Probabilistic and ensemble forecasting from short to seasonal time scales (SPARK session)


Probabilistic and ensemble forecasting from short to seasonal time scales (SPARK session)
Convener: Andrea Montani | Co-conveners: Jan Barkmeijer, Fernando Prates
Lightning talks
| Wed, 08 Sep, 14:00–15:30 (CEST)

Lightning talks: Wed, 8 Sep

Chairpersons: Andrea Montani, Fernando Prates, Jan Barkmeijer
Emanuele Silvio Gentile, Suzanne L. Gray, Janet F. Barlow, and Huw W. Lewis

Convective-scale ensemble prediction systems (EPS) are critical tools to accurately forecast damaging surface winds in the short range,  capturing the local details of their variability and providing guidance on the associated forecast uncertainty. Due to computational cost, operational convective-scale EPS are atmosphere-only models, which represent ocean and wave effects through sea-state independent parametrizations, and therefore do not account for the impact of an evolving ocean and wave state during the forecast. Benefits of integrating atmosphere, ocean, and wave feedbacks into a single coupled multimodel system have been shown by global-scale deterministic systems and EPS, and convective-scale deterministic systems. These benefits lead to the question of what are the corresponding benefits of coupling in convective-scale EPS. To address this question, we present the first convective-scale regional ensemble coupled system focused on the UK domain and surrounding seas (termed RECS-UK). We demonstrate the robustness of the impact of atmosphere-ocean-wave coupling and stochastic perturbations to model physics parametrizations on forecasts of extratropical cyclone Ciara and quantify the importance of these coupling impacts relative to initial condition error. 

Coupling to the ocean leads to localised reductions in the 10-m wind speeds due to cooling of sea surface temperatures, which increase the stability in the surface layer. However, these localised impacts on coupling to ocean are not apparent when comparing the ensemble strike probabilities of exceeding a storm wind threshold (set to 20 m s-1) for the atmosphere-ocean-coupled and control (atmosphere-only) ensembles. In contrast, coupling additionally to waves leads to substantial reductions in wind strike probability and consistently reduces, by up to 1 m s-1, the ensemble forecast median and mean of Ciara’s wind speeds at all simulation hours during which Ciara is in the model domain. Each atmosphere-ocean-wave coupled ensemble member simulates the dynamical response of wind speeds to the forced young ocean waves, with maximum reductions in high wind speed regions. The largest 10-m wind speed spread from stochastic and initial condition perturbations is found away from the strongest wind speed regions of Ciara, but the impact of coupling to waves is more enhanced in these strongest wind speed regions, and is also comparable in size there with the largest sensitivity to stochastic and initial condition perturbations. The implications of this work are that the impacts on 10-m wind speed of coupling convective-scale atmospheric models to ocean and wave models can be robust across an ensemble and be of comparable size to those of initial condition and stochastic physics perturbations. However, convection-permitting atmosphere-ocean-wave coupled EPS should be assessed in different meteorological conditions and further tested on longer timescales prior to operational implementation. 

How to cite: Gentile, E. S., Gray, S. L., Barlow, J. F., and Lewis, H. W.: The importance of atmosphere-ocean-wave coupling in ensemble regional convective-scale forecasts of midlatitude cyclones, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-62, https://doi.org/10.5194/ems2021-62, 2021.

Aristofanis Tsiringakis, Wim de Rooy, Sibbo van der Veen, and Jan Barkmeijer

In an ensemble prediction system (EPS) the uncertainty in the initial atmospheric conditions is usually represented via perturbation of the initial atmospheric state and different boundary conditions at the beginning and throughout the duration of the forecast. These approaches exclude the uncertainty due to the representation of physical processes within the parameterization schemes of a numerical weather prediction model (NWP). Much of the uncertainty in the presentation of physical process arises from uncertain parameter values regulating key physical processes in the boundary-layer and microphysics schemes. This uncertainty can be represented with a Stochastically Perturbed Parameterization (SPP) scheme, where parameter values for the different model grid points are randomly selected from a defined probability density function. The SPP scheme can improve model performance and increase ensemble spread, but may lead to unrealistic parameter values, which can introduce additional model bias. A potential solution is to use coupled/correlated perturbations for relevant SPP parameters to increase the model performance and ensemble spread, while maintaining physically realistic ranges for the parameters. In this study, we investigate the impact of coupled perturbations in key parameters within the boundary-layer and microphysics schemes of the HarmonEPS model using the new SPP scheme. The performance of the coupled perturbations experiment is evaluated against HarmonEPS experiments using independent parameter perturbations, and perturbations in the initial atmospheric state and boundary conditions for both a winter and a summer period.  We find that coupled perturbations in the SPP scheme can decrease model bias and increase the ensemble spread for the 2m temperature and relative humidity, 10m-wind speed and total cloud cover.

How to cite: Tsiringakis, A., de Rooy, W., van der Veen, S., and Barkmeijer, J.: Coupled parameter perturbations in the SPP scheme of the HarmonEPS, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-212, https://doi.org/10.5194/ems2021-212, 2021.

Alfons Callado, Pau Escribà, David Quintero, David Gil, Maria Cortès, and Joan Montolio

Thinking in terms of uncertainty in the operational forecast is the current goal for high resolution operational predictions. Fields like precipitation with convection, surface winds or fog, are very sensitive to model uncertainties and errors, resulting in a rapid loss of predictability in such meso-scales. The best tool that can quantify this uncertainty is a Short Range Ensemble Prediction System (SREPS). Since final 2016 AEMET runs operationally such a system and names it AEMET-γSREPS. It runs currently at 00 and 12UTC time in three domains over the Iberian Peninsula, Canary Islands and the Antarctic Peninsula, up to 48/60 at 00UTC. According to our operational forecasters, the main contribution of the system is to predict strong convective precipitation and its spatial variability. Besides it is useful to see the range of change of temperature from day to day or the localization of wind gusts in orographic areas and its associated spatial variability. The combination of HARMONIE-AROME with IFS deterministic models and AEMET-γSREPS is the best tool AEMET has for the short-range forecasting. In this talk we first briefly present the design of the system and the domains and times where and when the probabilistic predictions are computed. Then we describe the current status and last updates and the latest verification results. Special emphasis is focused on introducing data assimilation in the system. Finally, the future developments are shown. Although AEMET-γSREPS is developed by the AEMET Predictability group, it wouldn’t exist without the collaboration of the international partners ECMWF, MF, NCEP, NCAR, JMA, CMC and specially the HARMONIE community.

How to cite: Callado, A., Escribà, P., Quintero, D., Gil, D., Cortès, M., and Montolio, J.: AEMET-γSREPS, a fully operational system for high-resolution probabilistic forecasting over Iberian Pensinsula, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-369, https://doi.org/10.5194/ems2021-369, 2021.

Bas Crezee, Claire Merker, Jasmin Vural, Daniel Leuenberger, Alexander Haefele, Maxime Hervo, Giovanni Martucci, and Marco Arpagaus

The current atmospheric observing systems fail to provide observations of temperature and humidity in the planetary boundary layer (PBL) with satisfactory spatial and temporal resolutions despite their potential positive impact on numerical weather prediction (NWP). This is particularly critical for humidity, which exhibits a very high variability in space and time, and for the vertical profile of temperature, which determines the atmospheric stability. Therefore, the analyzed thermodynamic structure of the PBL can be prone to errors, leading to poor forecasts of warnings for relevant phenomena, such as severe storms due to intense summer convection or winter fog and low stratus.

One approach to improve the model’s representation of the PBL is to include novel, ground-based remote sensing profiler observations in the data assimilation system to improve the forecast initial conditions. This also improves the quality of downstream applications relying on a good representation of the PBL in the model, such as dispersion modelling for emergency response after nuclear, chemical or biological incidents.

In this contribution, we present results of the MeteoSwiss effort to include observations from Raman lidar and microwave radiometers into the 1km mesh-size ensemble data assimilation system KENDA-1. To this end, we have developed a forward operator for water vapor mixing ratio and temperature to assimilate profiles from the Raman lidar. Brightness temperatures from the microwave radiometers are assimilated using the RTTOV-gb forward operator. We produced extensive O-B statistics to validate the observations with respect to the model and to derive the error covariance matrices of the observations. Furthermore, we will present results of several data assimilation cycling experiments during summer-time convective situations.

How to cite: Crezee, B., Merker, C., Vural, J., Leuenberger, D., Haefele, A., Hervo, M., Martucci, G., and Arpagaus, M.: Towards operational assimilation of surface based remote sensing temperature and humidity profiler data at MeteoSwiss, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-259, https://doi.org/10.5194/ems2021-259, 2021.

Mika Rantanen, Matti Kämäräinen, Otto Hyvärinen, and Andrea Vajda

Sub-seasonal to seasonal scale forecasts provide useful information for city authorities for operational planning, preparedness and maintenance costs optimization. In the EU H2020 E-SHAPE project the Finnish Meteorological Institute aims at developing an operational service providing user-oriented sub-seasonal and seasonal forecast products for the City of Helsinki tailored for winter maintenance activities. To be able to provide skilful sub-seasonal to seasonal forecasts products, bias adjustment and evaluation of the used weather parameters, i.e. temperature and snow is crucial. 

In this study, we focus on the skill assessment of sub-seasonal temperature forecasts in Helsinki, Finland, experimenting with various methods to adjust the bias from the raw temperature forecasts. Due to its coastal location, skilful forecasting of temperatures for Helsinki is challenging. The temperature gradient on the coastline is especially strong during spring when inland areas warm considerably faster than the coastline. Therefore, raw point forecasts for Helsinki suffer from cold bias during the March-July period.

We use the 2 m temperature extended-range reforecasts obtained from the ECMWF S2S database and apply two bias adjustment techniques: removing the mean bias and the quantile mapping method. Reforecasts for a 20-years period, 2000-2019 with 10 ensemble members, run twice a week for 46 days ahead were calibrated and evaluated. Two datasets are used as reference, observations from Helsinki Kaisaniemi weather station and gridded ERA5 reanalysis data. Thus, these combinations yield in total five sets of forecasts which are evaluated against the observations.

The results of the experiments and the potential added value of bias correction will be presented for discussion. Based on the preliminary results, especially the cold bias in spring and early summer can be improved with the bias-correction methods. The bias-adjusted extended-range temperature forecasts are used in the development of sub-seasonal winter forecast products tailored for the needs of city maintenance.

How to cite: Rantanen, M., Kämäräinen, M., Hyvärinen, O., and Vajda, A.: Skill assessment of sub-seasonal 2-m temperature forecasts for Helsinki, Finland, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-138, https://doi.org/10.5194/ems2021-138, 2021.

Estíbaliz Gascón, Augustin Vintzileos, and Tim Hewson

Ideally, weather forecasts should be provided for points and not for the large regions represented by global model grid boxes. This requirement can be addressed by post-processing global forecast model output, as in “ecPoint”, an entirely new and innovative statistical technique developed by ECMWF that uses decision trees and non-local calibration. Products from ecPoint explicitly incorporate the expected sub-grid variability and gridscale bias correction (which both vary according to a diagnosed “grid-box weather type”).

The HIGHLANDER (HIGH-performance computing to support smart LAND sERvices) project is funded under the Connecting Europe Facility (CEF) – Telecommunication Sector Programme of the European Union. One of its main goals is data processing for more intelligent and sustainable management of natural resources and the territory. And one component of this, managed by ECMWF, is exploiting the CINECA supercomputer facilities in Bologna to extract maximum benefit from the ecPoint technique. The specific aims here are to improve probabilistic 24-h rainfall and 2m temperature representation in sub-seasonal forecasts and in the ERA5 reanalysis. Pre-existing 6-h (currently running in CINECA) and 12-h ecPoint-Rainfall forecasts products are currently being provided by ECMWF in real-time, using shorter range (day 1-15) twice daily predictions at 18km resolution. These probabilistic forecasts have exhibited clear improvements, in both reliability and resolution, relative to the raw model output. ecPoint benefits tend to be more significant when working at a lower resolution, so downscaling from 36km in the sub-seasonal forecast and 31 km in ERA5 can in principle deliver even greater improvements for users (relative to raw model and reanalysis output) than we have seen hitherto.

The final ecPoint sub-seasonal products will cover lead times of 16 to 30 days. They will provide more reliable climatological representations, for points on the land surface, than can raw model output (e.g. numbers of dry or wet days, days of freezing temperatures, which are both relevant for agricultural applications). Meanwhile, ERA5 ecPoint products will be the first probabilistic reanalysis products to have incorporated bias-corrected uncertainty information at point scale. ecPoint products for both systems will comprise 24-h accumulated rainfall and daily minimum, maximum and mean 2m temperature, for both percentiles (1, 2,..99) and (derived from these) probabilities of exceeding certain thresholds. Global outputs will be created, but a particular focus will be Italy, and surrounding Mediterranean areas, for agricultural planning purposes, to deliver benefits in both economic and resource availability terms.

In this presentation, we will introduce the ECMWF activities in the project and the methodologies applied to create the final probabilistic forecast and reanalysis products.

How to cite: Gascón, E., Vintzileos, A., and Hewson, T.: New ecPoint products for sub-seasonal forecasts and the ERA5 reanalysis - a HIGHLANDER project initiative, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-85, https://doi.org/10.5194/ems2021-85, 2021.

Seshagirirao Kolusu, Marion Mittermaier, Joanne Robbins, Caroline Jones, Raghavendra Ashrit, and Ashis K Mitra

The southwest monsoon rains in 2019 were the heaviest over India in a quarter of a century. The 2019 seasonal JJAS precipitation over the whole country was 110 % of its long period average (LPA) of 880mm. Precipitation is a cumulative field driven by many atmospheric processes both within nature and numerical prediction.  It’s a weather variable that impacts everyone’s life and hence is used routinely to assess the skill of modelling systems. In this study, we have analyzed the 2019 JJAS seasonal precipitation forecast skill of two global ensemble models: (1) the UK Met Office GloSea5 and (2) the National Center for Medium Range Weather Forecasting (NCMRWF) Global Ensemble Prediction System (NEPS-G). The Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG-GPM) rainfall and ERA5 winds with high spatial resolution and temporal data are used for verification of the model forecasts across a seamless range of time scales.  In order to compare a seamless range of time scales, we have summed forecast fields over time windows of forecast lead time from 1 day to 2 weeks. We also computed the actual skill and potential skill of the model ensemble forecasts at different lead windows. Our results for both models show large precipitation biases and reduced precipitation skills with forecast lead windows. We also found that the models’ actual and potential skill are sensitive to the number of ensemble members and type of ensemble generation. Moreover, the GloSea5 model actual skill is higher than the NEPS-G model over Indian homogeneous regions. To use the GloSea5 NWP forecast model ensemble members for more quantitative applications in downstream hazard and/or impact-based modelling and applications the between-ensemble-member bias introduced by the lagging needs to be addressed.

How to cite: Kolusu, S., Mittermaier, M., Robbins, J., Jones, C., Ashrit, R., and Mitra, A. K.: Seamless rainfall prediction skill comparison between GLOSEA5 and NEPS-G ensemble prediction systems, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-264, https://doi.org/10.5194/ems2021-264, 2021.

Marion Mittermaier, Seshagiri Rao Kolusu, and Joanne Robbins

The UK Met Office seasonal forecast system, Global Seasonal Forecast System version 5 (GloSea5), is an ensemble forecast prediction system providing sub-seasonal and seasonal forecasts over the globe with ~60 km resolution in the mid-latitudes. GloSea5 also produces hindcasts or historical re-forecasts. The system produces 4 members a day, initialised at 00UTC. Two members run out to 64 days and two run out to 216 days. We use these four members to generate a 40-member lagged ensemble with 10 days of lag time, i.e. for any forecast horizon the oldest members are always 10 days older. Due to this lag and the way these ensemble members are initialised, there is a considerable within-ensemble bias, even for a nominal “day 1” forecast. This within-ensemble bias evolves with increasing lead time horizon.

Traditionally hindcasts are used to correct for the so-called model drift. In this work the idea of using a distribution of daily rainfall amounts from short-lead time forecasts is used using the 2019 Indian monsoon season. Quantile mapping is trialled as a means of removing the “within-ensemble-member” bias to ensure that all ensemble members are drawn from a more consistent underlying distribution. Achieving this would enable the members to be used to drive downstream applications such as hazard or impact models, as such models require individual ensemble members.

The presentation will demonstrate the methodology and the impact it has on ensemble forecast skill, complementing the presentation by Kolusu et al. (same session in conference) which presents an evaluation methodology focusing on patterns for different accumulation lengths and forecast horizons.

How to cite: Mittermaier, M., Kolusu, S. R., and Robbins, J.: A novel way of correcting between-ensemble-member biases in a lagged S2S ensemble, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-345, https://doi.org/10.5194/ems2021-345, 2021.


Supporters & sponsors