RT6 WP 6.3 Home
Links to other Research Theme (RT) webpages: RT1 | RT2A | RT2B | RT3 | RT4 | RT5 | RT6 | RT7 | RT8
RT6 Home | News Archive | Meetings | Reports | Documents and Downloads | Members' Site | Participants | Links to other projects WP6.1 | WP6.2 | WP6.3 | WP6.0 workpages within RT6 with the details of their activity

This page is for WP 6.3 Impact modelling at seasonal-to-decadal time scales

Leader: UNILIV (Morse). Participants: UREADMM (Wheeler/Slingo), ARPA-SIM (Marletto), JRC-IPSC (Genovese), METEOSWISS (Liniger/Appenzeller), LSE (Smith), 
FAO (Gommes), WINFORMATICS (Norton), IRI (Thomson), EDF (Dubus), DWD (Biermann).

Information on the activity within the WP can be found below.

The first section gives the 60 month overview followed by the 18 month objectives and tasks. The section is completed by a list of deliverables and milestones.

WP 6.3 subpages:

The primary objective of this WP is to integrate application models within a probabilistic ESM system and within RCM systems. This integration links the human dimension to earth 
system modelling and allows subsequent socio-economic evaluation of the ENSEMBLES EPS. A number of research and development tasks are required to allow an effective integration
of the modelling systems. Output from the integrated EPS/application model runs will be validated in WP5.5 against ERA-40 and where available gridded station data driven application 
model runs.

State of Knowledge

The current state of the art for quasi-operational impact models running for the forthcoming season is that they make use of the observed climate through the first part of the season or 
through the preceding season. As an example wheat yield predictions in Europe are made using a dynamical crop growth model which is run to a given date using climate observations 
and then an end of season yield prediction is made through a statistical model relating crop development phase, at that date, to previous known wheat yields. A similar approach can be
made for malaria prediction using the seasonal rainfall totals through the early part of the rainy season to make a prediction, using a statistical model, of the forthcoming number of malaria 
cases that follow at the end of the rainy season. In both applications the lead-time is limited and the predictions become more reliable the later in the season they are produced. Probabilistic 
information is routinely used in weather risk management models. However this is mostly confined to distributions from climatological records, with occasional input from the mean of seasonal
forecasts. The tropical yield crop model GLAM has successfully been used to simulate yields over large areas in India using observed gridded data and reanalysis. Some preliminary work
has also been undertaken using the DEMETER ensembles for the period 1987-98.

During DEMETER a small number of application models were developed or modified to use the seasonal probabilistic forecasts to drive, daily time step, impact model integrations. In 
DEMETER the range of impact models run, the regions covered and number of ensemble members utilised was limited. Questions that arose during DEMETER regarding downscaling, 
bias correction and how to interpret the probabilistic impact model output were only partially addressed and these will be more fully investigated within ENSEMBLES. In ENSEMBLES,
a number of new impact models will run addressing a much larger potential user community.

Novel Questions

ENSEMBLES will attempt to answer a number of novel research questions

i. How to maximise the integrated model system skill using techniques such as ensemble dressing,

ii How to quantification the existing skill within an integrated modelling system,

iii. The estimation of the skill required from the driving seasonal forecasts to allow a skilful seasonal impact forecasts.

Tasks

Task 6.3.a: Downscaling and bias correction for ensemble hindcasts: This work is part of RT2B and the research and development will require the seasonal impacts groups to define their 
requirements and work in collaboration with RT2B partners. The nature of the downscaling requirements will not be uniform for each application model. The relative improvements of the 
various downscaling schemes will be assessed against un-downscaled ensembles for the impacts models.

Task 6.3.b: Integration of seasonal-to-decadal application models within an EPS. Initially the DEMETER datasets will be used to integrate the application models. As ENSEMBLES 
EPS (WP1.5 and WP2A.1) and RCM (WP2B.4 and WP3.5) output become available they will integrated into the modelling system. Probabilistic integrated application model runs will 
allow the 'skill in hand' of the current DEMETER EPS to be evaluated within WP5.5. The current skill-in-hand will act as the baseline from which to compare the developments produced 
within ENSEMBLES.

Task 6.3.c: Assessment of GCM vs. RCM driven seasonal-to-decadal application models. Probabilistic high-resolution regional climate scenarios at seasonal to decadal timescales from 
WP2B.4 and very high resolution RCM-derived scenarios for non-European areas from WP3.5 will be used to drive the application models and comparison will be made with GCM driven
application models in the same regions. The results will be evaluated in WP5.5.

Task 6.3.d: Gaining the maximum skill from an EPS seasonal-to-decadal scale integration: EPS is a new technology particularly when used for driving application models. Research and 
development will be undertaken to see how a forecast PDF can be utilised to give the maximum skill in the application forecasts e.g. ensemble dressing.

Task 6.3.e: Quantification of EPS skill requirements at seasonal-to-decadal timescales for application models. The integrated application model/ EPS model system is almost certainly 
non-linear for most cases. The defining of skill requirements from the application forecasts will link to the socio-economic investigations RT7. The skill definitions from the application 
groups will start to define the level of skill required from the driving seasonal-to-decadal EPS hindcast. Required skills levels in the hindcast will differ between different applications but 
these required levels would become the benchmark for targeted forecast skill in current and future EPS.

Proposed Examples

In addition to the methodological and theoretical developments outlined in the tasks above the integration of the impacts models within the ENSEMBLES EPS will lead to a number of 
measurable outputs including as examples

• Probabilistic estimates of wheat yield for Germany, Spain, France and Hungary will be driven using ENSEMBLES seasonal hindcasts and will be validated against EUROSTAT crop yield data and compared with the forecasts from the current state of the art MARS Crop Yield Forecasting System.
• Probabilistic hindcasts will used in a weather derivative model to calculate the expected value of standard swap and option weather derivative contracts at a number of European cities. The skill of the forecasts compared to a forecast of climatology will be validated against ERA-40 data interpolated onto the station locations.
• The General Large-Area Model for annual crops (GLAM) will be used to simulate the yield of major legume and cereal crops in India. Observed district-level yields will be used to assess the skill of the model output and examine the issue of spatial scale of predictions. Groundnut will be the first crop to be studied as the district-level yields for this crop are already available. Efforts will also be made to assemble datasets for some of the major tropical cereals.
• Probabilistic estimates of regional or pan African potential malaria transmission will be simulated driven by ENSEMBLES seasonal hindcasts and compared to an assumed 'perfect forecast' (probably using ERA-40 reanalysis). The perfect forecast approached will be utilised because pan African or even regional African interannual clinical malaria data are generally unavailable.
• A probabilistic heating degree-day prediction model will be driven with ENSEMBLES seasonal hindcasts with the output assessed for use in terms of prediction of weather risk and weather related insurance risk. ity for each deliverable and milestone will appear here with links to RT6 WPs

Notes: Deliverable Number 6.0 basic site up and running by month 3 ongoing improvements and expansion through course of ENSEMBLES

WP six monthly reports will appear here...