Towards optimal parameters for the prediction of near surface temperature and dewpoint

Towards optimal parameters for the prediction of near surface temperature and dewpoint
Technical memorandum
Date Published
Secondary Title
ECMWF Technical Memoranda
Anton Beljaars

Screen level temperature and dewpoint forecasts from ECMWF’s Integrated Forecasting System (IFS) show systematic errors with large scale geographical patterns, seasonal variation and pronounced diurnal cycles. The errors are hard to address because of the multitude of processes involved. To be able to optimize parameters it is important to have a fast and efficient testing environment. This report explores the potential benefit of so-called ”relaxation” integrations, where the upper air model fields are gently relaxed towards the analyses. It has the advantage that a full annual cycle, following a realistic synoptic evolution, can be reproduced rather efficiently. Three aspects are discussed: (i) evaluation of the relaxation integrations, (ii) verification of daily maximum/minimum temperature and corresponding dewpoint at screen level, and (iii) sensitivity experiments in an attempt to find more optimum parameter settings in the atmosphere/land coupling.

It is concluded that relaxation experiments are useful, although it is necessary to account for differences with respect to operations. Evaluation indicates that about half of the root mean square errors in temperature and dewpoint are systematic errors (in the sense that these error are present in the local monthly means). The systematic errors have complex large scale geographical, seasonal, and diurnal patterns. Some of the errors can be related to orography, snow and bare soil, and some to land-atmosphere coupling. Parameter sensitivity experiments show that improvement is possible on some aspects, but it is very hard to find combinations of parameters that do not lead to deterioration of other aspects. The combination of a daytime dry and cold bias in summer is of particular concern, because soil moisture data assimilation relies heavily on the assumption that the model is unbiased.

DOI 10.21957/yt64x7rth