Short-term risk forecasts of severe weather

Phys. Chem. Earth (B), Vol. 25, No. 10-12, pp. 1335-1338.2000

0 2000 Elsevier Science Ltd All rights reserved 1464-1909/00/.$ - see front matter

Pergamon

PII: S 1464- 1909(00)00204-5

Short-Term Risk Forecasts of Severe Weather W. Schmidl, S. Mecklenhurg’ and J. Joss2 ‘Atmospheric Science, 2MeteoSwiss,

Ch-6605

ETH, CH-8093 Locarno-Monti,

Zurich, Switzerland Switzerland

Received 16 June 2000; accepted 10 July 2000

Abstract. Methodologies

for risk forecasts of severe weather hardly exist on the scale of nowcasting (O-3 hours). In this contribution we discuss short-term risk forecasts of hail by using COTREURainCast: a procedure to extrapolate radar images into the near future. An error density function is defined using the error of location of the extrapolated radar patterns. The radar forecast is folded (“smeared’) with the density function, leading to a probability distribution of radar intensities. An algorithm to convert the radar data into signatures of hail provides the desired probability (or risk) of hail at any position within the considered window in space and time. This methodology is considered to be useful for risk forecasts of floods, heavy wind and snowfall or freezing rain as well. We will discuss the design of appropriate forecast models. 0 2000 Elsevier Science Ltd. All rights reserved. 1.

This contribution introduces a new methodology to obtain short-term risk forecasts of hail, based on high-resolution radar data. Hail is a local event, and the duration of a hailfall is short (typically some minutes). A deterministic hail forecast for a specific location and time is extremely difficult if not impossible. Any forecast of hail suffers from a considerable uncertainty. This contribution demonstrates how the estimated forecast errors can be converted into the desired risk forecasts. In the next section we summarize the method. Then, we analyse a severe hail event from summer 2000 in northern Switzerland. At the end, we will give an overview about the possibilities to use the same methodology for risk forecasts of other severe weather events.

2.

The method

2.1.

COTREC/RainCast

Introduction

Severe weather events, such as hail, floods, heavy wind, lightning, snowfall or freezing rain are a threat for many human activities. Therefore, their risk assessment and management are challenging tasks to protect human life and properties. Any risk forecast of severe weather may contribute to a better risk management. Probabilistic forecasts of weather or climate variables are obtained using numerical ensemble forecasts (e.g., Palmer et al., 1990) on the scale of some hours, days, or a season. In contrast, less effort has been made in developing methodologies for objective probability or risk forecasts on the scale of minutes to a few hours, This shorter time scale is essential for issuing wamings and getting prepared for the sudden onset of severe weather.

CGTREURainCast extrapolates radar images into the near future. COTRFC refers to a cross-correlation technique retrieving the motion and growth of radar patterns (Rinehart and Garvey, 1978; Li et al., 1995). Motion vectors typically refer to squares 40x40 km in size (Mecklenburg, 2OOOb). RainCast is the commercial version of COTREC (Schmid, 2000). Forecast products based on RainCast are distributed by an ETH-spinoff company. RainCast has the following properties: -

Correspondence to: Dr. Willi Schmid, Institute for Atmospheric Science, ETH-Zurich, CH-8093 Zurich, Switzerland, E-mail: [email protected]

-

1335

The full radar image is extrapolated, considering local variations of echo motions. Extrapolated images are stored up to 120 min, with a resolution of 1 min. New forecasts are calculated every 5 min. The computing time of the forecasts is less than 1 min on a modern workstation.

W. Schmid et al.: Short-Term Risk Forecasts

1336

RainCast is operated with the ‘TODAY” product of MeteoSwiss, a radar composite image (560x432 km) which covers Switzerland and parts of the neighboring countries. The pixel resolution of this product is 2x2 km. Joss et al. (1998) gives details about the Swiss radars and radar products. 2.2.

2.3.

Risk forecasts

Risk forecasts of precipitation at a given position of interest are obtained considering the absolute error (s hereafter) and assuming that the absolute value of the difference vector (v) between the forecasted and observed position of a radar pattern follows a 2-dimensional Gaussian distribution G(v):

Forecast quality G(v) = exp(-0.5(v/s)2)

A major effort has been done to assess the forecast errors quantitatively for various weather types (Mecklenburg et al., 2000a). The forecast errors are splitted into two categories: errors of position and intensity. The error values are obtained applying cross-correlation to the extrapolated and observed radar images. At this point, we introduce the absolute error: being defined as the mean value of the magnitude of the displacement vectors between the extrapolated and the real radar patterns. The absolute error is comparable to the distance errors proposed by Schmid (1991) and Johnson et al. (1998). Both studies present statistics about the mean distances between the predicted and observed mass centroids of convective cells. Fig. 1 shows the properties of the absolut& error derived from COTREC forecasts for convective precipitation. The figure indicates an almost linear relationship between the absolute error and the extrapolation time. As an example, an absolute error of about 5 km has to be considered on average for a forecast of 20 min.

I

I

I,

I

I

I,

I

I

I

-0ricjnbl model _ _Optlm@ed model .... ..-... .. .. .. . .. .. .. .. .. .. .. i... .. .. .. .. .. .. .. .. .. .. .. .. + John&n et al. Schmid f,1991)

cl

20 Extrapolation

40 time

Based on the findings summarized in Fig. 1, we use a linear relationship between s (given in meters) and the extrapolation time t (seconds): s=4.t

60 [min]

Fig. 1. The absolute error as a function of the extrapolation time, for different extrapolation models. Adapted from Mecklenhmg (2000b).

(2)

A network of grid points in the neighborhood of the position of interest is introduced. The mesh size of the network depends on the extrapolation time. The larger the extrapolation time, the wider the mesh size of the network, covering a square whose area is 362. The forecasted radar reflectivity Z for each grid point is converted into a rainfall intensity R using a standard Z-R relationship. Each rainfall value contributes with a weight G(v) to the frequency distribution flR)) of forecasted rainfall intensities at the position of interest. The frequency distribution is normalized such that the sum of flR) over all grid points corresponds to 100%. The frequency distributionflR) allows to calculate the risk of precipitation exceeding any given threshold of precipitation intensity. Attributing rainfall intensities > 100 mm/h to hail (e.g., Geotis, 1963; Waldvogel et al., 1978) allows to estimate the risk of hail for the position of interest. This procedure can be repeated for any position in the radar range, and for any extrapolation period between 1 and 60 min.

3.

A

(1)

Case study: June 5,200O

June 5 was the first major hail day of summer 2000 in northem Switzerland. Supercellular hailstorms occurred in northwestern, central and eastern Switzerland. To illustrate the described procedure, we calculated the mvimum risk of hail for a 30-min period, covering an extrapolation time between 30 and 60 min. We repeated the procedure for any location within a rectangular area of 130x100 km in size. The result is a risk map of hail, shown in Fig. 2a. The patterns of hail risk are mainly affected by three hailstorms. The centroids of the three storms at 1700 h, i.e. at the time of the last available radar image, are marked with a “x”. The risk patterns extend mainly in direction NE, i.e. in direction of the forecasted storm motion. The possible errors of location are well reflected by the overall size of the risk patterns. We find maximum risk values of the order of 20-50 %.

W. Schmid et al.: Short-Term Risk Forecasts

Fig. 2. (a) Map of a hail risk forecast for eastern Switzerland. The time of the last available radar image is 17oOh(local time). The risk forecast is calculated for 1730-18OOh(30-60 min extrapolation time). The centroids of three hailstorms. at 1700h are indicated with a “x”. (b) Forecast of the risk of precipitation for Dtfbendorf, indicated with a “+” in (a). The colored

areas refer to weak (blue), moderate (green), heavy (yellow) and extreme (red) precipitation. Extreme precipitation corresponds to hail. The red curve in the bottom represents rainfall accumulation, estimated from the last available radar images (1630-17OCh) and forecasted for the subsequent hour (1700-1800h).

W. Schmid et al.: Short-Term Risk Forecasts

1338

Fig 2b shows a risk forecast for the location of Diibendorf, marked with a “+” in Fig. 2a. The risk of precipitation is plotted for five intensity levels: dry (T), weak (L), moderate (M), strong (S), and extreme Q. We note, for instance, that the probability (or risk) for at least strong precipitation reaches about 85% at 1735 h, which corresponds to an extrapolation time of 35 min. The risk for extreme precipitation (attributed to hail),reaches 50% some min later. Hence, the figure provides a quick, but nevertheless detailed overview on the predicted time evolution of the spectrum of rainfall intensity, including hail. The diagram in the lower part of Fig. 2b shows the measured (past 30 min: 1630-1700 h) and forecasted rainfall accumulation. This information is useful for estimates of the flood potential.

Such forecasts would be useful for winter road services. Work is in progress to develop a proper forecast and warning system. A fundamental point, not considered in this contribution, is the validation of risk forecasts. Validation is complicated, because we have to compare a probabilistic forecast with a binary result (yes/no). Several methods dealing with this problem exist and are summarized by Zhang and Casey (2000). At present, we develop an operational environment for real-time validation of the RainCast forecasts. The possibilities to validate risk forecasts within this environment need to be further studied. Acknowledgement. We ate grateful to MeteoSwiss for providing the radar

data for this study.

4.

Outlook

A procedure to obtain short-term risk forecasts of precipitation, including hailfall, has been presented. The method implements the following components: A model for extrapolation of radar images into the near future b) Information about the extrapolation errors cl Conversion of the radar measurements into rain/hail intensity 4 Calculation of the risk of precipitation including hail

4

The scheme for calculating the risk factors can be refined. So far, we only considered the error of position. We intend to expand the present concept by considering the error of intensity and also the uncertainties associated with the conversion of the radar quantities into quantities of rain/hail intensity. In this study, we discussed risk forecasts of hailfall. However, the methodology can be adapted to risk forecasts of other severe weather phenomena related to precipitation, such as floods, lightning, wind, snowfall or freezing rain. Risk forecasts of snowfall or freezing rain, for instance, require at least two risk factors: a) b)

the risk of precipitation the risk of a temperature below or near the freezing level

References Geotis,S.G.. Some radar measurements of hailstorms. J. Appl. Meteor., 2, 270-275, 1963.

Johnson, J.T., MacKeen, EL., Witt, A., DeWayne, M., Stumpf, G.J., Eilts, M.D., and Thomas, K.W.. The storm cell identification and tracking algorithm: An enhanced WSR-88D algorithm. Wea. Forecasting, 13, 263-276. 1998. Joss, J., Sch&ller, B., Galli, Cl., Cavalli, R., Boscacci. M., Held, E., Dellabruna G.. Kappenberger, G.. Nespor, V., and Spiess, R., Operational use of radar for precipitation measurements in Switzerland. Technical Report, Emal Report NRP31, VDF Hochschulverlag an der ETH Zurich, CH-8092 Zurich. 110 tip, 1998. Li, L., S&mid, W., and Joss, J., Nowcasting of motion and gmwth of precipitation with radar over a complex orography. J. Appl. Meteor., 34, 12861300,1995. Mecklenburg, S. , Joss, J., and S&mid, W., Improving the nowcasting of precipitation in an Alpine region with an enhanced radar echo tracking algorithm. J. Hydml., submitted, 2ooOa. Mecklenburg, S., Nowcasting precipitation in an alpine region with a radar echo tracking algorithm. Diss. ETH Nr. 13608,118 pp., 2000. Palmer, T. N.. Mumau, R., and T. Molteni, T., The Monte Carlo forecast. Weather, 45,198-207,199O.

Rinehart, R.E. and Gawey, E.T., Three-dimensional storm motion detection by conventional weather radar. Nature, 273,287-289.1978. Schmid, W., The prediction of hail. Part II: the movement of hail cells. Ahnos. Rex, 28.71-91, 1991. Schmid, W.. Nowcasting winter precipitation with radar. Proc., 10th lnt. Road Weather Conference, Davos, Switzerland. Swiss Federal Roads . Office, CH-3003 Bern, 17-24.2000. Waldvogel, A., Federer, B.. S&mid, W.. aud Mexeix, JR, The kinetic energyof hailfalls.F’artII: Radar and hailpads. J. Appl. Meteor., 17, 1680-1693.1978. Zhang, H. and Casey, T., Verification of categorial probability forecasts. Wea. Fonmsiing,

15.80-89.2000.