIRLS (Short-range Warnings of Intense Rainstorms in Localized Systems) since 1998. The System was originally designed for short-range rainfall forecasting in support of the operation of the Observatory's Rainstorm Warning System. It has since evolved into a multi-function system with a wide spectrum of applications in flood warning, landslip warning and tropical cyclone warning.
SWIRLS is composed of a number of components each focusing on specific application domain:
- TREC - This component derives the multi-scale "wind" fields at multi-altitude levels by tracking the movement of individual radar reflectivity echoes using the TREC (Tracking Radar Echoes by Correlation) technique. TREC winds are also used to analyze the inner structure of
- rainstorm, to identify mesocyclone, hail storm as well as the wind distribution around a tropical cyclone (Fig.1).
- GTrack - tracks and predicts the movement of clusters of radar echoes using an object-oriented approach. GTrack is particularly useful in depicting the mutual interaction of weather systems.
- Dynamic radar-raingauge analysis - updates the Z-R relation dynamically using radar reflectivity and rain-gauge data in order to re-calibrate the radar rainfall in real-time. It is a critical step for generating realistic precipitation analysis and prediction.
- QPN & PQPN - extrapolates the rainfall re-calibrated radar echoes along the TREC wind direction using a modified semi-Lagrangian advection scheme to produce 1, 2 and 3-hr accumulated rainfall over the territory (Fig.2). To take into account the temporal/spatial/intensity variability of rainstorm, this component also groups a number of QPN grids to form an ensemble to produce a probability distribution of various rainfall intensity levels.
- Lightning - keeps track the movement of lightning flashes and predicts their future positions 30 minutes ahead by extrapolating the clusters along the TREC winds. Products are updated every 2 minutes.
- TephiViewer - calculates a number of thermodynamical instability indices based on upper-air observations and regional NWP models to alert forecasters on the potential of severe weather.
- GPS/PWV - continuously assesses the content of the water vapour in the atmosphere by analyzing the GPS data received by a local ground-based GPS receiving network. It is exceptionally useful for monitoring the growth/dissipation of rainstorms.
- LAPS - an efficient data assimilation system adapted from the Forecast Systems Laboratory of NOAA to produce rapidly updated (one hour interval or less) local scale (down to 1-km resolution) analyses of the state of the atmosphere. One special application of LAPS is to analyze the internal structure of tropical cyclone and to nowcast the wind and MSLP changes over Hong Kong during the passage of the tropical cyclone.
(i) Automatic TC identification - This project employs artificial intelligent methods, such as genetic algorithm, in combination with Kalman filtering, to locate and track the movement of tropical cyclone based on a sequence of radar images. The information derived could be integrated into the LAPS component to produce better TC analysis and intensity nowcast.
(ii) Rainstorm Pattern Recognition - It aims at identifying potential rainstorm development areas by incorporating GPS/PWV, stability indices, dynamical indices, etc. using Pattern Recognition techniques such as Artificial Neural Network. Results derived by this project could also be integrated into SWIRLS to further improve the QPN.
(iii) RAPIDS - A project tries to blend the QPN from SWIRLS with that from the Non-hydrostatic Model (NHM), adapted from JMA, to produce a frequently updated, high resolution, seamless precipitation forecast for the next 0 to 6 hours ahead. Products generated in this project are expected to have the optimized performance through this integrating process.
TIFS is an interactive nowcast system that has been employed by the Bureau of Meteorology, Australia since 2000 for the production of thunderstorm outlooks and nowcasts in the Sydney area. Based on initial nowcasting guidance provided by such systems as the Thunderstorm Identification, Tracking, Analysis and Nowcasting (TITAN) (Dixon and Weiner, 1993) nowcasters can graphically edit and modify attributes of thunderstorms represented as objects to produce a range of text and graphical warning products. Thunderstorm object attributes that can be modified by the nowcaster include cell existence, location, speed, direction, shape, size and intensity. In this way the nowcasters can provide quick routine monitoring and generation of thunderstorms nowcasts to the public and a range of specialist users. Typically, nowcasters aim to enhance the overall forecast process by providing consistent nowcast policy, quality control to overcome weaknesses and errors in initial guidance, and consideration of environmental factors not considered by the guidance e.g. conditions influencing cell growth, decay and motion.
TIFS was employed in the Sydney 2000 FDP as the WWRP forecast product preparation system and as a means to collate and view web-based summary displays of WWRP products both locally and remotely in a consistent manner. It enabled interactive modification of WWRP products, generation of storm "threat" areas and automated production of text and web-based products for dissemination to forecasters and end-user clients.
Over the past 15 years we have been examining and testing methods to assimilate radar data into numerical models for the purposes of analysis and short term forecasting. Most of this research has centered on 4Dvar assimilation with a meso-beta-scale cloud model. Recently, we have been testing 3DVar assimilation of multiple radar data into the Weather Research and Forecasting (WRF) model.
The following subjects will be discussed at the NowcastingWorkshop in Toulouse:
* The wind/mass balance that exists at the convective scale and the implications for data assimilation.
* Retrievability on the convective scale. The ability of an assimilation scheme using observed data to retrieve parameters in the underlying cloud model as well as conditions of the largescale environment will be examined.
* Wind retrieval using Doppler radar data; should wind retrievals be performed before assimilation?
* Issues surrounding assimilation of data from multiple Doppler radars.
We will conclude by showing examples of analyses and forecasts from a convective scale 4DVar data assimilation system (run at 4km horizontal resolution) which incorporates data from multiple Doppler radars (called the Variational Doppler Radar Analysis System or VDRAS). Examples will be shown from an operational realtime system that runs in the Chicago/Illinois area as well as a case study from the International H2O Project (IHOP).
