Accuracy of Arctic stratospheric temperature analyses and the implications for the prediction of polar stratospheric clouds

Lower stratospheric temperature analyses from European Centre for Medium‐Range Weather Forecasts (ECMWF) are compared with radiosonde temperatures in the Arctic in the winters 1994/95 and 1995/96. In February and March 1996, the biases between mean layer radiosonde and ECMWF temperatures are −0.1±1.2, 0.0±1.1, +0.1±1.4, and −0.7±1.8 K at 100, 70, 50, and 30 mb, respectively. Individual measurements between the radiosonde standard levels, however, exhibit substantially larger standard deviations, and at temperatures low enough for the formation of polar stratospheric clouds (PSCs) the biases are −0.6±1.5, −1.1±1.8, −1.4±1.9, and −1.6±2.3 K, respectively. This means that ECMWF temperatures predict too small an extent of PSCs. The cause is the inability of models with limited resolution to catch small‐scale temperature variations, and UK Meteorological Office (UKMO) temperature analyses exhibit similar biases. Before January 30, 1996, when ECMWF changed to a 3‐dimensional variational analysis, substantial biases to mean layer radiosonde temperatures exist.