A KGE-based weighted mean of stations’ ensemble to estimate the air temperature at Jungfraujoch since 1900

Air temperature is a key variable in the meteo-climatological fields because impacts the atmospheric stability and clouds formation, drives wind patterns and defines the kind of precipitation. However, there is a scarcity of long-term data, especially at high elevations (more than 2000 m). This study proposes a statistical-based methodology to reconstruct a long-term daily temperature record (maximum, mean, and minimum) for high-altitude sites. We have tested it at Jungfraujoch (3571 m a.s.l.), Switzerland, with a backward simulation extending to 1900. The methodology involves daily data from surrounding meteorological stations (thirty), within the MeteoSwiss database, located at elevations ranging 485-2691m a.s.l., providing uninterrupted observations spanning at least the period from 1971 to 2023. The methodology includes the following steps: 1) long-term temporal consistency was evaluated by removing observations with data gaps exceeding 30 days; 2) the mean monthly trend was removed using a non-linear trend estimation function; 3) for each meteorological station, during the calibration period (1988–2005), the daily temperature at Jungfraujoch was estimated as the sum of the temperature at the selected station plus a deterministic and stochastic component; 4) pairwise model performance was evaluated within two validation periods (1971–1985 and 2005–2023) by calculating biases, RMSE, correlation coefficients, rank-based metrics, and the Kling-Gupta Efficiency (KGE); 5) stations with a KGE greater than 0.9 were selected to calculate ensemble simulations, which were obtained as the weighted mean of these stations, extending back to the year 1900 ; 6) A validation was conducted by comparing the reconstructed time series with the closest grid point from two datasets: HISTALP and that provided by Imfeld et al. (2023).

The results suggest: i) comparable performance with existing datasets (HISTALP, Imfeld et al. 2023), despite using a highly parsimonious model that does not rely on additional variables such as relative humidity, cloud cover, wind velocity, or weather patterns; ii) the selection of stations with temporally consistent long-term observations is critical; iii) model performance, efficiency, and errors are primarily influenced by elevation, rather than latitude, longitude, exposure, or distance; iv) the Kling-Gupta Efficiency (KGE) is the most appropriate metric for selecting stations to be used in the ensemble; v) Temporally consistent time series generated by this methodology can provide a benchmark for evaluating observations anomalies and for deeper analysis of Elevation-Dependent Warming issue.