Exploring the Relationship Between Temperature Forecast Errors and Earth System Variables

Accurate subseasonal weather forecasts, from two weeks up to a season, can help reduce costs and impacts related to weather and corresponding extremes. The quality of weather forecasts has improved considerably in recent decades as models represent more details of physical processes, and they benefit from assimilating comprehensive Earth observation data as well as increasing computing power. However, with ever–growing model complexity, it becomes increasingly difficult to pinpoint weaknesses in the forecast models’ process representations which is key to improving forecast accuracy. In this study, we use a comprehensive set of observation–based ecological, hydrological and meteorological variables to study their potential for explaining temperature forecast errors at the weekly time scale. For this purpose, we compute Spearman correlations between each considered variable and the forecast error obtained from the ECMWF subseasonal–to–seasonal (S2S) reforecasts at lead times of 1–6 weeks. This is done across the globe for the time period 2001–2017. The results show that temperature forecast errors globally are most strongly related with climate–related variables such as surface solar radiation and precipitation, which highlights the model’s difficulties in accurately capturing the evolution of the climate–related variables during the forecasting period. At the same time, we find particular regions in which other variables are more strongly related to forecast errors. For instance, in central Europe, eastern North America and southeastern Asia, vegetation greenness and soil moisture are relevant, while in western South America and central North America, circulation–related variables such as surface pressure relate more strongly with forecast errors. Overall, the identified relationships between forecast errors and independent Earth observations reveal promising variables on which future forecasting system development could focus by specifically considering related process representations and data assimilation.