Ensembles, Forecasts and Predictability
147
the thick line is the observed evolution of the NIN03 SST. The panels correspond
to ensembles initialized in January 2000, November 1999 and September 1999.
The ensembles include the true evolution of the system and this is encouraging, but
the spread of the ensemble is substantial and the members predict, in the January
case for instance, from zero anomalies to one degree anomalies. They all agree to a
wanning trend, but there is a large uncertainty about the magnitude of the wanning. Reflecting that then these anomalies must be transfonned into atmospheric
anomalies by the response ofthe atmospheric model, it is possible to have a feeling
of the long road stilllaying ahead of us.
8.7 Conclusions
The seasonal forecasts are the first examples of the kind of environmental forecasts we are going to expect in the next century. There is an increasing demand for
forecasts of environmental parameters, like ice condition on roads, crop yield
results, even price of agricultural and energetic commodities that reflect the influence of climate variability. To respond to this growing demand we will have to
evolve our forecasting models from atmospheric and/or ocean models to Earth system models capable of giving estimates ofthe probability distribution ofthese various parameters.
As we move steadily away from parameters govemed by well-defined mathematical and physical laws there will be a growing need to introduce statistical
methods to extract the infonnation from our predictions. We will make a transition
from detenninistic forecasts to estimation of probability distribution of environmental parameter, using ensemble and MonteCarlo methods. The task is harder and
harder, but if we were able to dodge the predictability limit, at the end of the road
the reward will be big and exciting.
147
the thick line is the observed evolution of the NIN03 SST. The panels correspond
to ensembles initialized in January 2000, November 1999 and September 1999.
The ensembles include the true evolution of the system and this is encouraging, but
the spread of the ensemble is substantial and the members predict, in the January
case for instance, from zero anomalies to one degree anomalies. They all agree to a
wanning trend, but there is a large uncertainty about the magnitude of the wanning. Reflecting that then these anomalies must be transfonned into atmospheric
anomalies by the response ofthe atmospheric model, it is possible to have a feeling
of the long road stilllaying ahead of us.
8.7 Conclusions
The seasonal forecasts are the first examples of the kind of environmental forecasts we are going to expect in the next century. There is an increasing demand for
forecasts of environmental parameters, like ice condition on roads, crop yield
results, even price of agricultural and energetic commodities that reflect the influence of climate variability. To respond to this growing demand we will have to
evolve our forecasting models from atmospheric and/or ocean models to Earth system models capable of giving estimates ofthe probability distribution ofthese various parameters.
As we move steadily away from parameters govemed by well-defined mathematical and physical laws there will be a growing need to introduce statistical
methods to extract the infonnation from our predictions. We will make a transition
from detenninistic forecasts to estimation of probability distribution of environmental parameter, using ensemble and MonteCarlo methods. The task is harder and
harder, but if we were able to dodge the predictability limit, at the end of the road
the reward will be big and exciting.