Abstract

The fractions skill score (FSS) was one of the measures that formed part of the Intercomparison of Spatial Forecast Verification Methods project. The FSS was used to assess a common dataset that consisted of real and perturbed Weather Research and Forecasting (WRF) model precipitation forecasts, as well as geometric cases. These datasets are all based on the NCEP 240 grid, which translates to approximately 4-km resolution over the contiguous United States.

The geometric cases showed that the FSS can provide a truthful assessment of displacement errors and forecast skill. In addition, the FSS can be used to determine the scale at which an acceptable level of skill is reached and this usage is perhaps more helpful than interpreting the actual FSS value. This spatial-scale approach is becoming more popular for monitoring operational forecast performance.

The study also shows how the FSS responds to forecast bias. A more biased forecast always gives lower FSS values at large scales and usually at smaller scales. It is possible, however, for a more biased forecast to give a higher score at smaller scales, when additional rain overlaps the observed rain. However, given a sufficiently large sample of forecasts, a more biased forecast system will score lower. The use of percentile thresholds can remove the impacts of the bias.

When the proportion of the domain that is “wet” (the wet-area ratio) is small, subtle differences introduced through near-threshold misses can lead to large changes in FSS magnitude in individual cases (primarily because the bias is changed). Reliable statistics for small wet-area ratios require a larger sample of forecasts.

Care needs to be taken in the choice of verification domain. For high-resolution models, the domain should be large enough to encompass the length scale of the typical mesoscale forcing (e.g., upper-level troughs or squall lines). If the domain is too large, the wet-area ratios will always be small. If the domain is too small, fluctuations in the wet-area ratio can be large and larger spatial errors may be missed.

The FSS is a good measure of the spatial accuracy of precipitation forecasts. Different methods are needed to determine other patterns of behavior.

Abstract

The development of NWP models with grid spacing down to ∼1 km should produce more realistic forecasts of convective storms. However, greater realism does not necessarily mean more accurate precipitation forecasts. The rapid growth of errors on small scales in conjunction with preexisting errors on larger scales may limit the usefulness of such models. The purpose of this paper is to examine whether improved model resolution alone is able to produce more skillful precipitation forecasts on useful scales, and how the skill varies with spatial scale. A verification method will be described in which skill is determined from a comparison of rainfall forecasts with radar using fractional coverage over different sized areas. The Met Office Unified Model was run with grid spacings of 12, 4, and 1 km for 10 days in which convection occurred during the summers of 2003 and 2004. All forecasts were run from 12-km initial states for a clean comparison. The results show that the 1-km model was the most skillful over all but the smallest scales (approximately <10–15 km). A measure of acceptable skill was defined; this was attained by the 1-km model at scales around 40–70 km, some 10–20 km less than that of the 12-km model. The biggest improvement occurred for heavier, more localized rain, despite it being more difficult to predict. The 4-km model did not improve much on the 12-km model because of the difficulties of representing convection at that resolution, which was accentuated by the spinup from 12-km fields.

Abstract

A technique to produce high-water alerts from coinciding high astronomical tide and high mean sea level anomaly is demonstrated for the Pacific Islands region. Low-lying coastal margins are vulnerable to episodic inundation that often coincides with times of higher-than-normal high tides. Prior knowledge of the dates of the highest tides can assist with efforts to minimize the impacts of increased exposure to inundation. It is shown that the climate-driven mean sea level anomaly is an important component of total sea level elevation in the Pacific Islands region, which should be accounted for in medium-term (1–7 months) sea level forecasts. An empirical technique is applied to develop a mean sea level–adjusted high-water alert calendar that accounts for both sea level components and provides a practical tool to assist with coastal inundation hazard planning and management.

Abstract

A high-resolution data assimilation system has been implemented and tested within a 4-km grid length version of the Met Office Unified Model (UM). A variational analysis scheme is used to correct larger scales using conventional observation types. The system uses two nudging procedures to assimilate high-resolution information: radar-derived surface precipitation rates are assimilated via latent heat nudging (LHN), while cloud nudging (CN) is used to assimilate moisture fields derived from satellite, radar, and surface observations. The data assimilation scheme was tested on five convection-dominated case studies from the Convective Storm Initiation Project (CSIP). Model skill was assessed statistically using radar-derived surface-precipitation hourly accumulations via a scale-dependent verification scheme. Data assimilation is shown to have a dramatic impact on skill during both the assimilation and subsequent forecast periods on nowcasting time scales. The resulting forecasts are also shown to be much more skillful than those produced using either a 12-km grid length version of the UM with data assimilation in place, or a 4-km grid-length UM version run using a 12-km state as initial conditions. The individual contribution to overall skill attributable to each data-assimilation component is investigated. Up until T + 3 h, LHN has the greatest impact on skill. For later times, VAR, LHN, and CN contribute equally to the skill in predicting the spatial distribution of the heaviest rainfall; while VAR alone accounts for the skill in depicting distributions corresponding to lower accumulation thresholds. The 4-km forecasts tend to overpredict both the intensity and areal coverage of storms. While it is likely that the intensity bias is partially attributable to model error, both VAR and LHN clearly contribute to the overestimation of the areal extent. Future developments that may mitigate this problem are suggested.

ABSTRACT

Flooding is one of the costliest hazards in the United Kingdom. A large part of the annual flood damage is caused by surface water flooding that is a direct result of intense rainfall. Traditional catchment-based approaches to flood prediction are not applicable for surface water floods. However, given sufficiently accurate forecasts of rainfall intensity, with sufficient lead time, actions can be taken to reduce their impact. These actions require reliable information about severity and areas at risk that is clear and easily interpreted. The accuracy requirements, in particular, are very challenging, as they relate to prediction of intensities that occur only infrequently and that typically affect only small areas. In this paper, forecasts of intense rainfall from a new convection-permitting ensemble prediction system are evaluated using radar observations of intense rain and surface water flooding reports. An urban flooding case that occurred in Edinburgh in 2011 is first investigated and then a broader look is taken at performance through a 3-month period during the London Olympic and Paralympic Games in 2012. Conclusions are drawn about the value of the ensemble and the particular means of presenting the forecasts, and areas requiring further work are highlighted.

Abstract

With movement toward kilometer-scale ensembles, new techniques are needed for their characterization. A new methodology is presented for detailed spatial ensemble characterization using the fractions skill score (FSS). To evaluate spatial forecast differences, the average and standard deviation are taken of the FSS calculated over all ensemble member–member pairs at different scales and lead times. These methods were found to give important information about the ensemble behavior allowing the identification of useful spatial scales, spinup times for the model, and upscale growth of errors and forecast differences. The ensemble spread was found to be highly dependent on the spatial scales considered and the threshold applied to the field. High thresholds picked out localized and intense values that gave large temporal variability in ensemble spread: local processes and undersampling dominate for these thresholds. For lower thresholds the ensemble spread increases with time as differences between the ensemble members upscale. Two convective cases were investigated based on the Met Office United Model run at 2.2-km resolution. Different ensemble types were considered: ensembles produced using the Met Office Global and Regional Ensemble Prediction System (MOGREPS) and an ensemble produced using different model physics configurations. Comparison of the MOGREPS and multiphysics ensembles demonstrated the utility of spatial ensemble evaluation techniques for assessing the impact of different perturbation strategies and the need for assessing spread at different, believable, spatial scales.

Abstract

The realistic representation of rainfall on the local scale in climate models remains a key challenge. Realism encompasses the full spatial and temporal structure of rainfall, and is a key indicator of model skill in representing the underlying processes. In particular, if rainfall is more realistic in a climate model, there is greater confidence in its projections of future change.

In this study, the realism of rainfall in a very high-resolution (1.5 km) regional climate model (RCM) is compared to a coarser-resolution 12-km RCM. This is the first time a convection-permitting model has been run for an extended period (1989–2008) over a region of the United Kingdom, allowing the characteristics of rainfall to be evaluated in a climatological sense. In particular, the duration and spatial extent of hourly rainfall across the southern United Kingdom is examined, with a key focus on heavy rainfall.

Rainfall in the 1.5-km RCM is found to be much more realistic than in the 12-km RCM. In the 12-km RCM, heavy rain events are not heavy enough, and tend to be too persistent and widespread. While the 1.5-km model does have a tendency for heavy rain to be too intense, it still gives a much better representation of its duration and spatial extent. Long-standing problems in climate models, such as the tendency for too much persistent light rain and errors in the diurnal cycle, are also considerably reduced in the 1.5-km RCM. Biases in the 12-km RCM appear to be linked to deficiencies in the representation of convection.

Abstract

The statistical properties and skill in predictions of objectively identified and tracked cyclonic features (frontal waves and cyclones) are examined in the 15-day version of the Met Office Global and Regional Ensemble Prediction System (MOGREPS-15). The number density of cyclonic features is found to decline with increasing lead time, with analysis fields containing weak features that are not sustained past the first day of the forecast. This loss of cyclonic features is associated with a decline in area-averaged enstrophy with increasing lead time. Both feature number density and area-averaged enstrophy saturate by around 7 days into the forecast. It is found that the feature number density and area-averaged enstrophy of forecasts produced using model versions that include stochastic energy backscatter saturate at higher values than forecasts produced without stochastic physics. The ability of MOGREPS-15 to predict the locations of cyclonic features of different strengths is evaluated at different spatial scales by examining the Brier skill (relative to the analysis climatology) of strike probability forecasts: the probability that a cyclonic feature center is located within a specified radius. The radius at which skill is maximized increases with lead time from 650 km at 12 h to 950 km at 7 days. The skill is greatest for the most intense features. Forecast skill remains above zero at these scales out to 14 days for the most intense cyclonic features, but only out to 8 days when all features are included irrespective of intensity.

Abstract

In this second part of a two-part study of recursive filter techniques applied to the synthesis of covariances in a variational analysis, methods by which non-Gaussian shapes and spatial inhomogeneities and anisotropies for the covariances may be introduced in a well-controlled way are examined. These methods permit an analysis scheme to possess covariance structures with adaptive variations of amplitude, scale, profile shape, and degrees of local anisotropy, all as functions of geographical location and altitude.

First, it is shown how a wider and more useful variety of covariance shapes than just the Gaussian may be obtained by the positive superposition of Gaussian components of different scales, or by further combinations of these operators with the application of Laplacian operators in order for the products to possess negative sidelobes in their radial profiles.

Then it is shown how the techniques of recursive filters may be generalized to admit the construction of covariances whose characteristic scales relative to the grid become adaptive to geographical location, while preserving the necessary properties of self-adjointness and positivity. Special attention is paid to the problems of amplitude control for these spatially inhomogeneous filters and an estimate for the kernel amplitude is proposed based upon an asymptotic analysis of the problem.

Finally, a further generalization of the filters that enables fully anisotropic and geographically adaptive covariances to be constructed in a computationally efficient way is discussed.

Abstract

Many factors, both mesoscale and larger scale, often come together in order for a particular convective initiation to take place. The authors describe a modeling study of a case from the Convective Storms Initiation Project (CSIP) in which a single thunderstorm formed behind a front in the southern United Kingdom. The key features of the case were a tongue of low-level high θ_w air associated with a forward-sloping split front (overrunning lower θ_w air above), a convergence line, and a “lid” of high static stability air, which the shower was initially constrained below but later broke through. In this paper, the authors analyze the initiation of the storm, which can be traced back to a region of high ground (Dartmoor) at around 0700 UTC, in more detail using model sensitivity studies with the Met Office Unified Model (MetUM). It is established that the convergence line was initially caused by roughness effects but had a significant thermal component later. Dartmoor had a key role in the development of the thunderstorm. A period of asymmetric flow over the high ground, with stronger low-level descent in the lee, led to a hole in a layer of low-level clouds downstream. The surface solar heating through this hole, in combination with the tongue of low-level high θ_w air associated with the front, caused the shower to initiate with sufficient lifting to enable it later to break through the lid.