Predicting outliers in ensemble forecasts

An ensemble forecast is a collection of runs of a numerical dynamical model, initialized with perturbed initial conditions. In modern weather prediction for example, ensembles are used to retrieve probabilistic information about future weather conditions. In this contribution, we are concerned with ensemble forecasts of a scalar quantity (say, the temperature at a specific location). We consider the event that the verification is smaller than the smallest, or larger than the largest ensemble member. We call these events outliers. If a K-member ensemble accurately reflected the variability of the verification, outliers should occur with a base rate of 2/(K + 1). In operational forecast ensembles though, this frequency is often found to be higher. We study the predictability of outliers and find that, exploiting information available from the ensemble, forecast probabilities for outlier events can be calculated which are more skilful than the unconditional base rate. We prove this analytically for statistically consistent forecast ensembles. Further, the analytical results are compared to the predictability of outliers in an operational forecast ensemble by means of model output statistics. We find the analytical and empirical results to agree both qualitatively and quantitatively.

From ensemble forecasts to predictive distribution functions

The translation of an ensemble of model runs into a probability distribution is a common task in model-based prediction. Common methods for such ensemble interpretations proceed as if verification and ensemble were draws from the same underlying distribution, an assumption not viable for most, if any, real world ensembles. An alternative is to consider an ensemble as merely a source of information rather than the possible scenarios of reality. This approach, which looks for maps between ensembles and probabilistic distributions, is investigated and extended. Common methods are revisited, and an improvement to standard kernel dressing, called ‘affine kernel dressing’ (AKD), is introduced. AKD assumes an affine mapping between ensemble and verification, typically not acting on individual ensemble members but on the entire ensemble as a whole, the parameters of this mapping are determined in parallel with the other dressing parameters, including a weight assigned to the unconditioned (climatological) distribution. These amendments to standard kernel dressing, albeit simple, can improve performance significantly and are shown to be appropriate for both overdispersive and underdispersive ensembles, unlike standard kernel dressing which exacerbates over dispersion. Studies are presented using operational numerical weather predictions for two locations and data from the Lorenz63 system, demonstrating both effectiveness given operational constraints and statistical significance given a large sample.

Examining reliability of seasonal to decadal sea surface temperature forecasts: the role of ensemble dispersion

Useful probabilistic climate forecasts on decadal timescales should be reliable (i.e. forecast probabilities match the observed relative frequencies) but this is seldom examined. This paper assesses a necessary condition for reliability, that the ratio of ensemble spread to forecast error being close to one, for seasonal to decadal sea surface temperature retrospective forecasts from the Met Office Decadal Prediction System (DePreSys). Factors which may affect reliability are diagnosed by comparing this spread-error ratio for an initial condition ensemble and two perturbed physics ensembles for initialized and uninitialized predictions. At lead times less than 2 years, the initialized ensembles tend to be under-dispersed, and hence produce overconfident and hence unreliable forecasts. For longer lead times, all three ensembles are predominantly over-dispersed. Such over-dispersion is primarily related to excessive inter-annual variability in the climate model. These findings highlight the need to carefully evaluate simulated variability in seasonal and decadal prediction systems.Useful probabilistic climate forecasts on decadal timescales should be reliable (i.e. forecast probabilities match the observed relative frequencies) but this is seldom examined. This paper assesses a necessary condition for reliability, that the ratio of ensemble spread to forecast error being close to one, for seasonal to decadal sea surface temperature retrospective forecasts from the Met Office Decadal Prediction System (DePreSys). Factors which may affect reliability are diagnosed by comparing this spread-error ratio for an initial condition ensemble and two perturbed physics ensembles for initialized and uninitialized predictions. At lead times less than 2 years, the initialized ensembles tend to be under-dispersed, and hence produce overconfident and hence unreliable forecasts. For longer lead times, all three ensembles are predominantly over-dispersed. Such over-dispersion is primarily related to excessive inter-annual variability in the climate model. These findings highlight the need to carefully evaluate simulated variability in seasonal and decadal prediction systems.

Examining reliability of seasonal to decadal sea surface temperature forecasts: the role of ensemble dispersion

Useful probabilistic climate forecasts on decadal timescales should be reliable (i.e. forecast probabilities match the observed relative frequencies) but this is seldom examined. This paper assesses a necessary condition for reliability, that the ratio of ensemble spread to forecast error being close to one, for seasonal to decadal sea surface temperature retrospective forecasts from the Met Office Decadal Prediction System (DePreSys). Factors which may affect reliability are diagnosed by comparing this spread-error ratio for an initial condition ensemble and two perturbed physics ensembles for initialized and uninitialized predictions. At lead times less than 2 years, the initialized ensembles tend to be under-dispersed, and hence produce overconfident and hence unreliable forecasts. For longer lead times, all three ensembles are predominantly over-dispersed. Such over-dispersion is primarily related to excessive inter-annual variability in the climate model. These findings highlight the need to carefully evaluate simulated variability in seasonal and decadal prediction systems.Useful probabilistic climate forecasts on decadal timescales should be reliable (i.e. forecast probabilities match the observed relative frequencies) but this is seldom examined. This paper assesses a necessary condition for reliability, that the ratio of ensemble spread to forecast error being close to one, for seasonal to decadal sea surface temperature retrospective forecasts from the Met Office Decadal Prediction System (DePreSys). Factors which may affect reliability are diagnosed by comparing this spread-error ratio for an initial condition ensemble and two perturbed physics ensembles for initialized and uninitialized predictions. At lead times less than 2 years, the initialized ensembles tend to be under-dispersed, and hence produce overconfident and hence unreliable forecasts. For longer lead times, all three ensembles are predominantly over-dispersed. Such over-dispersion is primarily related to excessive inter-annual variability in the climate model. These findings highlight the need to carefully evaluate simulated variability in seasonal and decadal prediction systems.

Verification of European subseasonal wind speed forecasts

Analysis of the forecasts and hindcasts from the ECMWF 32-day forecast model reveals that there is statistically significant skill in predicting weekly mean wind speeds over areas of Europe at lead times of at least 14–20 days. Previous research on wind speed predictability has focused on the short- to medium-range time scales, typically finding that forecasts lose all skill by the later part of the medium-range forecast. To the authors’ knowledge, this research is the first to look beyond the medium-range time scale by taking weekly mean wind speeds, instead of averages at hourly or daily resolution, for the ECMWF monthly forecasting system. It is shown that the operational forecasts have high levels of correlation (~0.6) between the forecasts and observations over the winters of 2008–12 for some areas of Europe. Hindcasts covering 20 winters show a more modest level of correlation but are still skillful. Additional analysis examines the probabilistic skill for the United Kingdom with the application of wind power forecasting in mind. It is also shown that there is forecast “value” for end users (operating in a simple cost/loss ratio decision-making framework). End users that are sensitive to winter wind speed variability over the United Kingdom, Germany, and some other areas of Europe should therefore consider forecasts beyond the medium-range time scale as it is clear there is useful information contained within the forecast.

Improving flood forecasts for better flood preparedness in the UK (and beyond)

This winter (2013/14) coastal storms and an unprecedented amount of rainfall led to significant and widespread flooding across the southern UK. Despite much criticism and blame surrounding the flood events, the Flood Forecasting Centre, a recent development in national-level flood forecasting capabilities for the government and emergency response communities, has received considerable praise. Here we consider how scientific developments and organisational change have led to improvements in the forecasting and flood preparedness seen in this winter's flooding. Although such improvements are admirable, there are many technical and communication challenges that remain for probabilistic flood forecasts to achieve their full potential.

Global forecasting of thermal health hazards: the skill of probabilistic predictions of the Universal Thermal Climate Index (UTCI)

Although over a hundred thermal indices can be used for assessing thermal health hazards, many ignore the human heat budget, physiology and clothing. The Universal Thermal Climate Index (UTCI) addresses these shortcomings by using an advanced thermo-physiological model. This paper assesses the potential of using the UTCI for forecasting thermal health hazards. Traditionally, such hazard forecasting has had two further limitations: it has been narrowly focused on a particular region or nation and has relied on the use of single ‘deterministic’ forecasts. Here, the UTCI is computed on a global scale,which is essential for international health-hazard warnings and disaster preparedness, and it is provided as a probabilistic forecast. It is shown that probabilistic UTCI forecasts are superior in skill to deterministic forecasts and that despite global variations, the UTCI forecast is skilful for lead times up to 10 days. The paper also demonstrates the utility of probabilistic UTCI forecasts on the example of the 2010 heat wave in Russia.

An evaluation of decadal probability forecasts from state-of-the-art climate models

While state-of-the-art models of Earth's climate system have improved tremendously over the last 20 years, nontrivial structural flaws still hinder their ability to forecast the decadal dynamics of the Earth system realistically. Contrasting the skill of these models not only with each other but also with empirical models can reveal the space and time scales on which simulation models exploit their physical basis effectively and quantify their ability to add information to operational forecasts. The skill of decadal probabilistic hindcasts for annual global-mean and regional-mean temperatures from the EU Ensemble-Based Predictions of Climate Changes and Their Impacts (ENSEMBLES) project is contrasted with several empirical models. Both the ENSEMBLES models and a “dynamic climatology” empirical model show probabilistic skill above that of a static climatology for global-mean temperature. The dynamic climatology model, however, often outperforms the ENSEMBLES models. The fact that empirical models display skill similar to that of today's state-of-the-art simulation models suggests that empirical forecasts can improve decadal forecasts for climate services, just as in weather, medium-range, and seasonal forecasting. It is suggested that the direct comparison of simulation models with empirical models becomes a regular component of large model forecast evaluations. Doing so would clarify the extent to which state-of-the-art simulation models provide information beyond that available from simpler empirical models and clarify current limitations in using simulation forecasting for decision support. Ultimately, the skill of simulation models based on physical principles is expected to surpass that of empirical models in a changing climate; their direct comparison provides information on progress toward that goal, which is not available in model–model intercomparisons.

Towards improving the framework for probabilistic forecast evaluation

The evaluation of forecast performance plays a central role both in the interpretation and use of forecast systems and in their development. Different evaluation measures (scores) are available, often quantifying different characteristics of forecast performance. The properties of several proper scores for probabilistic forecast evaluation are contrasted and then used to interpret decadal probability hindcasts of global mean temperature. The Continuous Ranked Probability Score (CRPS), Proper Linear (PL) score, and IJ Good’s logarithmic score (also referred to as Ignorance) are compared; although information from all three may be useful, the logarithmic score has an immediate interpretation and is not insensitive to forecast busts. Neither CRPS nor PL is local; this is shown to produce counter intuitive evaluations by CRPS. Benchmark forecasts from empirical models like Dynamic Climatology place the scores in context. Comparing scores for forecast systems based on physical models (in this case HadCM3, from the CMIP5 decadal archive) against such benchmarks is more informative than internal comparison systems based on similar physical simulation models with each other. It is shown that a forecast system based on HadCM3 out performs Dynamic Climatology in decadal global mean temperature hindcasts; Dynamic Climatology previously outperformed a forecast system based upon HadGEM2 and reasons for these results are suggested. Forecasts of aggregate data (5-year means of global mean temperature) are, of course, narrower than forecasts of annual averages due to the suppression of variance; while the average “distance” between the forecasts and a target may be expected to decrease, little if any discernible improvement in probabilistic skill is achieved.

Improving probabilistic flood forecasting through a data assimilation scheme based on genetic programming

Opportunities offered by high performance computing provide a significant degree of promise in the enhancement of the performance of real-time flood forecasting systems. In this paper, a real-time framework for probabilistic flood forecasting through data assimilation is presented. The distributed rainfall-runoff real-time interactive basin simulator (RIBS) model is selected to simulate the hydrological process in the basin. Although the RIBS model is deterministic, it is run in a probabilistic way through the results of calibration developed in a previous work performed by the authors that identifies the probability distribution functions that best characterise the most relevant model parameters. Adaptive techniques improve the result of flood forecasts because the model can be adapted to observations in real time as new information is available. The new adaptive forecast model based on genetic programming as a data assimilation technique is compared with the previously developed flood forecast model based on the calibration results. Both models are probabilistic as they generate an ensemble of hydrographs, taking the different uncertainties inherent in any forecast process into account. The Manzanares River basin was selected as a case study, with the process being computationally intensive as it requires simulation of many replicas of the ensemble in real time.

A Bayesian state space modelling approach to probabilistic quantitative precipitation forecasting

The generation of very short range forecasts of precipitation in the 0-6 h time window is traditionally referred to as nowcasting. Most existing nowcasting systems essentially extrapolate radar observations in some manner, however, very few systems account for the uncertainties involved. Thus deterministic forecast are produced, which have a limited use when decisions must be made, since they have no measure of confidence or spread of the forecast. This paper develops a Bayesian state space modelling framework for quantitative precipitation nowcasting which is probabilistic from conception. The model treats the observations (radar) as noisy realisations of the underlying true precipitation process, recognising that this process can never be completely known, and thus must be represented probabilistically. In the model presented here the dynamics of the precipitation are dominated by advection, so this is a probabilistic extrapolation forecast. The model is designed in such a way as to minimise the computational burden, while maintaining a full, joint representation of the probability density function of the precipitation process. The update and evolution equations avoid the need to sample, thus only one model needs be run as opposed to the more traditional ensemble route. It is shown that the model works well on both simulated and real data, but that further work is required before the model can be used operationally. © 2004 Elsevier B.V. All rights reserved.

Wind power forecasts using Gaussian processes and numerical weather prediction

Since wind at the earth's surface has an intrinsically complex and stochastic nature, accurate wind power forecasts are necessary for the safe and economic use of wind energy. In this paper, we investigated a combination of numeric and probabilistic models: a Gaussian process (GP) combined with a numerical weather prediction (NWP) model was applied to wind-power forecasting up to one day ahead. First, the wind-speed data from NWP was corrected by a GP, then, as there is always a defined limit on power generated in a wind turbine due to the turbine controlling strategy, wind power forecasts were realized by modeling the relationship between the corrected wind speed and power output using a censored GP. To validate the proposed approach, three real-world datasets were used for model training and testing. The empirical results were compared with several classical wind forecast models, and based on the mean absolute error (MAE), the proposed model provides around 9% to 14% improvement in forecasting accuracy compared to an artificial neural network (ANN) model, and nearly 17% improvement on a third dataset which is from a newly-built wind farm for which there is a limited amount of training data. © 2013 IEEE.

High-level concept annotation using ontology and probabilistic inference

Image annotation is a significant step towards semantic based image retrieval. Ontology is a popular approach for semantic representation and has been intensively studied for multimedia analysis. However, relations among concepts are seldom used to extract higher-level semantics. Moreover, the ontology inference is often crisp. This paper aims to enable sophisticated semantic querying of images, and thus contributes to 1) an ontology framework to contain both visual and contextual knowledge, and 2) a probabilistic inference approach to reason the high-level concepts based on different sources of information. The experiment on a natural scene database from LabelMe database shows encouraging results.