Short-Range Forecasting System of Extreme Weather Events

1. Introduction

Operational and accurate short-term precipitation forecasts present a strong interest for a broad range of decision makers, including agriculturalists, winemakers, hydrologists, insurers, emergency managers, and various industrialists (Ebert et al., 2007; Kabir et al., 2018). By assimilating satellite and in-situ observations, numerical weather prediction (NWP) models remain the primary source of forecasting weather conditions. Moreover, technological advancements enable us to exploit much more data of global and local importance. McGovern et al. (2017) reviewed artificial intelligence techniques in applied meteorology for the assessment of storms duration, the prediction of severe winds, severe
hails and aviation turbulence.
Motivated by an industrial application, this study provides a meteorological short-range forecasting system to estimate the occurrence probability of extreme-weather events (PEWE) at the ground level in the Rhône valley (France). This PEWE study includes hail precipitations, cloud-to-ground lightnings and heavy rainfalls. The proposed probabilistic short-range forecasting system relies on the intensive input of weather measurements, remote-sensing observations, NWP model outputs, as well as archived feedbacks from precipitation-sensitive clients.

2. Decision-support tool

For better assisting its clients, Selerys (a company specialized in storm risk management) aims at developing a decision-support tool able to accurately alert about extreme-weather events.

To design nowcasting products, three main approaches can be distinguished in the related literature: (i) using only radar-based data, (ii) combining NWP model precipitation forecasts with extrapolation methods, (iii) applying advective-statistical techniques (Sokol and Pesice, 2012). The approach discussed in this study falls in the third category. Cintineo et al. (2014) showed that the combination of GOES derived-satellite data, NEXRAD radar data and NWP outputs were able to estimate the probability of occurrence of the severe weather within a 2-hour time range. The authors emphasized that cloud-to-ground lightning was a missing parameter to be investigated as a potential predictor.
In this study we investigate the relation between predicting parameters, namely (1) direct remote-sensing observations (satellite and radar) and model outputs (weather and air-quality forecasts) with (2) the occurrence of extreme-weather events at the ground level. The data used in this study cover the years 2016 and 2017, over the area of the Rhône valley, France. The size of collected data exceeds 150 Terabytes.

After an in-depth analysis, forecast predictors have been chosen:

• ECMWF operational weather forecast meteorological model,
• Selerys local network of radars,
• Meteosat Second Generation (MSG) with SEVIRI instrument used to monitor the atmosphere in 12 spectral bands,
• MOCAGE and MACC atmospheric chemistry models.

NWP models outputs are currently extensively used in MSG SEVIRI (NWC-SAF) derived products (Derrien and Gléau, 2010). Air quality triggers a significant role in clouds dynamics (Rosenfeld et al., 2008).

Ground-truth validation data encompass

• rainfall with the Météo-France PANTHERE product based on radar reflectivity (Tabary et al., 2007),
• cloud-to-ground lightning with Blitzortung database,
• hail precipitation reconstructed from PANTHERE product and crowdsourced local data.

The recent advancements in high-performance computing technologies made possible the intensive practice of machine-learning techniques, which offer a plenty of algorithms for dealing with highly non-linear systems (Cramer et al., 2017; Kühnlein et al., 2014).
The proposed short-range forecasting system resides in two main steps:

1. Performance evaluation of large-resolution forecasting models for the geographical area of interest.
2. Probabilistic forecasting system development based on the threshold specification and its validation. These threshold parameters are set according to the requirements of decision makers.

As the data acquisition is being done, the implementation of the proposed short-range forecasting system of extreme-weather events is currently in progress.

3. Conclusion

This paper presents a short-range extreme-weather event forecasting tool dedicated to supporting industrial needs. Large amounts of data are gathered, recorded and stored for explorating and forecasting purposes. The designed probabilistic system is currently under implementation and is expected to work efficiently in the industrial practice.

4. Acknowledgements

Our computating and data system is located at France-Grille, a French national cloud-computing consortium designed for scientific research.

References

Cintineo, J. L., Pavolonis, M.J.and Sieglaff, J., and Lindsey, D. (2014). An empirical model for assessing the severe weather potential of developing convection. Weather and Forecasting, 29:639–653.

Cramer, S., Kampouridis, M., Freitas, A. A., and Alexandridis, A. K. (2017). An extensive evaluation of seven machine learning methods for rainfall prediction in weather derivatives. Expert
Systems with Applications, 85:169–181.

Derrien, M. and Gléau, H. L. (2010). Improvement of cloud detection near sunrise and sunset by temporal-differencing and region-growing techniques with real-time seviri. International Journal of Remote Sensing, 31(7):1765–1780.

Ebert, E. E., Janowiak, J. E., and Kidd, C. (2007). Comparison of near-real-time precipitation estimates from satellite observations and numerical models. Bulletin of the American Meteorological Society, 88(1):47–64.

Kabir, E., Guikema, S., and Kane, B. (2018). Statistical modeling of tree failures during storms. Reliability Engineering and System Safety, 177:68 – 79.

Kühnlein, M., Appelhans, T., Thies, B., and Nauss, T. (2014). Improving the accuracy of rainfall rates from optical satellite sensors with machine learning — a random forests-based approach applied to MSG SEVIRI. Remote Sensing of
Environment, 141:129 – 143.

McGovern, A., Elmore, K. L., Gagne, D. J., Haupt, S. E., Karstens, C. D., Lagerquist, R., Smith, T., and Williams, J. K. (2017). Using artificial intelligence to improve real-time decision-making for high-impact weather. Bulletin of the American Meteorological Society, 98(10):2073–2090.

Rosenfeld, D., Woodley, W. L., Lerner, A., Kelman, G., and Lindsey, D. T. (2008). Satellite detection of severe convective storms by their retrieved vertical profiles of cloud particle effective radius and thermodynamic phase. Journal of Geophysical Research: Atmospheres (1984–2012), 113(D4).

Sokol, Z. and Pesice, P. (2012). Nowcasting of precipitation-advective statistical forecast model (SAM) for the Czech Republic. Atmospheric research, 103:70–79.

Tabary, P., Desplats, J., Do Khac, K., Eideliman, F., Gueguen, C., and Heinrich, J.-C. (2007). The new french operational radar rainfall product. part ii: Validation. Weather and Forecasting, 22(3):409–427.