Interpolation issues and data codes
SILO datasets are constructed from observational records provided by the Bureau of Meteorology. SILO interpolates the raw data, which may contain missing values, to derive datasets which are both spatially and temporally complete.
For a brief comparison of the products available from SILO and the Bureau, please see our product comparison .
For a detailed analysis of the interpolated surfaces provided by SILO and the Bureau, please see Beesley, C. A., Frost, A. J. and Zajaczkowski, J. (2009) A comparison of the BAWAP and SILO spatially interpolated daily rainfall datasets .
For a full list of publications relevant to SILO data, please refer to the publications and references page.
The accuracy of SILO’s interpolated datasets depends on both:
- the climate variable, and
- the number of observed values used to construct the gridded raster.
Variables such as temperature can be reliably interpolated over significant distances because air temperature is relatively stable. In contrast, rainfall is difficult to interpolate because it is typically “patchy”. In other words the rainfall received at one location can be quite different to the rainfall received at nearby location. Regardless of the climate variable, the accuracy of a gridded dataset will, in general, be strongly dependent on the number of observations used to construct it. As the number of stations reporting data for a given variable has varied considerably throughout history, the accuracy of SILO’s gridded rasters also varies depending on the period. SILO recommends users consider both the number and location of stations used to construct gridded rasters, as it can significantly affects the accuracy of interpolated estimates. (“number and location” should be a link to the new page)
Where possible (depending on the file format), the data are supplied with codes indicating how each datum was obtained.
||Official observation as supplied by the Bureau of Meteorology.
||Deaccumulated rainfall (original observation was recorded over a period exceeding the standard 24 hour observation period).
||Interpolated from daily observations for that date.
||Synthetic Class A pan evaporation, calculated from temperatures, radiation and vapour pressure.
||Interpolated from daily observations using an anomaly interpolation method.
||Interpolated from the long term averages of daily observations for that day of year.
1. Redistribution (code 15)
Much of the data in Australia have been collected by volunteer observers at Post Offices, Police stations, etc. Some of these workplaces only operate on weekdays. Consequently the rainfall measured on Monday may be for the period since 9am on the preceding Friday. In such cases, SILO redistributes the accumulated rainfall back to the days when it probably fell according to the amount and days that rain fell at nearby stations.
2. CLIMARC interpolation method (code 35)
An anomaly interpolation technique was used to construct the gridded surfaces for maximum and minimum temperatures, radiation and vapour pressure, for all years prior to 1957. The surfaces were derived, in part, using observational data made available under the CLIMARC project. Prior to the availability of the CLIMARC datasets, there were insufficient data to construct gridded surfaces so long term mean data were typically used.
The user should note that CLIMARC interpolations:
Are used throughout the period 1889-1956;
Are based on relatively few observations. For example, there are data for about 60 CLIMARC stations compared to several hundred stations reporting temperature in the post-1956 period;
Show less variation than the post-1956 data; and
Are derived from relatively old data which in some cases contain uncorrected instrument biases.
Consequently the interpolated CLIMARC data may not be suitable for some studies e.g. climate change detection, extreme events, number of frosts etc.
We expect the interpolated CLIMARC data to be significantly better than long term averages. Furthermore, the CLIMARC data should preserve the daily relationships between different elements better than long term means. It should however be noted that in some areas and years, the CLIMARC data are no different to the previously supplied long term averages. For example, in Western Australia there are very few observations before 1907, and Ceduna didn't commence reporting until about 1940. Please consult the CLIMARC Report for further information.
Brief history of CLIMARC
CLIMARC - Computerising the Australian climate archives” (QPI43) Nick Clarkson, Queensland Centre for Climate Applications (QCCA)/DPI&F, was a joint project between the first national Climate Variability in Agriculture Program (CVAP) , the Queensland Department of Primary Industries and Fisheries (DPI&F), the Queensland Department of Natural Resources, Mines and Energy (NRM&E), and the Australian Government Bureau of Meteorology. With the aim to increase the amount of data available electronically for variables other than rainfall, for the period before 1957. When the Bureau of Meteorology commenced storing observations electronically, it decided to computerise all rainfall data but not the climate data before 1957 for cost reasons. The CLIMARC project digitised the pre-1957 climate data for 50 locations. Prior to CLIMARC there were only 5 locations in Australia where the entire observational record had been digitised. A number of other locations had already been partly digitised either by the Bureau of Meteorology or other organisations. The pre-1957 data for the 50 CLIMARC locations, together with the small amount of data which were already available, made it possible to construct gridded surfaces for the pre-1957 period. The methodology is described in the CLIMARC Report .
Removal of suspect values
SILO uses a two pass interpolation system to identify and remove suspect values from the dataset used to construct a gridded surface. Suspect values are excluded from the interpolation so they do not adversely affect the gridded surface. Removal of suspect values from the interpolation also impacts SILO’s point datasets:
Point datasets at gridded locations are constructed using the gridded surfaces. Withholding a suspect value from the interpolation will result in localised changes to the fitted surface, so point datasets in the vicinity of the withheld datum will be affected.
Point datasets at station locations may not be consistent with interpolated data at the nearest grid point, because the station point dataset may contain observed data which were rejected by the interpolation system.
Grid point latitude, longitude and elevation
Interpolated estimates are computed at the latitude and longitude at the centre of the given pixel, and if appropriate, the mean elevation for the entire pixel.