Spatial analysis of sunshine duration in complex terrain by non-contemporaneous combination of station and satellite data

Climate monitoring and environmental modelling are in need of spatial datasets of sunshine duration or surface radiation. Their development is complicated by the coarse resolution of station measurements and the limited temporal extent and consistency of satellite retrievals. We present a method for spatial analysis that combines station and satellite data. Instead of merging contemporaneous measurements from both sources, our approach relies on statistical patterns distilled from satellite data over a limited time period. This permits application outside the satellite period and reduces inconsistencies across satellite generations. The non-contemporaneous merging builds on principal component analysis and kriging with external drift. Its strengths are in regions of complex orography. We develop the non-contemporaneous combination to derive km-scale analyses of relative sunshine duration (monthly and daily) for Switzerland (1971–2012). The monthly analysis has a mean absolute error of 3–5% (per cent relative sunshine duration), and it explains 60–80% of the spatial variance in individual months. Errors of the daily analysis are 7–10% with an explained spatial variance of about 60% but strongly varying from day to day. The integration of satellite data systematically improves the analysis compared to interpolation from station data alone. The improvement is largest in autumn and winter, particularly because of better representation of low-level stratus. The non-contemporaneous merging realizes more than half the accuracy gain from satellite data obtained with the more informative contemporaneous merging. Hence, our alternative involves only a moderate compromise for the return of making pre-satellite Campbell–Stokes measurements exploitable for long-term grid datasets.