Weather affects a wide variety of human activity. It is also the longest recorded and measured natural phenomenon, measured in lots of places and observed by many people. There are many aspects of weather - temperature, rainfall, humidity and cloud cover to name but a few - and they can be measured and recorded in different ways. Hence datasets on weather are among the most complicated.
There are many practical situations where weather can be used to better understand activities and for planning. Some projects where Data Analysis Australia has found it important to use statistical weather analysis include:
- Scheduling maintenance on electricity generation capacity to avoid peaks in demand (typically driven by extreme temperature).
- Monitoring water conservation efforts - reductions in water use might be due to conservation efforts or unusually 'kind' weather (lots of rain, no hot spells).
- In road safety, there is a recognised strong correlation between road and driving conditions (affected by rain, fog, etc) and accidents rates. Here the weather is often a "nuisance factor" that needs to be removed to better understand factors that can be used to improve road safety.
Through these and other projects Data Analysis Australia has developed advanced expertise in using weather data. If there is anything that this experience tells us, it is that weather data requires careful treatment if the most benefit is to be obtained. Some of Data Analysis Australia's "Weather Wisdom" is outlined below:
- A model including weather should include detail down to the individual day, since variations in weather are observed on a daily basis.
- The effects of weather are complex and non-linear. For example the effect of a 5°C change in temperature around the 40°C mark is much different to a 5°C change around 20°C.
- Weather can have effects that last several days. There is a relationship between the first day of rain and traffic accidents, for example. On the first day of rain, accident rates increase due to the mix of oil and water on the roads, but decrease on the day after as the oil has been washed away.
- The long historical records typical of weather data shouldn't be taken at face value. Quite often the data must be adjusted for small changes in the location of temperature readings or even local factors such as the type of lawn around the equipment. These are significant issues that can introduce error and bias into any analyses or modelling exercise.
Data Analysis Australia maintains a comprehensive database of day-to-day weather information using Bureau of Meteorology data. For further information on using weather data, please contact Data Analysis Australia at daa(at)daa.com.au or phone 08 9468 2533.