Gridded Daily Air Temperature at 14:00 CET for Austria at 1 km Resolution, 1961-2025

Overview This gridded meteorological dataset provides daily estimates of air temperature at 14:00 CET (T14) across Austria at a spatial resolution of 1 km for the period 1961-2025. Period: 01.01.1961 - 31.12.2025Temporal Resolution: dailyExtent: AustriaSpatial Resolution: 1 km × 1 kmProjection: ETRS89 / Austria Lambert (EPSG: 3416)Interpretation grid: grid coordinates represent the center point of the grid cellFormat: NetCDF Application Air temperature at 14:00 CET (T14) is a highly relevant variable because it was historically a standard time for manual observations at meteorological stations, resulting in consistent long-term time series. Consequently, T14 has been widely incorporated into climatic and agroclimatic indices. In Austria, it is used in the soil evaluation system (“Bodenschätzung”) and related applications, where derived indicators such as the heat sum ("Wärmesumme") and the K-index characterize vegetation-relevant thermal conditions and annual humidity or aridity (Harlfinger and Knees, 1999). T14 is also used in other indices such as the Fire Weather Index (FWI), which relies on standardized afternoon conditions to assess fire danger (Van Wagner, 1987). Many climate model datasets provide only daily minimum and maximum temperatures, lacking sub-daily information. This dataset helps bridge that gap by providing consistent estimates of T14, enabling the application of established indices and improving comparability between observations and model data. Methods T14 was estimated using a random forest model trained on station observations and applied to gridded predictor data. Step 1: Selection of GeoSphere Station Data All available GeoSphere Austria stations with daily records were considered (481 stations in total). The following variables were used: Short Name Description Unit tl_ii air temperature at 2 m at observation time II (14:00 CET) °C tlmin minimum air temperature at 2 m (between 19:00 CET of the previous day and 19:00 CET of the current day) °C tlmax maximum air temperature at 2 m (between 19:00 CET of the previous day and 19:00 CET of the current day) °C so_h sunshine duration (sum of values between 00:00 and 24:00 CET of the current day) h rr precipitation (sum of values between 07:00 CET of the current day and 07:00 CET of the following day) mm Step 2: Creation of the Input Data Set For each station, only data from the climatological reference period 1996-2025 were used for the input data set. Only stations with at least 50% data availability during this period were retained, resulting in 248 stations. A total of 11 predictor variables were included in the final model: Short Name Description Unit lon longitude ° lat latitude ° alt altitude m doy day of year - month month - tmin minimum temperature °C tmax maximum temperature °C trange temperature range (tmax - tmin) °C tmean mean temperature ((tmin + tmax) / 2) °C sun sunshine duration h precip precipitation mm The actual target variable is not directly T14 but a scale factor describing the relative position of T14 within the daily temperature range: $\text{scale_factor} = \frac{\text{T}_{\text{max}} - \text{T}_{14}}{\text{T}_{\text{range}}}$ The scale factor ranges between 0 and 1. A value of 0 indicates that T14 equals the daily maximum temperature, while a value of 1 indicates that T14 equals the daily minimum temperature. For numerical stability during model fitting, very small scale factor values (< 0.004) were randomly transformed to a uniform distribution between 0.004 and 0.01. Step 3: Fitting of the Random Forest Model A random forest model was trained using the ranger package in R (Wright, 2015). Based on permutation variable importance, the most influential predictors (in decreasing order) are: sun, trange, tmax, tmean, tmin, alt, doy, precip, month, lon, lat Step 4: Estimation on the SPARTACUS Grid The trained model was applied to the SPARTACUS gridded dataset from GeoSphere Austria at a spatial resolution of 1 km × 1 km across Austria for the period 1961-2025. The following gridded SPARTACUS variables were used: Short Name Description Unit TN minimum temperature, daily minimum measured between 19:00 CET of the previous day and 19:00 CET of the current day °C TX maximum temperature, daily maximum measured between 19:00 CET of the previous day and 19:00 CET of the current day °C SA sunshine duration, daily sum measured between 00:00 and 24:00 CET of the current day s RR precipitation, daily sum measured between 07:00 CET of the current day and 07:00 CET of the following day kg m⁻² All remaining predictors were derived from these variables. Units were harmonized to match those of the training dataset. Altitude was obtained from the elevation of the SPARTACUS grid. First, the scale factor was predicted using the random forest model. The final T14 values were then reconstructed as: $\text{T}_{14} = \text{T}_{\text{max}} - \text{T}_{\text{range}} \cdot \text{scale_factor}$ Step 5: Model Validation The resulting gridded T14 dataset was validated against all available station observations from 1961-2025. For direct comparison, grid values were elevation-corrected using a constant lapse rate of −6.5 K per 100 m, accounting for the elevation difference between the grid cell and the station. The resulting model performance metrics and goodness-of-fit are: Metric Value Mean Bias (MB) -0.198 °C Mean Absolute Error (MAE) 1.064 °C Root Mean Square Error (RMSE) 1.502 °C Explained Variance (R²) 0.976 Data Sources Station data: GeoSphere Austria, Messstationen Tagesdaten v2 (https://doi.org/10.60669/gs6w-jd70) Raster data: GeoSphere Austria, SPARTACUS v2.1 Tagesdaten (https://doi.org/10.60669/m6w8-s545) References Harlfinger, O. and G. Knees, 1999: Klimahandbuch der österreichischen Bodenschätzung: Klimatographie , Teil 1. Klimareferat der österreichischen Bodenschätzung. Van Wagner, C. E., 1987: Development and structure of the Canadian Forest Fire Weather Index System. Canadian Forestry Service, Headquarters, Ottawa. Forestry Technical Report 35. 35 p. Wright, M. N., 2015: ranger: A Fast Implementation of Random Forests. The R Foundation. doi:10.32614/cran.package.ranger.