Dr Mark Cresswell Statistical Forecasting [Part 2] 69EG6517 – Impacts & Models of Climate Change.

Dr Mark Cresswell

Statistical Forecasting [Part 2]

69EG6517 – Impacts & Models of Climate Change

Lecture Topics• Introduction

• SARCOF

• ACACIA

• Environment Canada

• UKMO forecasts

• Ocean models

• Summary

“The days are over, of hanging out the seaweed, examining the size of molehills, or studying animal entrails for portents ofcoming tempests – that is unless the computers are down!”

Palmer, 2000

• Statistical forecasts have often been the most commonly used forecasting method

• Only since good weather records (consistent observing) began in the mid 1700s did it become possible to exploit SF properly

• It has been the rise of computer technology and the associated growth of dynamical NWP models that has meant SF is less commonly used today

Introduction

• Where there is strong causality between a particular forcing (such as SST anomalies) and resultant weather patterns we can use SF very effectively

• Such forecasts work best in the Tropics and southern hemisphere where SST boundary conditions are more important

• Africa and South America are continents that use SF most often

Introduction

• Southern Africa Regional Climate Outlook Forum (SARCOF)

• Formed in early 1990s with UK, EU and US foreign aid (EU ENRICH)

• Generate “consensus” forecasts based on mainly SST conditions in Atlantic, Indian and Pacific Oceans

• Pre-season and post-season venues rotate in different Southern Africa countries

SARCOF

SARCOF

Rainfall maps generated – prob. terciles for Above/Normal/Below

• Method uses a blend of statistical forecasts (mainly using CCA and multi-variate regression) and dynamical NWP

• Representatives from each country (NMS) “fine tune” the tercile probabilities according to their own knowledge and experience (Analog forecasting)

• Zones of similar probabilities are grouped together

SARCOF

• Representatives of each weather-sensitive sector (agriculture, health, hydro-electric and tourism etc) meet to discuss the impacts of the coming (or past) rainfall season

• ECMWF, UKMO, NOAA-OGP and SAWB all contribute NWP model data as well

SARCOF

ACACIA

A Consortium for the Application of Climate Impact Assessments

Uses statistical methods to "downscale" GCM data to regional and local scales for use in impacts models in areas such as hydrology, agriculture, and ecosystem studies

• In 1996, Environment Canada began to produce surface air temperature and precipitation anomaly outlooks up to 3 to 9 months ahead using a statistical model based on Canonical Correlation Analysis (CCA)

• The forecasts are made at Canadian station locations up to nine months in advance

• The analysed field of sea surface temperature (SST) anomalies over the previous twelve months provides the forcing

Environment Canada

Environment Canada

• The Met Office has a long established statistical forecasting background

• UKMO weather records extend far back and cover many different regions of the world and the oceans – primarily due to the growth of the British Empire

UKMO

• Data from Africa, India, Asia, Canada and Australia/NZ provide a good historical database to explore global weather phenomena

• The UKMO model has been seen to work well in North East Brazil where forecasts of precipitation show good predictive skill

• Tropical North Africa uses March/April SST anomaly patterns to predict July rains

UKMO

• In tropical East Africa the December “short rains” may be forecast using linear regression based on SST anomalies – both globally and locally (Indian Ocean)

• These case studies are often held up as a “yardstick” for dynamical models to try and match in terms of skill and reliability

UKMO

• Given the fact that Sea-Surface Temperature (SST) conditions are a major source of climate predictability it is not surprising that the prediction of ocean conditions is important

• SST conditions vary less through a time period than one would observe of the atmosphere and land components

Ocean Models

• Palmer et al. (1994) has shown that model simulations of climate using observed SSTs were far superior than those using either climatological SSTs or persistence

• A significant improvement in predictability can be gained during ENSO years. Such anomalies can influence local and global weather patterns – acting to bias the future climate towards a particular state

Ocean Models

• Certain regions of the major oceans are intensively monitored – they provide a diagnostic of change

• The NINO-3 region of the Eastern Equatorial Pacific is one such region

Ocean Models

Ocean Models

Ocean Models

• Statistical models are amongst the oldest methods of generating predictions of future climate change (at different time-scales)

• They do not require large supercomputers

• Their predictions are constrained to the extremes of events observed during their training data

• Most success is in tropical local-scale models rather than global predictions

Summary