Prediction for Climate Services: Highlights SPECS Climate Prediction for Climate Services: Highlights F.J. Doblas-Reyes ICREA and Barcelona Supercomputing Center Barcelona, Spain

SPECS motivation http://www.specs-fp7.eu What: to produce quasi-operational and actionable local climate information Why: need information with improved forecast quality, a focus on extreme climate events and enhanced communication and services for RCOFs, NHMSs and a wide range of public and private stakeholders How: with a new generation of reliable European climate forecast systems, including initialised ESMs, efficient regionalisation tools and combination methods, and an enhanced dissemination and communication protocol Where: over land, focus on Europe, Africa, South America When: seasonal-to-decadal time scales over the longest possible observational period http://www.specs-fp7.eu Where we got started from ECOMS Conference, 7 October 2016

Climate data is not climate information SPECS Overview Strong links to EUPORIAS, but also NACLIM, IS-ENES2, PREFACE, … BSC Climate data is not climate information The project structure and motto ECOMS Conference, 7 October 2016

Coordinated experiments Tenths of climate prediction experiments, seven models, with different configurations and parameterisations Focus on both skill improvements and processes (case studies) 130 TB of output (and growing), most of it available from ESGF (a primer in climate prediction), curated in the long term Data management and experiment documentation are fundamental. They shouldn’t be underestimated The main message here is that the coordination of all those experiments and the availability of all these data requires a lot of effort from everyone, and needs to be taken into account from the early stages of the project. The legacy of the project depends largely on this. However, you could also mention that ESGF still poses a lot of problems and that the access to the data via other means is still necessary ECOMS Conference, 7 October 2016

Open access tools Portals are useful, but open access tools allow to go beyond what is initially considered by portal developers The packages created in SPECS can be better adapted to address specific problems in an interaction with the users There is no magic recipe, users should be accompanied to make an efficient use of these tools A big effort has been made to create tools that benefited the project partners, but also those in other projects (EUPORIAS) and the users in general ECOMS Conference, 7 October 2016

Forecast initialisation Initialisation matters, but the initialisation problem is far from being resolved; the initial condition uncertainty remains large (ocean, land surface and sea ice); better use of data assimilation techniques is needed The forecast drift is a huge problem that has been characterised; some links between drift and skill have been identified Simplified models can be very useful to explore solutions like coupled initialisation Models predict in a non-stationary state (because of the drift) and the systematic error is too large at times to benefit from good initial conditions A lot of work has gone into improving the initialisation of the systems (with the sensitivity to the initialisation of the sea ice, land surface, etc); improvements can be found at times with a better initialisation, including both better observations and better ensemble generation methods, but the systematic error (which is not stationary when the model is drifting) might prevent the model from making the most of having the best observations possible or the best way to perturb the system ECOMS Conference, 7 October 2016

Process understanding Case studies based on observed extreme events (European summer 2003 and 2010, North Atlantic warming and cooling) have been analysed Interesting sources of predictability for some events have been identified The process-based analysis of hits, misses and false alarms help improving the forecast systems if it is done in a coordinated multi-model context There was a lot of analysis of case studies, most of them chosen by the partners as a function of the extremes discussed at the GA3. The role of soil moisture, resolution, sea ice, etc was discussed in the improved representation of the events in the models. In some way, this is an analysis of the hits and misses conditioned on the observed events. However, there was very little about the analysis of the false alarms, which also make a lot of harm to the forecasts ECOMS Conference, 7 October 2016

Better forecast systems Improvements in the resolution, vegetation treatment, stochastic parameterisations, initialisation, sampling (ensemble and hindcast size), … Improvements in forecast quality are found, but statistical significance is an issue Improving the forecast systems takes long time; need to focus on those aspects that have a stronger impact for a wide range of users (e-g- NAO) Lots of results could be included here. I inserted a plot from Chloé about the impact of the horizontal resolution for Niño3.4 forecasts, the impact of the ensemble size from MiKlip and the improvement of the Z500 systematic error from Lauriane using the stochastic dynamics. The main point is that a lot of work has gone into improving the systems and transferring this information to the operational centres. However, the improvements are incremental, the impact of the schemes depend on the model configuration (the case of the stochastic physics is a good example) and requires testing in more than one model to obtain robust results ECOMS Conference, 7 October 2016

Much more than global models The operationalisation of decadal forecasts, the creation of robust empirical benchmarks, the exhaustive illustration of the (limited) benefits of downscaling and model combination, the generation of entry-level documentation, have been rewarding aspects of the project activity Multidisciplinarity is fundamental in climate services, and this should include climate modellers and forecasters, which requires profiles that are not readily available The idea here is to make clear that the modelling part should be complemented by many other indispensable stages: downscaling, calibration, combination, documentation, operationalisation, dissemination of the knowledge within and outside the climate research community. The rightmost plot is an example of downscaling for the Philippines from José’s group ECOMS Conference, 7 October 2016

Legacy SPECS has tried to make a difference in engaging with multiple communities like WCRP (WGSIP drift project), Polar Prediction Project (role of sea ice and snow in climate prediction), GFCS (tools and examples for the RCOFs), Copernicus (verification and standards), … The sustainability of the outcome of research projects in an international context can be ensured by the operationalisation of their conclusions 10 ECOMS Conference, 5 October 2016 ECOMS Conference, 7 October 2016

Climate data is not climate information SPECS Overview Strong links to EUPORIAS, but also NACLIM, IS-ENES2, PREFACE, … Climate data is not climate information … but there is no climate information without reliable climate data BSC This might sound obvious, but somehow the climate services community shouldn’t consider that the climate modelling problem is solved ECOMS Conference, 7 October 2016

SPECS Climate Prediction for Climate Services