About the long-range forecasts

About the long-range forecasts

About extreme rainfall and temperture long-range forecasts

Video about long-range forecasts   

Free training: Interpreting climate outlooks (long-range forecasts)

Long-range forecast video, summaries, maps and graphs

• The overview includes forecast highlights, climate influences, season maps, and a fortnightly video.
• Summaries include key rainfall and temperature long-range forecasts across Australia.
• Maps of Australia and the global views can be zoomed for detail. Maps include rainfall scenarios, difference from average, extremes, medians, and model accuracy.
• NEW maps include the chance of particular 3-day rainfall totals.
• Graphs show long-range forecasts for any location in Australia. Search or tap maps for your location.
  • Bar graphs show likely long-range forecast ranges below key forecast and historical values.
  • Timeline graphs show observations and forecasts over weeks or months at your location.
  • Line graphs are available for rainfall, and show the chance of exceeding a range of totals.
  • Tap/click graphs for rainfall or temperature details.

Long-range forecast maps and graphs

• The rainfall and temperature climate long-range forecast maps and popup location details show the percentage chance of experiencing wetter/drier (and warmer/cooler) than median¹ weather at different timescales for the upcoming four months. forecasts are available for weeks, fortnights, months and three-months ahead.
• For the monthly to seasonal timescales, the forecast issued closest to the end of a calendar month provides the Bureau's best advice on the likely temperature and rainfall patterns for the three months ahead. For example, forecasts issued in the last week of June, for the three months July to September, will generally be more accurate than forecasts issued earlier in June.
• More maps:
  • Rainfall: the likelihood of exceeding particular totals (e.g., the chance of receiving at least 10 mm), as well as the specific chances (e.g., rainfall totals that have a 75% chance being exceeded). These types of forecasts are particularly useful for the weekly and fortnightly timescales.
    Details: Rainfall scenarios
  • Rainfall: the likelihood of exceeding particular totals over three days. Available for the weekly and fortnightly timescales. Useful for applications requiring or avoiding sustained wetness.
    Details: Three-day totals rainfall scenarios
  • Temperature: difference from average (anomaly) maps. Available for the weekly and fortnightly timescales. They show how far above or below the 1981–2018 average the temperature is likely to be (1981–2018 is the model period; see Medians tab). The values provided are the middle of the predicted scenarios—also known as the ensemble mean.
  • Temperature: the likelihood of unusually cool/warm or wet/dry conditions is provided by the chance that the forecast will result in conditions in the top or bottom 20% of historical observations.
    Details: Extremes

Differences between 'Long-range forecasts' and 'Weather forecasts'

• Model: The long-range forecasts are generated from the Australian Community Climate Earth-System Simulator–Seasonal (ACCESS-S), the Bureau's dynamical (physics based) climate model.
  Details: Long-range forecast model
• While weather forecasts can tell you what the temperature will be tomorrow and how much rain to expect, long-range forecasts cannot be this specific. This is because the further we look ahead, the more small random changes can amplify into different weather patterns. However, longer-term forecasts of seasonal statistics, such as whether rainfall or temperature will generally be above or below median, are possible to accurately predict. Advances in modelling and computing power have allowed us to look at shorter timescales, making possible accurate predictions for sub-monthly timescales.
  Details: Past accuracy
• Probability-based long-range forecasts are designed to be used as one of several planning tools within risk management and decision-making. The greatest benefits of using Bureau long-range forecasts will accrue from use over a number of seasons or years.
  Read our blog: What to expect from the climate when the forecasts are neutral

Long-range forecast update schedule

• One and two week forecasts: Daily around 3pm AEST/AEDT.
• One and three month forecasts: Thursdays at 3pm AEST/AEDT.
• Climate and water long-range forecast videos: twice a month on Thursdays.
• Download schedule as calendar file (ICS).

Notes

¹ Median is a measure of what is considered typical rainfall or temperature for a specific location; similar to mean. The median is the middle value when a set of values are ranked from lowest to highest. Due to the high variability of rainfall, and that in some locations just one or two extremely wet years can substantially change the overall average, using the median is the best representation of typical rainfall that would be expected.
Details: Medians

Rainfall long-range forecasts

• Long-range forecasts are given as a probability (or chance) of exceeding a specific threshold. For the Bureau’s rainfall long-range forecast, it is provided as the chance of rainfall being above median, expressed as a percentage.
• For those interested in specific rainfall amounts (e.g. 200 mm for the season), or who make decisions at specific probabilities (e.g. if there is a 75% chance it will be drier than average), the Bureau uses the spread of possible rainfall amounts from the long-range forecast model to transform output into rainfall scenarios that can be viewed in a different way:
  • Chance of at least: the chances that rainfall for the selected forecast period will exceed particular thresholds, e.g., chance of at least 200 mm over the coming three months, or 10 mm in a week.
  • Rainfall scenarios: rainfall amounts that are likely at a particular percentage chance, e.g., 25% chance of receiving the given rainfall amount for the period.
• In the shorter-term, for those interested in specific rainfall amounts within a short period of time can look at 3-day totals maps, which indicate the chance of receiving a particular rainfall amount within 3-days over a week or fortnight.
• The long-range forecast data is prepared on a 60 km by 60 km grid. For the 'Chance of at least', 'Rainfall scenario', and '3-day totals' maps, a statistical technique is used to interpolate to a smaller 5 km by 5 km region. This is the resolution of these maps.

Rainfall scenario maps

• Rainfall scenario maps present the long-range forecast information as rainfall amounts which have a 75%, 50% or 25% chance of occurring. To illustrate, the map on the right shows the rainfall forecast from 1 September to 30 November 2019. The colours on the map show the amount of rainfall (mm) that has a 75% chance of occurring during this period. The location highlighted with the black circle in northwest Victoria, has a 75% chance of receiving at least 25 mm and possibly up to 50 mm of rain.

Chance of at least maps

• Chance of at least maps also show the seasonal rainfall forecast, by displaying the chance of receiving a specified rainfall amount. You can choose from 15 different rainfall amounts (in mm) for the different time periods.
• The adjacent example map shows the chance of receiving a total rainfall amount of at least 50 mm during 1 September to 30 November 2019. The colours on the map show the percentage chance of 50 mm of rain occurring. The location highlighted with the black circle in northwest Victoria has a 50 to 75% chance of receiving 50 mm of rain during the period.
• Chance of at least rainfall long-range forecast maps are consistent with the Rainfall scenarios. For the same location in the examples above, the chance of at least 50 mm of rainfall is 50 to 75%, which is consistent with the rainfall scenario which shows a 75% chance of at least 25 mm, and possibly up to 50 mm.

Chance of 3-day totals maps

• Chance of 3-day totals maps show the percentage chance of receiving at least the selected total over three consecutive days within the forecast period. These are often called rainfall bursts. Forecast periods include weeks and fortnights.
• Four specific thresholds are available: 15 mm, 25 mm, 50 mm or 75 mm. These four available quantities align well with agricultural and livestock requirements. The three higher thresholds approximately represent 1 to 3 inches.
• The 15 mm and 25 mm thresholds are typically better suited for southern regions from late autumn to early spring, while the 50 mm and 75 mm thresholds are more suited to the northern tropics from late spring to early autumn (i.e., the northern Australian wet season).
• The example map shows the chance of receiving a total rainfall accumulation of at least 15 mm spread over three consecutive days during the week of 25 June to 1 July 2022. The colours on the map show the percentage chance of at least 15 mm of rain occurring. The location highlighted with the black circle in south-west Victoria has a 40% to 60% chance of receiving 15 mm of rain across three days within the period.
• The climate model uses a 60 km by 60 km grid. These forecasts are calibrated to observations on a 5 km by 5 km grid.

Rainfall and temperature medians

• Median is a measure of what is considered typical rainfall or temperature for a specific location - similar to the mean. Due to the high variability of rainfall, and that in some locations, one very wet year can substantially change the overall mean, using the median (the middle number when you rank past rainfall or temperature from lowest to highest) is the best representation of typical rainfall.
  
• The medians (which are the same as the 50th percentile) for Bureau long-range forecasts are calculated over the 1981 to 2018 period, as this is the period for which the climate model has been run in historical (or hindcast) mode.
• Median maps for the 1981 to 2018 period for all months and seasons are available from the long-range forecast Map archive and Average maps. The median maps on the long-range forecasts website will differ slightly from other median maps on the Bureau's website. This is because the dynamical model forecasts use a period of 1981 to 2018 to calculate medians, while other median maps typically use the full Bureau climate record (which extends for more than 100 years).
• The 20th percentile maps are also provided for the 1981 to 2018 period. Rainfall totals or temperature which falls below the 20th percentile are considered unusually dry or unusually cool by the Bureau, with the 20th percentile one of the two thresholds used in the extremes maps. This means that the forecast extremes fall into the bottom 20% of observations. Typically, it would be expected that these would occur roughly once every five years under a normal climate so for example, the coolest June out of five years, or the driest spring in five years. The 20th percentile maps are available on the long-range forecast Average maps page.
• Similarly, the 80th percentile maps are also provided for the 1981 to 2018 period. Rainfall totals or temperature which are above the 80th percentile are considered unusually wet or unusually warm and are the upper threshold used in the extremes maps. This means the forecast extremes fall into the top 20% of observations. Typically, it would be expected that these would occur roughly once every five years under a normal climate so for example, the warmest June out of five years, or the wettest spring in five years. The 80th percentile maps are available on the long-range forecast Average maps page.

Chance of extremes maps

• Long-range forecasts are able to be expressed as a percentage likelihood of extreme rainfall or temperature for those interested in possible weather extremes.
• Chance of extremes maps show the likelihood of unusually wet or dry, or warm or cool conditions. They are displayed by the chance that the long-range forecast will result in rainfall or temperature in the top or bottom 20% of historical observations for the selected forecast period. In addition to the percentage chance of this occurring, the legend also shows how this could be displayed as the number of times more likely. For example, a 40% chance of being in the lowest 20% of rainfall records is also two times the usual level of likelihood. A 60% chance would mean it is three times as likely.
• Rainfall totals or temperatures which fall below the 20th percentile are considered unusually dry or unusually cool by the Bureau. This means the forecast extremes fall into the bottom 20% of observations. Typically, it would be expected that these would occur roughly once every five years under a normal climate. For example, the coolest June out of five years, or the driest spring in five years.
• Rainfall totals or temperature which are above the 80th percentile are considered unusually wet or unusually warm. This means the forecast extremes fall into the top 20% of observations. Typically, it would be expected that these would occur roughly once every five years under a normal climate. For example, the warmest August out of five years, or the wettest spring in five years.
• The climate model uses a 60km by 60km grid. These forecasts are calibrated to observations on a 5km by 5km grid.

Long-range forecasts for your location

Summary values and bar graphs showing likely long-range forecast ranges

• The location-based bar graphs show the forecast probability of rainfall or temperature being in a particular climatological range for your selected location. This includes the likelihood of being in the bottom 20% (decile 1 and 2) of historical records, the top 20% (decile 9 and 10), or the three ranges in between: decile 3 and 4; decile 5 and 6; and decile 7 and 8.
• The long-term average probability ("usual chance") for each category is 20%, shown by the horizontal dashed line. If the grey bar (the forecast) is above this line it means there is a stronger than usual chance that category will occur. If it is below the line there is less chance than normal.
• In the example shown here for July to September rainfall for Canberra, the odds are stacked towards having a wetter season than usual in Canberra, with more than double the usual likelihood of having decile 9 or 10 rainfall (i.e., greater than 198.9 mm). There is a reduced chance of having a very dry season (less than 88.9 mm).
• The rainfall bar graphs are less useful for areas that are climatologically very dry. In these areas, the median rainfall (and even the higher thresholds) are often close to zero and dividing them into 5 categories is not as meaningful.
• The information above the bar graphs shows the observed median for the selected location for the 1981-2018 period. Underneath this are the percentage chance of experiencing unusually dry (cool) conditions, the chance of exceeding the median, and the chance of experiencing unusually wet (warm) conditions. In the case of rainfall for example, a higher chance of experiencing unusually wet conditions would also typically mean a high chance of exceeding the median rainfall. Similarly, a high chance of experiencing unusually dry conditions would typically mean a low chance of exceeding the median rainfall.
• The stars on the right-hand side of the top panel refer to the accuracy of the forecast. If the model has performed well in forecasting for that location/period/time of year in the past, it will be indicated by three solid stars–meaning high past accuracy. If the model has low past accuracy for that forecast period, it will only have one solid star and two open stars. Refer to the past Accuracy section for more information on accuracy.
• Like the extremes maps, the pop-up boxes use a 5 km by 5 km grid. Pop-up boxes use the closest 5 km grid box.

Graphs of observations and forecasts over time

• Timeline graphs include recent observations and forecasts for weeks or months. Click/tap respective buttons to change the timeline display between weeks and months. Hover/tap on the graph area to view rainfall or temperature details for particular dates.
• The observations used in the graph (solid black line) are the same as that used in the Bureau's rainfall and temperature analysis. Refer to Rainfall map information for details on the rainfall analysis, and Temperature map information for details on the temperature analysis. For the weekly plots, the date listed at the bottom of the graph is the first day of the week; hover over/touch the line on the graph to see the full week information as well as the amount of rainfall over that week. For the example to the right, May 30th refers to the week of 30 May to 5 June. Canberra had approximately 34 mm of rainfall during this week.
• The dashed black line shows the forecast period, specifically indicating the median of the forecast. The first week of the forecast may have been modified to respect the official weather forecast, while the subsequent three weeks of forecasts are fully from the climate forecast system ACCESS-S.
• The forecast for each week is shown as a box and whisker plot. The green line in the centre of the box is the median of the forecast, with the outer edges of the box the 25th and 75th percentiles of the forecast. The outer edges of the lines are 10th and 90th percentiles. Hovering over a forecast point will give you the full range of forecast values. In the example to the right, the full range of values is shown for the forecast for Canberra for the week of 27 June to 3 July. The forecast median for the week is 3.5 mm, with the 25th percentile at 0.3 mm (meaning a 75% chance of at least 0.3 mm), and the 75th percentile at 10.4 mm (meaning a 25% chance of at least 10.4 mm during this week).
• For the monthly timeline graphs, the current month is a combination of observations and forecast data. This means as the end of the month is approached, the spread of likely outcomes for that first month becomes smaller, meaning the box and whiskers are smaller and closer together.
• The stars above each respective forecast at the top of the graph area refer to its accuracy. If the model has performed well in forecasting for that location, period, and time of year in the past, it will be indicated by three solid stars–meaning high past accuracy. If the model has low past accuracy for that forecast period, it will only have one solid star and two open stars. Refer to the past Accuracy section for more information on accuracy.
• The coloured background of the graph helps put the forecasts into historical perspective. It is the observed range of values split into five bins; the white area is the middle 20% of past observed rainfall totals, while the blue/green in the rainfall are the top 40%, and the yellow/brown is the bottom 40%. For temperatures, the reds are the top 40%, and the blues are the bottom 40%. These values are calculated over 1981–2018.
• Both observations and long-range forecast information use the closest 5 km grid box to your location.

Rainfall line graphs showing the chance of exceeding a range of totals

• Line graphs show the percentage chance of exceeding a range of rainfall totals.
• These graphs are also known as Probability of Exceedance (PoE) curves.
• There are two lines shown on the graph. The red line shows the historical chance at that location of receiving a particular rainfall total. Low rainfall totals have a high chance of occurring, while higher rainfall totals will have a much lower chance. For example, the adjacent graph shows an forecast for the week of 25 June to 1 July at Dubbo. Dubbo typically has a 60% chance of receiving around 2 mm of rainfall during this week, a 40% chance of 8 mm, and a 20% chance of 20 mm. The The blue line is the forecast probabilities. When the blue line is below the climatological red line, this indicates a drier forecast, while a blue line above it indicates a wetter forecast. The example shows a drier forecast, with only a 40% chance of receiving 2 mm, and a 20% chance of 8 mm, with just under a 10% chance of 20 mm.
• Tapping/hovering the mouse over the graph brings up the percentage chance for the corresponding rainfall depth in a tooltip. These forecasts allow users to obtain information for specific rainfall amounts that are of interest for their specific application.
• Like the timeline graphs, the coloured background of the graph also helps put the forecasts into historical perspective. It is the observed range of values split into five bins; the white area is the middle 20% of past observed rainfall totals, while the blue/green in the rainfall are the top 40%, and the yellow/brown is the bottom 40%. These values are calculated over 1981–2018.

The model

The Bureau of Meteorology's climate forecast system for weekly to seasonal and longer-range climate forecasts is called the Australian Community Climate Earth-System Simulator – Seasonal (ACCESS–S). It is a state-of-the-art dynamical (physics-based) forecast modelling system, which uses ocean, atmosphere, ice and land observations to initiate forecasts for the season ahead. The ACCESS–S climate model is a collaboration between the Bureau of Meteorology and the UK Meteorological Office (UKMO).

The atmosphere and land model components of ACCESS–S operate at an approximate resolution of 60 km in the Australian region. At this resolution, the model can represent the markedly different climates of the Great Dividing Range and the eastern seaboard in Australia's east.

The ocean model component of ACCESS–S operates at an approximate resolution of 25 km in the Australian region. At this resolution, the model can resolve small-scale currents and eddies.

ACCESS-S forecasts are based on a 99-member ensemble. This is a common climate forecasting technique where the model is run 99 times with slightly different initial conditions to capture a range of likely future scenarios.

Being a dynamical model, ACCESS–S is undergoing continuous research and development. Advances in the science of seasonal prediction, improvements in the observations and how they are fed into the model, as well as increases in supercomputing power are just some of the ways the model's accuracy will increase over time.

ACCESS–S replaced POAMA in August 2018. POAMA, also a dynamical climate model, was used for official Bureau climate forecasts from May 2013 until ACCESS–S was brought into operation. Prior to 2013, the Bureau used a statistical method to generate climate forecasts.

Technical details

• Wedd R et al. (2022) ACCESS-S2: the upgraded Bureau of Meteorology multi-week to seasonal prediction system. Journal of Southern Hemisphere Earth Systems Science 72(3), 218-242.

Read more about ACCESS-S.

Past accuracy

• Model accuracy (also known as model confidence or model skill) is a measure of how well the model has performed at that time of year in the past. The Bureau measures the accuracy of its climate models by comparing how often the real outcomes matched the forecast. This measurement of accuracy is known as percent consistent (or past accuracy), and has been tested for the Bureau's climate forecast model over the period from 1981 to 2018.

Accuracy assessment for chance of exceeding median long-range forecasts

• The accuracy measure of the chance of exceeding median long-range forecasts is based on the percent correct approach, but is weighted by the size of the observed difference from median. This accuracy measure looks at the percentage of time the forecasts correctly predicted being above or below median during the 1981–2018 period for any given location, and is then weighted by the size of the observed difference from the historical median.
• So if the model correctly favoured an area that had an anomalously large rain event, it scores highly, but if it misses that event, points are deducted from the weighting calculation.
  More information on the weighted percent correct accuracy metric.
• Weighted percent consistency is only one of the ways the Bureau assesses its long-range forecasts, and it is presented here as it is one of the simplest and most informative measurements of accuracy of the chance of exceeding the median forecasts. Other aspects of overall model ability (skill) are also routinely assessed.

Accuracy assessment for extreme long-range forecasts

• A different measure of accuracy is used to assess the ability of the model to correctly predict the occurrence of an extreme or historically unusual condition (for rainfall or temperature). The relative operating characteristic (ROC) is a skill score which measures the ability of the model to get the correct outcome for each forecast probability. A correct extreme forecast is known as a "hit", and an extreme forecast that was forecast but did not eventuate is known as a "false alarm". These hits and false alarms are assessed over the 1981–2018 period. A "ROC curve" is then plotted using pairs of hit rates and false alarm rates across the range of forecast probabilities (0–100%). The ROC curve shows the ability of the model to consistenty forecast a higher probability of occurrence of extreme conditions than it would do for a non-occurrence. The area under the curve (AUC) varies from 0 (model has no skill) to 1 (model has perfect skill), or from 0 to 100% in the percentage sense. More information on the ROC curve.

• To make things simpler, the past accuracy of the model is presented on the same scale for all products (although different evaluation methods are used).
• Past accuracy maps for all months and seasons are available from the climate long-range forecasts Map archive and Accuracy maps. On these maps, the higher the percentage value for an area (i.e. the greener/darker the map), the greater the accuracy of forecasts has been in that area for that time of year and the more confidence can be placed in future forecasts. Areas of the map that are not shaded have a lower record of accuracy in that area for that time of the year. In the least accurate areas, the forecasts perform no better than random chance (equivalent to the flip of a coin). As a guide, the Bureau uses the terminology in the table on the right when referring to the accuracy of the long-range forecasts.
• The accuracy for the second month ahead is generally less than the accuracy for the first month ahead. This is to be expected as the second month looks further into the future and is further from when the forecast period starts. Similarly, the accuracy for one and three month forecasts issued closest to when the forecast period starts is typically higher than the accuracy for forecasts issued earlier in the month. This same principle can be applied to weekly and fortnightly forecasts; the first week or fortnight typically has higher accuracy than the second week or fortnight.

Accuracy assessment for 3-day totals maps

• Model accuracy for the 3-day totals long-range forecasts is based on a skill score that is often used for rare weather events (e.g., 50 mm and 75 mm in 3 days is quite uncommon for most regions).
• The skill score, known as the Symmetric Extremal Dependence Index (SEDI), looks at all past forecasts of 3-day rainfall totals over 1981–2018 for a specific forecast period to provide an indication of the accuracy.
• The SEDI skill score incorporates information on how often the Bureau's long-range forecast model has correctly forecast the occurence of these 3-day totals. A forecast that is confirmed to have happened is called a hit, while a false alarm is when the model predicts a rainfall event which was not observed.
• A score greater than zero indicates that the forecasts perform better than random chance (flip of the coin).
• More information on the SEDI accuracy metric is described here.
• Accuracy maps are presented in Bureau Research Report 063 (Appendix, Section 6.3):
• Maps are provided for each 3-day rainfall total (15 mm, 25 mm, 50 mm, and 75 mm), showing the accuracy for specific forecast periods for each month.
• Skill scores lie between −1 and 1, with, for example, a value of 0.2 indicating a 20% improvement over a random forecast, with higher values representing better accuracy.
• White areas on the maps represent locations where no rainfall events were forecast within the 1981–2018 timeframe, and hence, past accuracy could not be determined (e.g., regions where 75 mm in 3-days does not occur).

Climate drivers

• Australia's climate can vary greatly from one year to the next. The links below, present some key influences that drive the Australian climate. They have varying levels of impact in different regions and at different times of year. See: Australian climate influences.

Australian climate influences