Reconstruction of long-term consistent air temperature grids for Austria back to 1781

Access to spatially comprehensive information of climate variables spanning multiple decades is crucial for various applications, including ecosystem modelling, climate monitoring, and the evaluation of climate models. However, existing observational temperature datasets for climate monitoring in Austria often exhibit limitations in either temporal extension or spatial comprehensiveness. The HISTALP dataset provides homogenized monthly observation series of air temperature for the greater Alpine region, with records dating back to the 19^th or even 18^th century, but with limited spatial coverage. Conversely, the Austrian spatial climate observation dataset SPARTACUS offers daily-resolved high-resolution spatial grids of air temperature but is restricted to the period after 1961.

This study aimed to address these limitations by constructing a temporally consistent grid dataset of monthly air temperature for Austria, covering the period from 1781 to 2020. Combining the strengths of both the HISTALP and SPARTACUS datasets, we applied a statistical reconstruction technique called „Reduced Space Optimal Interpolation“ (RSOI), involving a Principal Component Analysis (PCA) and Optimal Interpolation (OI). This methodology allowed us to merge long-term, continuous, and homogeneous mean air temperature series from HISTALP with the high-resolution grids derived from SPARTACUS. A further advantage of this method is the possibility to reconstruct the temperature evolution during the early instrumental period even in regions where direct observations were lacking at that time.

The resulting grid dataset, named SOCRATES (Spatial Reconstruction of Climate in Austria Combining SPARTACUS and HISTALP Datasets), provides monthly grids of air temperature anomalies back to 1781 with respect to the reference period 1961-1990. These anomaly grids allow the derivation of absolute temperature grids as well as seasonal and annual aggregates. Beside details on the method, we will present some results of the evaluation. The comparison of the reconstruction with observations by applying a leave-one-out cross validation showed a bias close to zero across all reconstruction periods and seasons. The mean absolute error (MAE) decreased over the considered reconstruction periods, i.e. from 0.35 K for 1781-2020 to 0.22 K for 1951-2020, regarding full years. Furthermore, the MAE showed a seasonal dependence with the lowest errors in summer and highest errors in winter. The applicability of the reconstructions further depends on the regions within Austria. In low-lying parts of northern and eastern Austria, the results demonstrated high reconstruction skill, even for the earliest reconstruction period, while for southern Austria and high elevations it is recommended to consider reconstruction periods starting in 1851 or later. Overall, the results emphasized the capability of SOCRATES in achieving high temporal consistency, which is essential for its use in the long-term spatial climate monitoring in Austria.