TY - UNPB

T1 - Validation in models of climate change and forecasting accuracy

AU - Fildes, R A

AU - Kourentzes, N

PY - 2010

Y1 - 2010

N2 - Forecasting researchers, with few exceptions, have ignored the major forecasting controversy facing the world in the early 21st Century: namely, whether and by how much the earth is warming; and the role of climate modelling in reaching any conclusions on this challenging topic. In contrast, scientists from climatologists, through hydrologists to fluid dynamicists, have engaged in this modelling and forecasting controversy. In this discussion paper, we first describe briefly the atmospheric-ocean general circulation models (AOGCM) used in most climate forecasting, in particular by the Intergovernmental Panel on Climate Change (IPCC). This discussion paper takes a forecaster’s perspective in a review of established principles for the validation of such large-scale simulation models. One key principle is that such models should reproduce the ‘stylised facts’ or 'dominant modes' of dynamic behaviour that characterize key model outputs: here taken as the aggregate annual changes in world and regional temperatures. By developing various time series models and input-output dynamic models of atmospheric carbon dioxide and temperature that capture current trends, we compare the results with dynamic forecasts produced by one well-established AOGCM model, the Hadley Centre’s HadCM3. Time series models are shown to perform strongly and by using encompassing tests, structural deficiencies are identified in the AOGCM model and its corresponding forecasts. The paper concludes with some implications for climate modellers when producing decade-ahead forecasts from global climate models. If forecasting accuracy is the focus, methods that combine standard time series methods with the structure of a GCM should be used. This has implications for the effectiveness of control policies, focussed on carbon dioxide emissions alone..

AB - Forecasting researchers, with few exceptions, have ignored the major forecasting controversy facing the world in the early 21st Century: namely, whether and by how much the earth is warming; and the role of climate modelling in reaching any conclusions on this challenging topic. In contrast, scientists from climatologists, through hydrologists to fluid dynamicists, have engaged in this modelling and forecasting controversy. In this discussion paper, we first describe briefly the atmospheric-ocean general circulation models (AOGCM) used in most climate forecasting, in particular by the Intergovernmental Panel on Climate Change (IPCC). This discussion paper takes a forecaster’s perspective in a review of established principles for the validation of such large-scale simulation models. One key principle is that such models should reproduce the ‘stylised facts’ or 'dominant modes' of dynamic behaviour that characterize key model outputs: here taken as the aggregate annual changes in world and regional temperatures. By developing various time series models and input-output dynamic models of atmospheric carbon dioxide and temperature that capture current trends, we compare the results with dynamic forecasts produced by one well-established AOGCM model, the Hadley Centre’s HadCM3. Time series models are shown to perform strongly and by using encompassing tests, structural deficiencies are identified in the AOGCM model and its corresponding forecasts. The paper concludes with some implications for climate modellers when producing decade-ahead forecasts from global climate models. If forecasting accuracy is the focus, methods that combine standard time series methods with the structure of a GCM should be used. This has implications for the effectiveness of control policies, focussed on carbon dioxide emissions alone..

KW - validation

KW - long ranger forecasting

KW - simulation models

KW - Global circulation models

KW - neural nets

M3 - Working paper

T3 - Management Science Working Paper Series

BT - Validation in models of climate change and forecasting accuracy

PB - The Department of Management Science

CY - Lancaster University

ER -