Nowcasting

Meteorological services worldwide experience an ever increasing demand of nowcasts, i.e. forecasts with short lead-times with a high temporal and spatial resolution, and a high update frequency. Moreover, also more and more nowcasting applications find their way to the broad public through mobile devices and platforms. At the Radar and Lightning Detection section, our nowcasting research is strongly oriented towards operational applications. Precipitation nowcasting is the main focus, but also nowcasts for other parameters are generated.

Operational nowcasting: INCA-BE

In 2010, the INCA-BE nowcasting system (Integrated Nowcasting through Comprehensive Analysis Belgium) was introduced at RMI by our section. The system on which INCA-BE is based, has been developed at the meteorological service of Austria (ZAMG). This core system from ZAMG was adapted for a domain covering Belgium and for RMI's observational data flow. It was integrated in RMI's operational chain in spring 2012. It was first used internally by the RMI forecasters, and later on also by external users like the regional hydrological services or the other Belgian met services at Skeyes or the Belgian defence (Meteo Wing). Over the years, INCA-BE is heavily adapted and finetuned, and new extensions were developed. Currently, the output has found its way to numerous downstream applications and users.

INCA-BE system is not confined to precipitation only, but also nowcasts for other parameters are generated, like (2m and ground) temperature, humidity, wind and cloudiness. Furthermore, also some derived fields are available, like precitpitation type, wind chill, wind gusts and visibility. For the calculation of these fields, INCA-BE ingests a wide range of observational and model data: the radar composite from the RADQPE system, data from automatic weather stations, satellite data and NWP output. The nowcasts of the fields gradually converge towards the NWP forecast for longer lead times.

An overview of all INCA-BE input sources.

Probabilistic precipitation nowcasting: STEPS-BE

In parallel to the operational nowcasting system INCA-BE, our section has also implemented and evaluated STEPS‐BE. STEPS‐BE is the local implementation of the probabilistic Short‐Term Ensemble Prediction System that was jointly developed by the Australian Bureau of Meteorology and the UK MetOffice, and adapted for implementation at RMI. The  system generates an ensemble of precipitation nowcasts by perturbing a deterministic radar extrapolation nowcast with spatially and temporally correlated stochastic noise. The STEPS‐BE system runs on our operational HPC infrastructure, allowing for the generation of an ensemble of 48 members every five minutes up to a lead time of two hours.

An essential part of the STEPS system is the decomposition of the rainfall field into different spatial scales, the so-called "cascade". The predictability of the larger scales is much higher than the smaller ones, hence the smaller rainfall features are replaced by stochastic noise in the subsequent nowcasting steps.

Probabilistic nowcasting with NWP: STEPS-ALARO

In order to progressively evolve towards a seamless forecasting system with a 12 hours lead time, blending extrapolation-based nowcasts with NWP output, preferably in a probabilistic framework, is necessary. At RMI, we have implemented the blending of the nowcasts produced by STEPS-BE with precipitation fields from the NWP model ALARO with a horizontal resolution of 1.3 km. ALARO uses a multi-scale package for microphysics and convection, and has been adapted to produce precipitation output with a frequency of five minutes, the same as the radar images.

In this highly sophisticated nowcasting system, coined STEPS-ALARO, the blending of the STEPS-BE cascade with the NWP cascade (i.e. the decomposition of the rainfall field in different scales) is performed level-by-level, meaning that the weight of the NWP contribution can differ between larger and smaller scales. The scale-dependent weights of the extrapolation and NWP forecasts are computed from the recent skill of the respective forecasts, and converge to a climatological value.

Nowcasting research: pySTEPS

The complex code base of STEPS and its closed license were the main motivation to start the PySTEPS initiative: a community driven open-source Python library for probabilistic precipitation nowcasting. PySTEPS (Pulkkinen et al., 2019) is built on the same principles as the original STEPS system, but in addition it supports various input/output file formats and implements several optical flow methods. Also important additions were introduced in the stochastic noise generation, visualisation and forecast verification.

The PySTEPS initiative revitalised the international collaboration of the research and development of operationally oriented nowcasting. PySTEPS is a modular library with interchangeable components, making it a convenient platform for research purposes. Researchers, from their side, are encouraged to actively contribute to the open source code repository.

At RMI, all recent precipitation nowcasting research is conducted in the framework of PySTEPS, and active contributions to the code repository are expected. Moreover, we foresee PySTEPS to become the main precipitation nowcasting system at the RMI in the future.

The bigger picture: interdepartemental project IMA

Precipitation nowcasting reseach and development has become more and more an interdisciplinary effort, with different internal stakeholders like the forecasters, the research department and the developers of downstream applications (e.g. smartphone app). A joint coordination was set up under the form of project IMA. IMA is the first part of the Seamless Prediction programme of the RMI, and with this project, we aim to address the growing demand for calibrated observation-driven seamless probabilistic short-term forecasts from users in areas such as urban hydrology, renewable energy, as well as from the general public. For this we combine (probabilistic) nowcasting techniques (e.g. PySTEPS) with a Mini-EPS of different numerical weather prediction (NWP) models in which novel and high-resolution observations are assimilated.

Schematic overview of the main characteristics of project IMA, and how they interact.

Cookies saved