This collection provides hourly near-surface (1.5 m) air temperature grids for Australia with a spatial resolution of 1 km. These data are spatially interpolated using quart-variate thin plate splines (full spline dependence on easting, northing, elevation, and time- and space-varying coastal distance index). This collection includes: (i) a 30-year climatology for every hour on every 5th day of the year (1991-2020), (ii) a time-series of hourly air temperature beginning on 01/Jan/2015, (iii) an animation of the 30-year climatology from (i), and (iv) supporting metadata and point-based climatologies for weather station locations to enable comparative statistical analyses.
Further details are provided in the README file.
Credit
We at TERN acknowledge the Traditional Owners and Custodians throughout Australia, New Zealand and all nations. We honour their profound connections to land, water, biodiversity and culture and pay our respects to their Elders past, present and emerging. This work was funded by the Terrestrial Ecosystem Research Network (TERN), an Australian Government NCRIS enabled project, and is supported by the use of TERN infrastructure.
Purpose
These data were developed to support numerous scientific applications, including the Surface-Air Temperature Difference Anomaly (SATDA), an early-warning method for detecting vegetation drought stress (
Cai et al., 2025).
Lineage
Full details are provided in Stewart et al. (2024), including the Supplementary Material.
The original station data were supplied by the METAR data stream from BoM from January 1990 to March 2023 for 621 stations across Australia. A direct spatial interpolation approach (i.e., models are re-fitted for each time step using all available observations and model covariates) was applied to generate air temperature grids across all Australia using the ANUSPLIN v4.4. software (Hutchinson and Xu, 2013). Two key data products are included in this collection:
1. Hourly air temperature climatologies (every 5th DOY, 1991-2020)
Hourly air temperature climatologies were interpolated with quart-variate thin plate splines as a function of easting, northing, elevation, and a time- and space-varying coastal distance index. Elevation (m) was exaggerated by a factor of 100 to represent the differences in horizontal and vertical synoptic scales typical for spline-based climate interpolation (Hutchinson et al. 2009). The coastal distance index is calculated as a limiting transformation of the generalised distance to coast (Hutchinson et al. 2021), and corresponds to sea breeze dynamics, particularly during spring and summer in the afternoon and evening hours. Observations of the same hour within +/- 5 DOYs (i.e., 5 DOYs before and 5 DOYs after the target DOY for each hour) are included to increase the number of values available for calculating stable climatologies. A total of 498 stations were used for interpolating climatologies using the v3 methodology representing 1991 to 2020.
2. Hourly air temperature (every hour, 01/Jan/2015 onwards)
Hourly air temperature was directly interpolated with quart-variate thin plate splines as a function of easting, northing, elevation, and a time- and space-varying coastal distance index (i.e., following the same approach as the climatologies, but for every hour). A total of 551 stations were used for interpolating hourly air temperature using the v3 product methodology between 01/Jan/2015 and 23/Jan/2024.
Version history:
v2 - All interpolations now use a direct spatial interpolation method, 80–95 % of the data points as knots, and include a time-varying coastal distance index that allows the models to represent the effects of sea breeze dynamics on air temperature. These changes reduced cross-validated root mean squared error (RMSE) by 14% for air temperature climatologies (v2 RMSE = 0.75 °C) and 7% for hourly interpolations (v2 RMSE = 1.56 °C) when pooled across all available times and stations. Note v2.0 is described in Stewart et al. (2024).
v3 - All interpolations now use a time- and space-varying coastal distance index that optimises the covariate surface by day-of-year and local solar time. This allows for the coastal distance index to respond to varying solar geometry at specific times (e.g., solar exposure on east coast vs west coast) and therefore better represent thermal gradients at the coast in different locations. Air temperature climatologies were also reprocessed to represent 1991-2020. The v3 product results in cross-validated root mean squared error (RMSE) of 0.72 °C and 1.52 °C for climatologies (1991-2020) and hourly time-series (01/Jan/2015 to 23/Jan/2024). Comparisons with v2 are not provided due to the use of different assessment periods.
References:
