Abstract

Statistical postprocessing is commonly applied to reduce location and dispersion errors of probabilistic forecasts provided by numerical weather prediction (NWP) models. If postprocessed forecast scenarios are required, the combination of ensemble model output statistics (EMOS) for univariate postprocessing with ensemble copula coupling (ECC) or the Schaake shuffle (ScS) to retain the dependence structure of the raw ensemble is a state-of-the-art approach. However, modern machine learning methods may lead to both a better univariate skill and more realistic forecast scenarios. In this study, we postprocess multimodel ensemble forecasts of cloud cover over Switzerland provided by COSMO-E and ECMWF-IFS using (i) EMOS + ECC, (ii) EMOS + ScS, (iii) dense neural networks (dense NN) + ECC, (iv) dense NN + ScS, and (v) conditional generative adversarial networks (cGAN). The different methods are verified using EUMETSAT satellite data. Dense NN shows the best univariate skill, but cGAN performed only slightly worse. Furthermore, cGAN generates realistic forecast scenario maps, while not relying on a dependence template like ECC or ScS, which is particularly favorable in the case of complex topography.

Abstract

The authors have analyzed global station data and created a gridded dataset of monthly precipitation for the period of 1900–88. Statistical analyses suggest that discontinuities associated with instrumental errors are large for many high-latitude station records, although they are unlikely to be significant for the majority of the stations. The first leading EOF in global precipitation fields is an ENSO-related pattern, concentrating mostly in the low latitudes. The second leading EOF depicts a linear increasing trend (∼2.4 mm decade⁻¹) in global precipitation fields during the period of 1900–88. Consistent with the zonal precipitation trends identified in previous analyses, the EOF trend is seen as a long-term increase mostly in North America, mid- to high-latitude Eurasia, Argentina, and Australia. The spatial patterns of the trend EOF and the rate of increase are generally consistent with those of the precipitation changes in increasing CO₂ GCM experiments.

The North Atlantic oscillation (NAO) accounts for ∼10% of December–February precipitation variance over North Atlantic surrounding regions. The mode suggests that during high-NAO-index winters, precipitation is above normal in northern (>50°N) Europe, the eastern United States, northern Africa, and the Mediterranean, while below-normal precipitation occurs in southern Europe, eastern Canada, and western Greenland.

Wet and dry months of one standard deviation occur at probabilities close to those of a normal distribution in midlatitudes. In the subtropics, the mean interval between two extreme events is longer. The monthly wet and dry events seldom (probability < 5%) last longer than 2 months. ENSO is the single largest cause of global extreme precipitation events. Consistent with the upward trend in global precipitation, globally, the averaged mean interval between two dry months increased by ∼28% from 1900–44 to 1945–88. The percentage of wet areas over the United States has more than doubled (from ∼12% to >24%) since the 1970s, while the percentage of dry areas has decreased by a similar amount since the 1940s. Severe droughts and floods comparable to the 1988 drought and 1993 flood in the Midwest have occurred 2–9 times in each of several other regions of the world during this century.

Abstract

In the Project for Intercomparison of Land-Surface Parameterization Schemes phase 2a experiment, meteorological data for the year 1987 from Cabauw, the Netherlands, were used as inputs to 23 land-surface flux schemes designed for use in climate and weather models. Schemes were evaluated by comparing their outputs with long-term measurements of surface sensible heat fluxes into the atmosphere and the ground, and of upward longwave radiation and total net radiative fluxes, and also comparing them with latent heat fluxes derived from a surface energy balance. Tuning of schemes by use of the observed flux data was not permitted. On an annual basis, the predicted surface radiative temperature exhibits a range of 2 K across schemes, consistent with the range of about 10 W m⁻² in predicted surface net radiation. Most modeled values of monthly net radiation differ from the observations by less than the estimated maximum monthly observational error (±10 W m⁻²). However, modeled radiative surface temperature appears to have a systematic positive bias in most schemes; this might be explained by an error in assumed emissivity and by models’ neglect of canopy thermal heterogeneity. Annual means of sensible and latent heat fluxes, into which net radiation is partitioned, have ranges across schemes of30 W m⁻² and 25 W m⁻², respectively. Annual totals of evapotranspiration and runoff, into which the precipitation is partitioned, both have ranges of 315 mm. These ranges in annual heat and water fluxes were approximately halved upon exclusion of the three schemes that have no stomatal resistance under non-water-stressed conditions. Many schemes tend to underestimate latent heat flux and overestimate sensible heat flux in summer, with a reverse tendency in winter. For six schemes, root-mean-square deviations of predictions from monthly observations are less than the estimated upper bounds on observation errors (5 W m⁻² for sensible heat flux and 10 W m⁻² for latent heat flux). Actual runoff at the site is believed to be dominated by vertical drainage to groundwater, but several schemes produced significant amounts of runoff as overland flow or interflow. There is a range across schemes of 184 mm (40% of total pore volume) in the simulated annual mean root-zone soil moisture. Unfortunately, no measurements of soil moisture were available for model evaluation. A theoretical analysis suggested that differences in boundary conditions used in various schemes are not sufficient to explain the large variance in soil moisture. However, many of the extreme values of soil moisture could be explained in terms of the particulars of experimental setup or excessive evapotranspiration.

Abstract

In the PILPS Phase 2a experiment, 23 land-surface schemes were compared in an off-line control experiment using observed meteorological data from Cabauw, the Netherlands. Two simple sensitivity experiments were also undertaken in which the observed surface air temperature was artificially increased or decreased by 2 K while all other factors remained as observed. On the annual timescale, all schemes show similar responses to these perturbations in latent, sensible heat flux, and other key variables. For the 2-K increase in temperature, surface temperatures and latent heat fluxes all increase while net radiation, sensible heat fluxes, and soil moistures all decrease. The results are reversed for a 2-K temperature decrease. The changes in sensible heat fluxes and, especially, the changes in the latent heat fluxes are not linearly related to the change of temperature. Theoretically, the nonlinear relationship between air temperature and the latent heat flux is evident and due to the convex relationship between air temperature and saturation vapor pressure. A simple test shows that, the effect of the change of air temperature on the atmospheric stratification aside, this nonlinear relationship is shown in the form that the increase of the latent heat flux for a 2-K temperature increase is larger than its decrease for a 2-K temperature decrease. However, the results from the Cabauw sensitivity experiments show that the increase of the latent heat flux in the +2-K experiment is smaller than the decrease of the latent heat flux in the −2-K experiment (we refer to this as the asymmetry). The analysis in this paper shows that this inconsistency between the theoretical relationship and the Cabauw sensitivity experiments results (or the asymmetry) is due to (i) the involvement of the β _g formulation, which is a function of a series stress factors that limited the evaporation and whose values change in the ±2-K experiments, leading to strong modifications of the latent heat flux; (ii) the change of the drag coefficient induced by the changes in stratification due to the imposed air temperature changes (±2 K) in parameterizations of latent heat flux common in current land-surface schemes. Among all stress factors involved in the β _g formulation, the soil moisture stress in the +2-K experiment induced by the increased evaporation is the main factor that contributes to the asymmetry.

Abstract

Twenty-one land surface schemes (LSSs) performed simulations forced by 18 yr of observed meteorological data from a grassland catchment at Valdai, Russia, as part of the Project for the Intercomparison of Land-Surface Parameterization Schemes (PILPS) Phase 2(d). In this paper the authors examine the simulation of snow. In comparison with observations, the models are able to capture the broad features of the snow regime on both an intra- and interannual basis. However, weaknesses in the simulations exist, and early season ablation events are a significant source of model scatter. Over the 18-yr simulation, systematic differences between the models’ snow simulations are evident and reveal specific aspects of snow model parameterization and design as being responsible. Vapor exchange at the snow surface varies widely among the models, ranging from a large net loss to a small net source for the snow season. Snow albedo, fractional snow cover, and their interplay have a large effect on energy available for ablation, with differences among models most evident at low snow depths. The incorporation of the snowpack within an LSS structure affects the method by which snow accesses, as well as utilizes, available energy for ablation. The sensitivity of some models to longwave radiation, the dominant winter radiative flux, is partly due to a stability-induced feedback and the differing abilities of models to exchange turbulent energy with the atmosphere. Results presented in this paper suggest where weaknesses in macroscale snow modeling lie and where both theoretical and observational work should be focused to address these weaknesses.

Abstract

The Project for Intercomparison of Land-Surface Parameterization Schemes phase 2(d) experiment at Valdai, Russia, offers a unique opportunity to evaluate land surface schemes, especially snow and frozen soil parameterizations. Here, the ability of the 21 schemes that participated in the experiment to correctly simulate the thermal and hydrological properties of the soil on several different timescales was examined. Using observed vertical profiles of soil temperature and soil moisture, the impact of frozen soil schemes in the land surface models on the soil temperature and soil moisture simulations was evaluated.

It was found that when soil-water freezing is explicitly included in a model, it improves the simulation of soil temperature and its variability at seasonal and interannual scales. Although change of thermal conductivity of the soil also affects soil temperature simulation, this effect is rather weak. The impact of frozen soil on soil moisture is inconclusive in this experiment due to the particular climate at Valdai, where the top 1 m of soil is very close to saturation during winter and the range for soil moisture changes at the time of snowmelt is very limited. The results also imply that inclusion of explicit snow processes in the models would contribute to substantially improved simulations. More sophisticated snow models based on snow physics tend to produce better snow simulations, especially of snow ablation. Hysteresis of snow-cover fraction as a function of snow depth is observed at the catchment but not in any of the models.