Seamless prediction systems are designed to deliver frequently updated forecasts that spanvarious timescales by combining extrapolations of the most recent observations - such as weatherradar data - with numerical weather prediction models. End users such as hydrological services,local authorities, smartphone users and the renewable energy sector, require increasingly earlyand accurate forecasts, especially for fields with a high spatiotemporal variability such asprecipitation. Moreover, downstream impact models such as (urban) hydrological models havea strong nonlinear dependence on the meteorological input. So in addition to being fast andaccurate, we also need calibrated ensembles of forecasts to conduct a proper error propagation toassess the impact uncertainty.In response to these requirements, many national meteorological services have introducedseamless prediction systems, with notable examples including DWD{\textquoteright}s SINFONY, Met Office{\textquoteright}sIMPROVER, and Geosphere Austria{\textquoteright}s SAPHIR. In Belgium, Project IMA (after the Japaneseword for ”now” or ”soon”) represents the country{\textquoteright}s seamless prediction approach, utilising theRMI{\textquoteright}s observations network, the gauge-corrected quantitative precipitation estimate RADQPE,the pysteps-be probabilistic rainfall nowcasting system, the INCA-BE analysis and nowcastingsystem, and the ACCORD NWP model configurations ALARO and AROME. Unlike many otherseamless systems, Project IMA features seamless ensemble precipitation nowcasts from 0 to 6hours, aimed to improve predictions of flash floods and their uncertainty.We present the advancements within Project IMA and especially the novelties in pysteps.We share a glimpse of deep learning-based blending methods to extend forecast horizons andimprove calibration, sharpness, and general utility for hydrologists, crisis managers, and otherstakeholders.Project IMA aims to integrate research swiftly into operational processes. It encouragescontributing to open-source software such as pysteps, promoting transparency, reproducibility,and international collaboration, and supporting the UN{\textquoteright}s initiative for “Early Warnings for All”by 2027.
De Cruz, L, Van Ginderachter, M, Reyniers, M, Deckmyn, A, De Kock, S, Dehmous, I, Dewettinck, W, Erdmann, F, Imhoff, R, Moraux, A, Reinoso-Rondinel, R, Veldhuizen, M, Casey, JJ, Faleu Kemajou, L, Kumar, A & Van Nieuwenhuize, V 2025, 'Project IMA: seamless predictions at the Royal Meteorological Institute of Belgium', Precipitation Processes - Estimation and Prediction, Bonn, Germany, 16/03/25 - 21/03/25.
De Cruz, L., Van Ginderachter, M., Reyniers, M., Deckmyn, A., De Kock, S., Dehmous, I., Dewettinck, W., Erdmann, F., Imhoff, R., Moraux, A., Reinoso-Rondinel, R., Veldhuizen, M., Casey, J. J., Faleu Kemajou, L., Kumar, A., & Van Nieuwenhuize, V. (2025). Project IMA: seamless predictions at the Royal Meteorological Institute of Belgium. Abstract from Precipitation Processes - Estimation and Prediction, Bonn, Germany.
@conference{789984adc9f14a5cbef2de82180081c8,
title = "Project IMA: seamless predictions at the Royal Meteorological Institute of Belgium",
abstract = "Seamless prediction systems are designed to deliver frequently updated forecasts that spanvarious timescales by combining extrapolations of the most recent observations - such as weatherradar data - with numerical weather prediction models. End users such as hydrological services,local authorities, smartphone users and the renewable energy sector, require increasingly earlyand accurate forecasts, especially for fields with a high spatiotemporal variability such asprecipitation. Moreover, downstream impact models such as (urban) hydrological models havea strong nonlinear dependence on the meteorological input. So in addition to being fast andaccurate, we also need calibrated ensembles of forecasts to conduct a proper error propagation toassess the impact uncertainty.In response to these requirements, many national meteorological services have introducedseamless prediction systems, with notable examples including DWD{\textquoteright}s SINFONY, Met Office{\textquoteright}sIMPROVER, and Geosphere Austria{\textquoteright}s SAPHIR. In Belgium, Project IMA (after the Japaneseword for ”now” or ”soon”) represents the country{\textquoteright}s seamless prediction approach, utilising theRMI{\textquoteright}s observations network, the gauge-corrected quantitative precipitation estimate RADQPE,the pysteps-be probabilistic rainfall nowcasting system, the INCA-BE analysis and nowcastingsystem, and the ACCORD NWP model configurations ALARO and AROME. Unlike many otherseamless systems, Project IMA features seamless ensemble precipitation nowcasts from 0 to 6hours, aimed to improve predictions of flash floods and their uncertainty.We present the advancements within Project IMA and especially the novelties in pysteps.We share a glimpse of deep learning-based blending methods to extend forecast horizons andimprove calibration, sharpness, and general utility for hydrologists, crisis managers, and otherstakeholders.Project IMA aims to integrate research swiftly into operational processes. It encouragescontributing to open-source software such as pysteps, promoting transparency, reproducibility,and international collaboration, and supporting the UN{\textquoteright}s initiative for “Early Warnings for All”by 2027.",
author = "{De Cruz}, Lesley and {Van Ginderachter}, Michiel and Maarten Reyniers and Alex Deckmyn and {De Kock}, Simon and Idir Dehmous and Wout Dewettinck and Felix Erdmann and Ruben Imhoff and Arthur Moraux and Ricardo Reinoso-Rondinel and Mats Veldhuizen and Casey, {Joseph James} and {Faleu Kemajou}, Loic and Anshul Kumar and {Van Nieuwenhuize}, Viktor",
year = "2025",
month = mar,
day = "4",
language = "English",
note = "Precipitation Processes - Estimation and Prediction, PrePEP ; Conference date: 16-03-2025 Through 21-03-2025",
url = "https://indico.kit.edu/event/4015/page/1131-schedule",
}