Prediction Scheme Project
The aim of this project was to produce statistical-based daily weather predictions from on our ongoing Daily Analysis of Rainfall & Temperature (DART) Study. The project was conceived during the strict COVID-19 restrictions in 2020, and was intended for personal interest.
The resulting development is a statistical prediction scheme that produces daily predictions of temperature and rain in XML format for the following calendar year.
The scheme has desired outcomes to evaluate its potential application and guide any further development. The predictions would be verified against the observations to assess the performance of the prediction scheme.
Predictions
Temperature predictions are expressed in whole degrees. For each date it represents the mean maximum and minimum temperatures in the DART Study. To conform to our standard observing procedure, the maximum temperature applies to the 24 hours beginning at 9.00 am, whilst the minimum temperature applies to the 24 hours ending at 9.00 am.
Rainfall predictions are given as a chance and possible amount. To conform to our standard observing procdure, the rainfall applies to the 24 hours ending at 9.00 am.
- Chance is expressed as a percentage. For each date the number of rainfall events in each rainfall range (< 1 mm, < 10 mm, < 25 mm and ≥ 25 mm) in the DART Study are ranked from most to least. The rainfall range with the most events is then compared against the number of rainfall observations. In the event that two or more rainfall ranges tie as the most common, the greater rainfall range is used. The rainfall range 0 mm is only used if 100% of the rainfall observations are in this category.
- Possible amount is expressed as a range: 0 - 1 mm, 1 - 10 mm, 10 - 25 mm and ≥ 25 mm. For each date it represents the most common rainfall range in the DART Study. In the event that two or more rainfall ranges tie as the most common, the greater rainfall range is used. The rainfall range 0 mm is only used if 100% of the rainfall observations are in this category.
Desired Outcomes
The future of the scheme, including any further development, depends on how it performs against these outcomes.
- A high percentage of rain predictions (rain or dry) proven to be correct, conversely a low percentage of rain predictions proven to be incorrect.
- A high percentage of predicted rainfall is in the same range or overestimate the observed rainfall.
- A low percentage of predicted rainfall that underestimate the observed rainfall.
- A high percentage of predicted maximum temperatures from April to September within +/- 2° C of the observed temperatures.
- A high percentage of predicted maximum temperatures from October to March within +/- 5° C of the observed temperatures (preferably +/- 2° C).
- A high percentage of predicted minimum temperatures from April to September within +/- 2° C of the observed temperatures.
- A high percentage of predicted minimum temperatures from October to March within +/- 5° C of the observed temperatures (preferably +/- 2° C).
- Increased reliability over time as the volume of daily weather data expands further.
Verification
Verification is an automated four stage process using the scheme's inbuilt verification algorithms. Stage one deals with temperature, the remaining stages deal with rain.
Stage one rounds the predicted and observed maximum and minimum temperatures to the nearest whole degree, and then calculates the difference by subtracting the predicted value from the observed value.
Stage two classifies each rain prediction as either correct or incorrect.
Stage three selects the correct predictions of rain and then further classifies each case according to whether the predicted rainfall range: (1) matches the observed, (2) overestimates the observed, or (3) underestimates the observed.
Stage four selects the occasions when rain was predicted and did not eventuate. The predictions are based on chance and it is therefore reasonable to expect that not all will eventuate, particularly if the chance is low. The verification estimates the number of cases that could reasonably be expected to be dry, and recommends an adjustment to the assessment in stage two.
Project Status
This 2020 project ran from mid 2020 to December 2025. It achieved the desired outcomes and is now operational. View the catalogue of daily predictions.