Search

The decadal hindcast simulations performed for the Mittelfristige Klimaprognosen (MiKlip) project are evaluated using satellite-retrieved cloud parameters from the CM SAF cLoud, Albedo and RAdiation dataset from AVHRR data (CLARA-A1) provided by the EUMETSAT Satellite Application Facility on Climate Monitoring (CM SAF) and from the International Satellite Cloud Climatology Project (ISCCP). The forecast quality of two sets of hindcasts, Baseline-1-LR and Baseline-0, which use differing initialisations, is assessed. Basic evaluation focuses on multi-year ensemble mean fields and cloud-type histograms utilizing satellite simulator output. Additionally, ensemble evaluation employing analysis of variance (ANOVA), analysis rank histograms (ARH) and a deterministic correlation score is performed. Satellite simulator output is available for a subset of the full hindcast ensembles only. Therefore, the raw model cloud cover is complementary used. The new Baseline-1-LR hindcasts are closer to satellite data with respect to the simulated tropical/subtropical mean cloud cover pattern than the reference hindcasts (Baseline-0) emphasizing improvements of the new MiKlip initialisation procedure. A slightly overestimated occurrence rate of optically thick cloud-types is analysed for different experiments including hindcasts and simulations using realistic sea surface boundaries according to the Atmospheric Model Intercomparison Project (AMIP). By contrast, the evaluation of cirrus and cirrostratus clouds is complicated by observational based uncertainties. Time series of the 3-year mean total cloud cover averaged over the tropical warm pool (TWP) region show some correlation with the CLARA-A1 cloud fractional cover. Moreover, ensemble evaluation of the Baseline-1-LR hindcasts reveals potential predictability of the 2–5 lead year averaged total cloud cover for a large part of this region when regarding the full observational period. However, the hindcasts show only moderate positive correlations with the CLARA-A1 satellite retrieval for the TWP region which are hardly statistical significant. Evidence for predictability of the 2–5 lead year averaged total cloud cover is found for parts of the equatorial to mid-latitudinal North Atlantic.

Decadal climate predictions and their verification are part of ongoing research. This article studies different methods applied to decadal hindcasts of three-dimensional atmospheric variables to evaluate the MiKlip (Mittelfristige Klimaprognosen) prediction system. Variables such as upper air temperature are tight to the core of the prediction system and hence help to reveal its power and deficiencies. The verification uses both, necessary and sufficient probabilistic measures. We analyze annual and multi-year averages of air temperature and geopotential height and the parametrized quantity net water flux at the ocean surface, the so-called freshwater flux, also known as E‑P (evaporation minus precipitation), as an important variable for atmosphere-ocean coupling. The model data stem from various versions of the MiKlip prediction system and constitute different sets of ensemble hindcasts covering 1979–2012. The results reveal that the freshwater flux is far more sensitive to model deficiencies than the basic dynamical variables and the predictability decays much earlier with prediction lead time. Initializing the atmospheric component is more important for the predictability than the difference in resolution between two model versions. The combined initialization of atmosphere and ocean has the effect of increasing the predictability in the inner tropics from 1 to 2 years compared to the ocean only initialization. For prediction year 7–10, the hindcasts are still closer to each other than to the uninitialized historical runs indicating that the prediction system is still influenced by the initial conditions. The skill for prediction year 7–10 is, however, only marginally larger than the skill of the uninitialized ensemble. The three-dimensional skill analysis reveals a clear indication of a mid-tropospheric temperature error developing in the tropical Pacific area.

A framework of ensemble forecast verification tools is discussed which is founded on the concept of information entropy. It can be based on a common yardstick namely that of "correlation". With these measures calibration is deduced from the balance between ensemble sharpness and resolution. With the same units these features can be put into one diagram for continuous time series from Gaussian processes and exceedance probabilities, the latter usually tested with the reliability term from the Brier score. The sharpness and resolution terms allow to use the same vocabulary of over- and underdispersion which is established for frequency histograms. The concept is based on the fact that mutual information (MI) of two Gaussian processes is directly related to Pearson's anomaly correlation. Further MI can be written as the Kullback-Leibler divergence of the conditional probability of observations given the model forecasts and the unconditioned observations. Thus the MI is a measure of resolution. The mean of the UTILITY defined by (Kleeman, 2002) is the corresponding measure of sharpness. For Gaussian processes the mean UTILITY is very close to the ratio of ensemble mean variance to mean ensemble variance (ANOVA) which is the analysis of variance factor when time is taken as treatment. The ensemble spread score (ESS) (Palmer et al., 2006) is shown to be a measure of calibration if model and observed data are scaled with their respective means and standard deviations. For exceedance probabilities the resolution term of the divergence score (Weijs et al., 2010) is already defined as a MI term and it is here complemented with a mean UTILITY formed similarly to the resolution term but with forecasts only. The entropy terms are then rescaled to the "correlation" yardstick. The concept is applied to temperature data from the German project on decadal climate prediction, Mittelfristige Klimaprognose (MiKlip). It is shown that both over – and underdispersion can be found for the 2m temperature forecasts. Increasing ensemble sharpness of surface ocean temperature with lead year in the southern ocean hints at model-data inconsistencies at some locations in the ocean. Finally empirical orthogonal functions (EOF) of northern hemisphere annual mean surface temperature for ERA-40/ERA-Interim and MiKlip retrospective hindcasts are determined. For both data sets the respective first EOF represents the low frequency temperature development. The time coefficients of the EOF are used to compare resolution and sharpness of continuous data and exceedance probabilities in one diagram.

A time series of wind stresses computed from European Centre for Medium Range Weather Forecast (ECMWF) wind data is compared to the climatology of Hellermann & Rosenstein (HR) for the Southern Hemisphere. ECMWF stresses are generally stronger, especially in the westerly belt. However they have an overall lower meridional component than the HR data. The dominance of the half annual cycle relative to the annual wave in the zonal stress at middle to high latitudes, which is documented for independent data sets, is seen in the ECMWF but not in the HR data. ECMWF winds are also compared with measurements from three expeditions to the Weddell Sea by RV Polarstern. Good correlations between Polarstern and ECMWF winds are found but for single dates the differences are above 10 ms −1. The differences are found to be uncorrelated in space and are thus due to observational errors and to the unresolved small scale variance in the ECMWF analysed winds.

Analyses indicate that the Atlantic Ocean SST was considerably colder at the beginning than in the middle of the century. Parallel to this a systematic change of the North Atlantic SLP pattern was observed. To find out whether the analyzed SST and SLP changes are consistent, which would indicate that the SST change was real and not an instrumental artifact, a response experiment with a low resolution (T21) atmospheric GCM was performed. Two perpetual January simulations were conducted which differ solely in Atlantic Ocean (40$ - 60N) SST: the ”cold” simulation utilizes the SSTs for the period 1904-13, the ”warm” simulation uses the SST’s for the period 1951—60. Also a ”control” run, with the model’s standard SST somewhat between the ”cold” and ”warm” SST, was made. For the response analysis a rigorous statistical approach was taken: First the null hypothesis of identical horizontal distributions was subjected to a multivariate significance test. Second, the level of recurrence was estimated. The multivariate statistical approaches are based on hierarchies of test models. We examined three different hierarchies: a scale dependent hierarchy based on spherical harmonics (S), and two physically motivated ones, one based on the barotropic normal modes of the mean 300 hPa flow (B) and one based on the eigenmodes of the advection diffusion operator at 1000 hPa (A). The intercomparison of the ”cold” and ”warm” experiments indicates a signal in the geostrophic stream function that is in the S-hierarchy significantly nonzero and highly recurrent. In the A—hierarchy, the low level temperature field is identi— fied as being significantly and recurrently affected by the altered SST distribution. The SLP signal is reasonably similar to the observed SLP change. Unexpectedly the upper level streamfunction signal does not appear significantly nonzero in the B—hierarchy. If, however the pairs of experiments ”warm vs. control” and ”cold vs. control” are examined in the B—hierarchy, a highly significant and recurrent signal emerges. We conclude that the ”cold vs. warm” response is not a ”small disturbance” which would allow the signal to be described by eigenmodes of the linearized system. An analysis of the three—dimensional structure of the signal leads to the hy— pothesis that two different mechanisms are acting to modify the model’s mean state. At low levels, the local heating and advection are dominant, but at upper levels the extratropical signal is a remote response to modifications of the tropical convection.

An analysis of the signature of the North Atlantic Oscillation (NAO) in the Atlantic-European sea level pressure (SLP) is presented for observed (German Weather Service) and ECMWF T21 model data. The former time series consists of 1881–1984 January to December fields and the latter of 42 monthly fields from 3 permanent January simulations. The NAO is shown to be one of the dominant eigenmodes of SLP for all calendar months. A very similar NAO anomaly pattern is filtered from the T21 model data, whereby the correspondence between model and reality is far better than for the mean states. Typical extreme states of the NAO, i.e. high- and low-index situations are reconstructed from the EOF (empirical orthogonal functions) and the mean fields. The annual cycle of these states and the differences between observed and simulated January extremes of the NAO are discussed. DOI: 10.1034/j.1600-0870.1990.t01-3-00001.x