Australian Government - Bureau of Meteorology Home | About Us | Contacts | Help | Feedback |

Global | Australia | NSW | Vic. | Qld | WA | SA | Tas. | ACT | NT | Ant. |
Weather & Warnings | Hydrology | Climate | Numerical Prediction | About Services | Learn About Meteorology | Registered User Services |

Bureau of Meteorology Research Centre link image
BMRC is now part of CAWCR: The Centre for Australian Weather and Climate Research.
For more information on The Centre please go to http://www.cawcr.gov.au

Climate Dynamics HOME Climate Dynamics STAFF Climate Dynamics EXPERIMENTAL RESULTS Climate Dynamics COLLABORATIONS Climate Dynamics ANNUAL REPORT

A Downscaling Technique

Constructing Regional Climate Change Scenarios for Australia
Introduction
A Statistical Model (SM) has been developed to downscale large scale predictors given by Global Climate Models (GCMs). This is a complementary approach to the dynamical modeling of regional climate change using high resolution nested models. It allows to bridge the scale difference (“downscaling”) between coarse grid Coupled Atmosphere-Ocean GCMs and the finest temporal and spatial scales required for regional and environmental impact studies of climate change.

The analogue technique was chosen for this study as it has proven successful in the past for mid-latitude climate and in particular for forecasting in Australia. Furthermore, statistical techniques allow the reconstruction of time series and therefore consider extreme events such as spells of anomalous temperatures.

Analogues are based on large scale atmospheric predictors (e.g. Mean Sea Level Pressure, geopotential height or atmospheric thickness) for which GCM are considered to be more reliable than for grid-average of surface observations. Predictor series are then obtained from these functions with a high temporal distribution (e.g. daily). Hence, climatic issues related to extreme events can be investigated (e.g. frost days, anomalous spells and return of extremes). This is a key area of research where insufficient effort has been devoted so far.

The aim of this research is to:

  • Develop a system to statistically downscale information provided by climate models at local observational level.
  • Obtain sufficient skill to match local extreme temperature observations in two major Australian agricultural areas.
  • Insure that temperature series constructed with the statistical model are more realistic than direct outputs from climate models.
  • Assess the usefulness of such reconstructed series for impact studies in particular for anomalous spells
The Method Used
Analogues have been used by meteorologists for forecasting as a recognition tool or simply as a main driver of the subjective forecaster process. More recently, analogue techniques have been successfully applied for downscaling in climate studies.
They have been shown to be successful at mid-latitude, producing unbiased series with the right spatial correlation structure. The latter is a very important feature in Australian landscape.

Finally the experience, gained by the Australian Bureau of Meteorology since 1980, while using analogues as a forecasting tool has shown the potential of this technique for many variables.
The available dataset is split in two sub sets of independent data: 1970-1981 and 1982-1993. The SM is applied on the second sample, the temperature series are reconstructed using analogue situations extracted from the first period.
The model is adapted for each season and location and parameters are set up during this validation. The main validation tool is the comparison of the reconstructed series with the observed one. Differences of mean and variance are compared, as well as the correlation between the two series and the Root Mean Square Error (RMSE).
The Local Predictants
In this particular project emphasis has been put on local surface air temperature a key parameter for impact studies on ecosystem, agriculture and human health.
In Australia, it has been proven that climate variability and global changes have a strong impact on ecological and economic fluctuations.
Two areas of interest have been selected according to their importance in term of agricultural production: the South-West Corner (SWC) of Western Australia and the Murray-Darling Basin (MDB). The latter is of particular concern as it appears to be affected by abrupt changes in long term climatologies.
The density of high quality data in the areas of interest is rather good and superior to most other parts of Australia. Most of the 29 and 22 stations used in the MDB and the SWC are part of a subset of high quality record. This set of high quality observations covering the entire continent has been used to assess long term climatic trends over Australia.


Extraction of temperature series from the complete historical dataset.
Selection was based on the quality of the observations and ensure a comprehensive
coverage of the two areas of interest.

Some additional stations have been added to extend the coverage. These extra stations are of high quality for the entire period of interest 1970-1993. Amongst these data, five stations in each domain are of particular interest as temperature is recorded every 3-hours The reliability of the data is high, very few missing data are reported. 50% of the stations report less than 1% of missing data, however a few stations report up to 10%. Results indicate no dependence of the statistical model’s skill on the percentage of missing data.
The Large Scale Predictors
The database for predictors, METANAL, is derived from historical synoptic analyses.
From 1970 to 1993, at 00 UTC and 12 UTC, grid point analyses for Australian region have been archived, by the Australian Bureau of Meteorology. The analyses are available on a regular 1.5 x 1.5 grid from 50 S to 10 S and from 90 E to 170 E.
This dataset contain Mean Sea Level pressure (MSLP), geopotential at various height, Temperature at 850 hPa and wind speed at 500 hPa.

Several atmospheric predictors were tested, the selection was based on the following criteria: 1. Being realistically simulated by GCMs
2. Having a strong predictive skill for surface temperature
3. Being complementary to the other atmospheric forcing

The best combination was found to be MSLP and Temperature at 850 hPa. The domain size applied to predictors in order to optimize recognition of synoptic systems from unnecessary noise is a key parameter of the SM. Both primary fields and their decomposition in to Principal Components (PCs) were used. In the latter case the number of PCs used varies and has been optimized for each domain. This number decreases with the size of the domain as more variance is explained by fewer leading PCs. In large domains PCs, by filtering synoptic signal from remanent noise, increase the predictive skill from the raw fields. But in smaller domains, primary variables yield to better results.

The search for the best matching analogue, using Euclidean distance, is based on several metrics taking into account the weather situation on the day or over several days and thus the evolution of the atmosphere. The latter has been particularly useful in improving the representation of anomalous spells as it partially incorporates the auto-correlation of surface temperature.


Different sizes were tested, ranging from a minimal size just encompassing
the domain of interest to the entire METANAL grid.
Application to Climate Model
The SM, developed and adapted during the validation is then applied to GCM predictors. No tunning is made on the parameters defined previously. The climate run was from a 1979-1988 AMIP simulation. The model is a state of the art atmospheric model developed by the Bureau of Meteorology and Research Center (BMRC).
The horizontal resolution of the model provides information for grid boxes of approximately 250 Km in latitude by 350 Km in longitude over Australia.
Analogues are chosen amongst the entire set of available observations: 1970-1993. The benefit of using a downscaling method is seen by comparing the reconstructed series with the 2m-Temperature provided by the Global Model. Biases average for each season and per area of interest show the large reduction of GCM errors.


Probability Distribution Function: the major bias observed of Climate Model surface Temperature are corrected and the PDF of the reconstructed series match the observed ones. Anomalous hot (HSD) and cold (CSD) spells: the threshold used are more realistic for the series obtained from the analogues. In most cases the anomalous spells of the reconstructed series are well in agreement with those from the observations.
Validation of the Statistical Model

Correlation between the observation during 1982 to 1993 compared with the reconstructed series based on analogues selected during the 1970-1981 period. Correlation in summer are higher in the SWC than in the MDB in particular in the northern, more Tropical part of the Basin, and for T_max than for T_min.


Biases are very small, global values are averaged from contrasting individual results.
No particular positive or negative trend is seen.


Although the values are small compared to the variance of the original series, reconstructed series using analogues underestimate the observed variance.


In order to better assess the results of the SM, skillscores were compared with persistence. The SM was able to surpass persistence in most cases. Best correlation are achieved for the transition seasons Spring and Autumn.
Conclusions and Future Plans
In this study, a statistical model was developed to downscale large scale predictors. Such a technique complements dynamical approaches for climate change studies performed with CAOGCM and allows finer time and spatial resolutions for impact studies.
  • a SM, based on the recognition of analogues, was developed for two agricultural areas of Australia: the Murray-Darling Basin and the South West corner of the continent.
  • The SM was found to be more skillful than persistence in all cases apart for winter (JJA) in the MDB. In the MDB, results are contrasted between the southern part where analogues are successful and the Northern Tropical part where skill is low.
  • Extreme events such as anomalous spells were found to be well captured by the SM. The quality of the spell reproduction is dependent on the skill achieved.
  • Those results were confirmed when GCM outputs were used as predictors. The ability to reproduce a particular PDF is maintained with no particular bias on the mean. Furthermore the spells are similarly captured.

In the future, further work is needed to improve the SM and compare this approach with other downscaling methods:

  • Using 40 years, of re-analyses available from NCEP or ECMWF would give access to more predictors and would train the model on a dataset with different properties from that used for validation. This is a key validation prior to downscaling of climate changes.
  • Some intercomparisons are planned in collaboration with other groups interested by downscaling over Australia, and using a non-homogeneous hidden Markov Model. Comparison with PDF obtained from a dynamical downscaling model such as a nested regional climate model is also planned.
  • The SM is to be applied to transient scenarios proposed by coupled models. A particular emphasis will be put on applying analogues to several coupled models in order to investigate if statistical downscaling reduces the scatter observed between climate change scenarios at regional scale.
  • Impact studies are to be conducted in collaboration with other groups, in particular for areas which have shown high skill, such as the lower part of the Murray-Darling Basin.
   
For more information, please contact:
Dr Bertrand Timbal (Email: b.timbal@bom.gov.au)
BMRC, GPO Box 1289, Melbourne, VIC 3001, AUSTRALIA
http://www.bom.gov.au/bmrc/clfor/cfstaff/btimbal.htm
BMRC, Bureau of Meteorology Research Centre
Department of Environment and Heritage
Commonwealth of Australia

 


  Bureau Home   ||   BMRC Home  ||  Search  ||  Contact BMRC Webmaster
Experimental results described in these pages are from research systems developed in BMRC and are not part of the Bureau of Meteorology's operational products & services.


Home | About Us | Learn about Meteorology | Contacts | Search | Help | Feedback
Weather and Warnings | Climate | Hydrology | Numerical Prediction | About Services | Registered Users | SILO
© Copyright Commonwealth of Australia 2008, Bureau of Meteorology (ABN 92 637 533 532)
Please note the Copyright Notice and Disclaimer statements relating to the use of the information on this site and our site Privacy and Accessibility statements. Users of these web pages are deemed to have read and accepted the conditions described in the Copyright, Disclaimer, and Privacy statements. Please also note the Acknowledgement notice relating to the use of information on this site. No unsolicited commercial email.