Report on Presao 10
Written by JP Céron (Météo-France)
With participation of O N’Diaye (IRI) and A Kamga (Acmad)
1. Introduction :
The Presao (RCOF dedicated to West Africa) has started in 1998, taking the opportunity of an El Niño event over the tropical Pacific Ocean and with the support of the Clips project from WMO. For the 10th session of the Presao and taking into account some evolution in the framework of Seasonal to Interanual Forecasting (SIF), a Pre-Forum was set up in collaboration between Acmad, IRI and Météo-France, leading to a new and important evolution of the process.
The key elements which led us to this direction are :
Ø The new WMO organisation which is proposed for the SIF theme and which was ratified at the last CBS session notably by nominating 9 Global Producing Centres of Long Range Forecasts (GPCs) and additionally 2 other potential candidates. This organisation start from large scale products provided by GPCs to the benefit of RCCs and NMHS where these products should be adapted to the sub-regional and national scales including the tailoring to specific needs of potential users. One has to highlight that GPCs provide the SIF products on a regular base (minimal list of requirements) and should provide also to RCCs and NMHS (on request), under the umbrella of WMO, the corresponding verifications.
Ø The recurrent problem of the lack of real use of SIF at least in West Africa (but probably not only) notably because of the lack of relevant and adapted products (nature of the product, forecasted period, … ). Here one can quote the positive feeback wheel which come from the provision of relevant products: Adapted products = used products = more confidence in SIF products = feebacks from users to providers = better products = …
Ø The needs of recalibration of statistical models used in the frame of the RCOF (mainly calibrated over the 1961-1990 period) notably because of the multi decadal variability.
Ø The potential of Couple General Circulation Models (CMCG), notably looking to the possible forecasted periods, potential products and future improvements.
Ø The availability of friendly sofware (like CPT) which allow to easily and efficiently adapt GCM outputs as provided by GPCs.
Ø Some examples of SIF product uses which clearly demonstrate that, at this stage, substantial skill improvement can be gotten in correcting bias GCMs (notably spatial bias) and adapting GCM outputs notably in terms of spatial scales for the benefit of SIF product users. To be more precise, one get more added value into the forecasting suite by correcting and adapting the GCM ouputs rather than trying to provide large scale forecasts using other models like the statistical ones. To be pointed out here that statistical techniques are very relevant for these corrections/adaptations.
Ø The science of SIF which should start from the large scale signal and large scale forecast before then to perform downscaling and adaptation. In the frame of the Presao (and more generally RCOFs) this is the reverse process which is more usual (one start from national forecasts and the one try to produce a consensus forecast at the scale of the region of interest – e.g. West Africa).
Ø The weakness of NMHS ressources dedicated to SIF.
Ø Last the lack of continuous and coordinated activity during the remaining period beween to Preaso.
2. The proposed evolution :
The general idea is to use in the most efficient way the GPCs’ opportunity by building a system, operational in the next, which allow NMHS first to get GPCs’ products and then to adapt them to the scale of their countries and the needs of their users. In this evolution, one important step was to convince NMHS that this evolution benefit to them and open for futur perspective which are without equivalence with respect to Presao’s «classical» methods; namely SST vs precipitation regression. The evaluation of GPCs’ forecast quality (verification) and the possibilities to use other predictors than SSTs are, obviously, relevant arguments in this context.
Thus, the Pre-Forum has been design in order to support this idea of evolution. Consequently, a first part was dedicated both to statistical model verifications and adaptation models directly calibrated on GCM outputs. One part of the Pre-Forum was also devoted to discussions on expected evolutions and additionnaly on continuous work proposals which should lead NMHS to be more confortable and confident on the use of GCM output and their adaptation. The targetted objective is to converge for the next session of Presao to adapted forecast for each country using operationnal GPCs products; This information being introduced into relevant sofware (like CPT). This work shoul allow to provide relevant products in relevant periods to the benefit of applications and users.
One has also done the state of activites over the last year and try to understand what were the sensitive points in the framework of the proposed evolution.
3. The Pre-Forum:
The main part of presentations and practical was on the verification topic. Notably, it was introduced the most usual criteria, notably within the GPCs’ community (ROC, Reliability, RPSS, MSSS). The practical took advantage of the CPT software (provided by IRI) and the available possibilities about the used of country data in this context. The GCM ECHAM (4.5) forecasts were also prepared by IRI for these practical.
4. The Forum:
The discussion about consensus forecast has been oriented toward the large scale forecast. Notably, possible solutions and probable SST evolution has been deeply discussed. The spread of the different solutions proposed by different GCMs has allowed to review the GCMs forecasts and the consequence for statistical models (mainly SST evolutions - e.g. SST in the Guinean Gulf).
All the countries have presented both national forecasts using «classical» SSTs’ relationship and forecasts issued using GCM outptut adaptation; as starting point the used predictor being the forecasted zonal wind at 925hPa by the ECHAM4.5 model.
This exercice was very beneficial , it has demonstrated both the potentiality of these appraoches (we have to discuss with the GPCs’ community about available predictors in operational mode) and most of the time, the consistence between adapted GCMs output and SSTs-Rainfall based relationships forecasts.
5. The discussions :
State of actions: very few actions conducted during the past intersession. The follow-up problem (notably via the coordination committee) was pointed out; a specific recommendation of the forum was addressed to Acmad on this point. The lach of human resources was also highlighted. Then the data access and the hardware/software problem seem to be limiting factors (notably some PC are from the starting date of Presao ; they are now obsolete et make the work clearly more difficult to achieve – e.g. data transfert).
Sensitive points :
a. Softwares:
This is clearly a crucial point for the success of expected changes. Their documentation is absolutely necessary for downscaling and adaptations of GCMs outputs. Without maintained and easy access software, giving sufficient flexibility to NMHS for different methods and way to proceed but preserving a rigorous approach for method application, this kind of evolution as poor chance to be successful in the majority of the countries. It still remains an open question : software like tool boxes (like CPT) offering a secured solution for method uses versus more open software (like R) which offer may be more possibilities but with some risks of wrong uses of methods. The question of the evolution of CPT has been addressed to IRI and the answer can determine the effort to achieve on this topic and the best way to proceed. WMO should also contacted for notably facilitate the access to specific softwares developed in the frame of research projects (e.g. ENSEMBLE) including their adaptation to the African context.
b. Data:
The access to GPCs’ product should not be a priori a real problem. Nevertheless, one has to take care with easy access to these data (notably available format for the forecasted files) probably prepared for the sub-region (West Africa). In this context, Acmad has probably an important role to play (focal point for GPCs?). On can also quote that the Demeter dataset can be easily used in order to assess the potentialities of GCMs’ output and their adaptations over regions, countries or for specific users. In that case, it remains to recalibrate the models already developed in this “research” framework using hindcast experiences of operational models. Here again, the role of Acamd could be important in the data access facilitation and the organisation of relevant training for the targeted research goals. To be noted that the access to good quality datset from the user/application domain is crucial for the success of such adaptations.
c. New products:
The new orientation as a matter of evidence promote the research and production of new products. Looking to the preparation of such new products, one essential question is the understanding of the whole chain of use of forecasts. Notably the knowledge of the characteristic time scale of decisions and actions of the users are a crucial information for giving the more relevant and best response to the application domain. The example of agriculture, notably through the monsoon onset, has been widely discussed with a conclusion that no evidence appears on the use of the onset date with respect to the seasonal time scale (probably that at least partly of the answer could come from the monthly forecast or even medium range forecast). Other products of interest have been discussed like the number of rainy day, the number of dry day, the length and/or frequency of dry spells, …
The preparation of specific products for applications must be promoted with a high priority because it greatly benefits to the whole Seasonal Forecasting chain et beyond SIF to NMHS and their countries.
d. Identification of forecasting periods:
Here again, the new framework is able to falilitate the issuance of forecasts for the most relevant periods. Additionnally to the different rainy seasons, eventually twice for regions close to the equator, dry seasons (notably for menigitus transmission) or low flow periods (for water resources management including fresh water) have been stressed. One has to build an «operational» system which should be able to deliver timely the necessary and relevant products ; the role of Acmad will be probably crucial in this future organisation.
e. Coordination of the process :
The involvment of the coordination and follow-up comitee, created during Presao 9 but never sollicitated by Acamd, should be effective for both the follow-up of the work of the different working groups and also for the organisation and preparation of the next Presao, including the choice of the date.
6. The inter-session work : (cf. details annexe 1)
In order to allow NMHS in the sub-region to become more comfortable and familiar with the GCMs’ output adaptations a specific work, to be perform between October 2007 and march 2008 has been proposed to NMHS.
This work consist in the use on one hand of the ECHAM model output and on the other hand of the DEMETER project dataset which is composed of SIF coming from 3 different models. It should be quoted that the ensemble of these models are already available or close to operational versions as well that one can found at IRI or GPCs (Met Office, ECMWF, Météo-France). Additionally, NMHS should perform the verification of statistical models and consequently be able to compare the different methods. One has to point out that the set of predictors prepared for this work is voluntary limited but that it could (and should) be extended in the next.
This work has been schedule on a detailed base which start beginning of September by an official letter from Acmad to NMHS explaining the goal of this work and the detailed scheduling : real start of the job in october, a partial review of the state of the job by the end of november (notably verification available) and the end of the job end of march including a technical report from each country sended to Acmad. This work will be reviewed by the team which assume the technical follow-up and will be eventually published by Acmad (with the relevant support – e.g. WMO). The technical follow-up will be assumed by IRI (O Ndiaye et S Trzaska) and Météo-France (JP Céron).
Other optional works have been also proposed:
Ø The documentation of the definition and use of the onset date notably focusing on the user chain; for whom, for what action, in which time/deadline.
Ø The documentation of others relevant forecast periods where the seasonal climate information could be useful (same remark like previous lines: for whom, for what action, in which time/deadline).
Annex 1 - Joint Activities WG2&3 for 2007-2008
Main activity: verifications of forecasts based on General Circulation Models’ outputs.
Goals: Evaluate the performance of the forecasts based on GCM outputs at sub-national (zone) and station levels; select best prediction system for each zone/group of stations; Evaluate potential for forecasting other seasons (where relevant); practice techniques and interpretation of the main verification scores introduced at the pre-PRESAO 10 training session; produce a technical report summarizing major findings in each country (to be presented at PRESAO 11).
Possible other activities (only when the main one has been completed):
- compile definitions of onset of the rainy season used in different contexts and relative decisions ( who, when, which lead-time)
- gather information about potential needs/users in other seasons (where relevant): who, when , what kind of decision, lead time.
Verification of GCM-based forecasts:
Seasonal Rainfall forecasts are issued using statistical methods to translate GCM outputs into sub-regional or station rainfall probabilities (MOS correction). Currently low-level winds (as monsoon proxy) and seasonal rainfall amounts are the main GCM outputs used as predictors. The activity consists on using CPT and the RPC methods herein to evaluate the skill of such forecast to predict sub-regional or station rainfall probabilities using reliability diagram, ROC and RPSS.