Probabilistic Weather Forecasts Research Articles

Leveraging Deterministic Weather Forecasts for In Situ Probabilistic Temperature Predictions via Deep Learning

Abstract We propose a neural network approach to produce probabilistic weather forecasts from a deterministic numerical weather prediction. Our approach is applied to operational surface temperature outputs from the Global Deterministic Prediction System up to 10-day lead times, targeting METAR observations in Canada and the United States. We show how postprocessing performance is improved by training a single model for multiple lead times. Multiple strategies to condition the network for the lead time are studied, including a supplementary predictor and an embedding. The proposed model is evaluated for accuracy, spread, distribution calibration, and its behavior under extremes. The neural network approach decreases the continuous ranked probability score (CRPS) by 15% and has improved distribution calibration compared to a naive probabilistic model based on past forecast errors. Our approach increases the value of a deterministic forecast by adding information about the uncertainty, without incurring the cost of simulating multiple trajectories. It applies to any gridded forecast including the recent machine learning–based weather prediction models. It requires no information regarding forecast spread and can be trained to generate probabilistic predictions from any deterministic forecast. Significance Statement Weather is difficult to predict a long time in advance because we cannot measure the state of the atmosphere precisely enough. Consequently, it is common practice to run forecasts several times and look at the differences to evaluate how uncertain the prediction is. This process of running ensemble forecasts is expensive and consequently not always feasible. We propose a middle ground where we add uncertainty information to forecasts that were run only once, using artificial intelligence. Our method increases the value of these forecasts by adding information about the uncertainty without incurring the cost of multiple full simulations.

Everything Hits at Once: How Remote Rainfall Matters for the Prediction of the 2021 North American Heat Wave

AbstractIn June 2021, Western North America experienced an intense heat wave with unprecedented temperatures and far‐reaching socio‐economic consequences. Anomalous rainfall in the West Pacific triggers a cascade of weather events across the Pacific, which build up a high‐amplitude ridge over Canada and ultimately lead to the heat wave. We show that the response of the jet stream to diabatically enhanced ascending motion in extratropical cyclones represents a predictability barrier with regard to the heat wave magnitude. Therefore, probabilistic weather forecasts are only able to predict the extremity of the heat wave once the complex cascade of weather events is captured. Our results highlight the key role of the sequence of individual weather events in limiting the predictability of this extreme event. We therefore conclude that it is not sufficient to consider such rare events in isolation but it is essential to account for the whole cascade over different spatiotemporal scales.

The sensitivity of probabilistic convective‐scale forecasts of an extratropical cyclone to atmosphere–ocean–wave coupling

AbstractThe benefits of dynamical atmosphere–ocean–wave coupling in probabilistic weather forecasts generated using convective‐scale ensemble prediction systems are to date unknown. We investigate the respective impacts of atmosphere–ocean–wave coupling, and initial condition (IC), lateral boundary condition (LBC), and stochastic physics perturbations within a convective‐scale ensemble coupled system for an extratropical cyclone case study. Towards this aim, we developed the first 18‐member, 2.2 km grid spacing ensemble regional coupled system (Ensemble‐RCS) with domain covering the British Isles and surrounding seas. Ensemble‐RCS coupled and uncoupled simulations of cyclone (February 2020) were performed. Adding stochastic perturbations to the model physics parametrizations enhances the ensemble spread of the uncoupled atmosphere‐only ensemble driven by IC and LBC perturbations, while slightly reducing (by up to 0.5 ms) the median of the ensemble 95th percentile 10‐m wind speeds from its value of about 24 ms at peak time. A substantial proportion of this impact is attributable to Charnock parameter perturbations alone. By coupling the atmosphere‐only ensemble, with stochastic physics, to the ocean, the ensemble median and spread is mainly unaffected. However, additional coupling to waves reduces the median wind speed by 1 ms, which leads to reductions of up to 70% in strong wind strike probability, and halving of the spatial coverage of high values (>50%) of this probability. Finally, we demonstrate the usefulness of two metrics originally developed for precipitation verification – the neighbourhood‐based Fraction Skill Score (FSS) and the object‐based Structure, Amplitude, Location (SAL) – for examining the spread in convective‐scale ensemble forecasts. It is concluded that coupling has a consistent impact across the ensemble members. Remarkably, the impact of coupling to waves is found to be comparable in size to that of adding IC, LBC and stochastic physics perturbations to the uncoupled atmosphere‐only ensemble simulation, implying that the dynamical coupling to ocean and sea‐state are important aspects of model uncertainty.

Combining distribution‐based neural networks to predict weather forecast probabilities

AbstractThe success of deep learning techniques over the last decades has opened up a new avenue of research for weather forecasting. Here, we take the novel approach of using a neural network to predict full probability density functions at each point in space and time rather than a single output value, thus producing a probabilistic weather forecast. This enables the calculation of both uncertainty and skill metrics for the neural network predictions, and overcomes the common difficulty of inferring uncertainty from these predictions. This approach is data‐driven and the neural network is trained on the WeatherBench dataset (processed ERA5 data) to forecast geopotential and temperature 3 and 5 days ahead. Data exploration leads to the identification of the most important input variables. In order to increase computational efficiency, several neural networks are trained on small subsets of these variables. The outputs are then combined through a stacked neural network, the first time such a technique has been applied to weather data. Our approach is found to be more accurate than some coarse numerical weather prediction models and as accurate as more complex alternative neural networks, with the added benefit of providing key probabilistic information necessary for making informed weather forecasts.

Area-covering postprocessing of ensemble precipitation forecasts using topographical and seasonal conditions

Probabilistic weather forecasts from ensemble systems require statistical postprocessing to yield calibrated and sharp predictive distributions. This paper presents an area-covering postprocessing method for ensemble precipitation predictions. We rely on the ensemble model output statistics (EMOS) approach, which generates probabilistic forecasts with a parametric distribution whose parameters depend on (statistics of) the ensemble prediction. A case study with daily precipitation predictions across Switzerland highlights that postprocessing at observation locations indeed improves high-resolution ensemble forecasts, with 4.5% CRPS reduction on average in the case of a lead time of 1 day. Our main aim is to achieve such an improvement without binding the model to stations, by leveraging topographical covariates. Specifically, regression coefficients are estimated by weighting the training data in relation to the topographical similarity between their station of origin and the prediction location. In our case study, this approach is found to reproduce the performance of the local model without using local historical data for calibration. We further identify that one key difficulty is that postprocessing often degrades the performance of the ensemble forecast during summer and early autumn. To mitigate, we additionally estimate on the training set whether postprocessing at a specific location is expected to improve the prediction. If not, the direct model output is used. This extension reduces the CRPS of the topographical model by up to another 1.7 % on average at the price of a slight degradation in calibration. In this case, the highest improvement is achieved for a lead time of 4 days.

Embracing Uncertainty: Using Probabilistic Weather Forecasts to Make Ensemble Hydraulic Predictions at Remote Low-Water Crossings

AbstractLow-water crossings are structures designed to be overtopped during high river flows. These structures are usually constructed in remote locations, making timely emergency response difficult in case of flooding. In this work, five historical flooding events were hindcasted at a remote low-water crossing in central Texas. An ensemble of model-simulated precipitation forcing cascades uncertainty through hydrologic and hydraulic models. Each precipitation ensemble member corresponds to an independent model run, resulting in an ensemble 24-h streamflow forecast initialized at 0000 UTC. In addition to the hydrologic conditions, the forecast is expanded to predict river hydraulics, through flow velocity and depth. Analysis of the five hindcast events indicates that cascading probabilistic precipitation through hydrologic and hydraulic models adds robustness to river forecasts compared to deterministic methods. The approach provides a means to communicate the uncertainty of predictions through the ensemble spread. Analysis of deterministic hazard thresholds suggest that a hydraulic stability threshold, calculated as the multiplication of flow velocity and depth, is a useful alternative approach to NWS high-water categories for communicating hydrologic/hydraulic risk, as well as associated model uncertainty in the simplest manner possible.

Artificial Learning Dispatch Planning with Probabilistic Forecasts: Using Uncertainties as an Asset

Weather forecast uncertainty is a key element for energy market volatility. By intelligently considering uncertainties on the schedule development, renewable energy systems with storage could improve dispatching accuracy, and therefore, effectively participate in electricity wholesale markets. Deterministic forecasts have been traditionally used to support dispatch planning, representing reduced or no uncertainty information about the future weather. Aiming at better representing the uncertainties involved, probabilistic forecasts have been developed to increase forecasting accuracy. For the dispatch planning, this can highly influence the development of a more precise schedule. This work extends a dispatch planning method to the use of probabilistic weather forecasts. The underlying method used a schedule optimizer coupled to a post-processing machine learning algorithm. This machine learning algorithm was adapted to include probabilistic forecasts, considering their additional information on uncertainties. This post-processing applied a calibration of the planned schedule considering the knowledge about uncertainties obtained from similar past situations. Simulations performed with a concentrated solar power plant model following the proposed strategy demonstrated promising financial improvement and relevant potential in dealing with uncertainties. Results especially show that information included in probabilistic forecasts can increase financial revenues up to 15% (in comparison to a persistence solar driven approach) if processed in a suitable way.

The Met Office Weather Game: investigating how different methods for presenting probabilistic weather forecasts influence decision-making

Abstract. To inform the way probabilistic forecasts would be displayed on their website, the UK Met Office ran an online game as a mass participation experiment to highlight the best methods of communicating uncertainty in rainfall and temperature forecasts, and to widen public engagement in uncertainty in weather forecasting. The game used a hypothetical “ice-cream seller” scenario and a randomized structure to test decision-making ability using different methods of representing uncertainty and to enable participants to experience being “lucky” or “unlucky” when the most likely forecast scenario did not occur. Data were collected on participant age, gender, educational attainment, and previous experience of environmental modelling. The large number of participants (n>8000) that played the game has led to the collation of a unique large dataset with which to compare the impact on the decision-making ability of different weather forecast presentation formats. This analysis demonstrates that within the game the provision of information regarding forecast uncertainty greatly improved decision-making ability and did not cause confusion in situations where providing the uncertainty added no further information.

Constrained Quantile Regression Splines for Ensemble Postprocessing

Abstract Statistical postprocessing of ensemble forecasts is widely applied to make reliable probabilistic weather forecasts. Motivated by the fact that nature imposes few restrictions on the shape of forecast distributions, a flexible quantile regression method based on constrained spline functions (CQRS) is proposed and tested on ECMWF Ensemble Prediction System (ENS) wind speed forecasting data at 125 stations in Norway. First, it is demonstrated that constraining quantile functions to be monotone and bounded is preferable. Second, combining an ensemble quantile with the ensemble mean proved to be a good covariate for the respective quantile. Third, CQRS only needs to be applied to about 10 equidistant quantiles, while those between can be obtained by interpolation. A comparison of CQRS versus a mixture model of truncated and lognormal distributions showed slight overall improvements in quantile score (less than 1%), reliability, and to some extent also sharpness. For strong wind speed forecasts the quantile score was improved by up to 4.5% depending on lead time.

A simulated annealing approach to 3D strategic aircraft deconfliction based on en-route speed changes under wind and temperature uncertainties

We tackle the problem of minimizing the number of aircraft potential conflicts via speed regulations and taking into account uncertainties on aircraft position due to wind and temperature. The resolution is done at a strategic level, before any of the aircraft has departed. Owing to the complexity of this kind of optimisation problem, a simulated annealing metaheuristic approach is employed. A scenario with four hours of traffic overflying the Spanish (structured, continental) airspace has been selected. Inputted traffic provides routes, Mach profiles (considered to be constant), and altitude profiles as in their flight plans. Probabilistic weather forecasts from an Ensemble Prediction System are employed. Solutions provide constant speed profiles that slightly differ from those in the flight plans. It is shown that the number of conflicts can be significantly reduced by slightly modifying flight plan speeds while not altering the routes, nor the altitudes, both selected by the airspace user. The impact of this resolution strategy in flight efficiency is also analyzed.

Promoting the use of probabilistic weather forecasts through a dialogue between scientists, developers and end‐users

Today's ensemble weather prediction systems provide reliable and sharp probabilistic forecasts—yet they are still rarely communicated to outside users because of two main worries: the difficulty of communicating probabilities to lay audiences and their presumed reluctance to use probabilistic forecasts. To bridge the gap between the forecasts available and their use in day‐to‐day decision making, we encourage scientists, developers, and end‐users to engage in interdisciplinary collaborations. Here, we discuss our experience with three different approaches of introducing probabilistic forecasts to different user groups and the theoretical and practical challenges that emerged. The approaches range from quantitative analyses of users' revealed preferences online to a participatory developer–user dialogue based on trial cases and interactive demonstration tools. The examples illustrate three key points. First, to make informed decisions, users need access to probabilistic forecasts. Second, forecast uncertainty can be understood if its visual representations follow validated best practices from risk communication and information design; we highlight five important recommendations from that literature for communicating probabilistic forecasts. Third, to appreciate the value of probabilistic forecasts for their decisions, users need the opportunity to experience them in their everyday practice. With these insights and practical pointers, we hope to support future efforts to integrate probabilistic forecasts into everyday decision making.

Probabilistic load forecasting for buildings considering weather forecasting uncertainty and uncertain peak load

This paper presents a new probabilistic building load forecasting model considering uncertainties in weather forecasts and abnormal peak load. Two basis function components are developed, including the probabilistic normal load forecasting and the probabilistic uncertain peak load (or abnormal peak load) forecasting. The probabilistic normal load forecasting model is built using the artificial neural network (ANN) and the probabilistic temperature forecasts. The probabilistic abnormal peak load forecasting model consists of two models quantifying the probabilistic occurrence and magnitude of the peak abnormal differential load respectively. The test results show that the ANN deterministic load forecasting model can achieve satisfactory performance with the average mean absolute percentage error (MAPE) of 5.0%. The probabilistic occurrence model can forecast the occurrence frequency of the peak abnormal differential load with the satisfactory agreement, and the probabilistic magnitude model can well forecast the magnitudes of the peak abnormal differential load with the Kolmogorov-Smirnov error of 0.09. Real-time application case studies are conducted by different means of using the probabilistic weather forecasts. The results show that the probabilistic normal load forecasts have satisfactory accuracies and the load forecasts based on the one-day-ahead probabilistic weather forecasts are the best.

Comment

Comment

Flow-Dependent Reliability: A Path to More Skillful Ensemble Forecasts

AbstractWhile chaos ensures that probabilistic weather forecasts cannot always be “sharp,” it is important for users and developers that they are reliable. For example, they should not be overconfident or underconfident. The “spread–error” relationship is often used as a first-order assessment of the reliability of ensemble weather forecasts. This states that the ensemble standard deviation (a measure of forecast uncertainty) should match the root-mean-square error on the ensemble mean (when averaged over a sufficient number of forecast start dates). It is shown here that this relationship is now largely satisfied at the European Centre for Medium-Range Weather Forecasts (ECMWF) for ensemble forecasts of the midlatitude, midtropospheric flow out to lead times of at least 10 days when averaged over all flow situations throughout the year. This study proposes a practical framework for continued improvement in the reliability (and skill) of such forecasts. This involves the diagnosis of flow-dependent deficiencies in short-range (∼12 h) reliability for a range of synoptic-scale flow types and the prioritization of modeling research to address these deficiencies. The approach is demonstrated for a previously identified flow type, a trough over the Rockies with warm, moist air ahead. The mesoscale convective systems that can ensue are difficult to predict and, by perturbing the jet stream, are thought to lead to deterministic forecast “busts” for Europe several days later. The results here suggest that jet stream spread is insufficient during this flow type, and thus unreliable. This is likely to mean that the uncertain forecasts for Europe may, nevertheless, still be overconfident.

Expressing Flood Likelihood: Return Period versus Probability

Abstract The likelihood of floods and other potentially destructive natural phenomena is often expressed as a return period or recurrence interval, such as a 100-yr flood. However, the expression might give users the impression that the event will occur exactly once within the described period, obscuring the intended probabilistic meaning. If so, users may think a flood is less likely when one has just occurred or more likely when it has not, leading to a “flood is due” effect. This hypothesis was tested experimentally in two studies reported here. Participants were given either a return period or a probability expression and asked to rate flood likelihood and concern. Flood recency was also manipulated. The results from both studies support a flood is due effect when the return period expression is used. In the return period condition alone, participants rated floods as more likely and expressed greater concern when no flood had occurred recently. When no likelihood information was conveyed in the control condition, the opposite effect was observed. Participants rated flood likelihood as higher and expressed greater concern when a flood had occurred recently. Participants using the percent chance expression were least affected by flood recency. This adds to the growing body of research suggesting that nonexperts can benefit from probabilistic weather forecasts.

Gridded probabilistic weather forecasts with an analog ensemble

This study extends the analog ensemble (AnEn) technique to generate probabilistic forecasts of 10 m wind speed over a grid. The AnEn has been widely used to generate short‐term probabilistic predictions of meteorological variables, air quality, wind and solar power, and to effectively downscale reanalysis fields. It is based on a historical dataset including observations paired with corresponding deterministic predictions. For each forecast lead time and location, the AnEn is generated using the observations corresponding to the past deterministic predictions that are more similar to the current forecast. So far, the AnEn has been used to generate predictions at specific locations with available observations. Here, it is extended over a two‐dimensional grid, where each grid point is treated independently, using meteorological analysis instead of observations. The forecast range is extended up to 6 days ahead to evaluate the performance of the AnEn in both the short and medium range. The proposed approach allows to extend the AnEn to applications requiring gridded probabilistic products that are often generated with dynamical ensemble models. The AnEn forecasts are generated using the European Centre for Medium‐Range Weather Forecasts (ECMWF) deterministic and analysis model. Also, the ECMWF Ensemble Prediction System (EPS) is used for comparison. Given that the AnEn predictions are generated independently at any location and lead time, the resulting spatial and temporal correlation may be degraded by noise. A reordering technique called Schaake Shuffle (SS) is then applied to the ensemble members to recover the spatio‐temporal correlation. The AnEn outperforms a calibrated version of EPS for the first two days of the prediction. During the third forecast day, the AnEn remains competitive with the calibrated EPS, while the latter is more skilful from 72 to 144 h ahead. The AnEn uses about one sixth of the computational resources necessary to generate the real‐time EPS prediction.

Exploring the Usability of Probabilistic Weather Forecasts for Water Resources Decision-Making in the United Kingdom

Abstract Over the last two decades, probabilistic weather forecasts have been developed to quantify the uncertainties inherent in modeling the climate system. The skill of these forecasts has steadily increased, but the question of whether they are usable for water resources management remains open. The interdisciplinary study described in this paper combined a modeling approach with qualitative methods to identify technical and nontechnical factors that enhance or constrain the usability of probabilistic weather forecasts for reservoir management, using a case study of drought management decision-making by a water supply company in northwestern England. The modeling approach calibrated and applied probabilistic medium- and extended-range precipitation forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) to a simplified water resources system to study the technical quality of the forecasts that, in principle, could have informed management decisions during a drought event in 2010. The qualitative approach comprised initial semistructured interviews with water managers and regulators and follow-up discussions using the model experiment results to elicit further insights into the potential for incorporating probabilistic forecast information into decision-making processes. The technical analysis showed that even these postprocessed forecasts did not have skill beyond the medium range; this constrains the type of management decisions the forecasts can inform. Regulatory frameworks and attitudes to risk in the water sector also inhibit the take-up of probabilistic forecasts for drought management decisions owing to the high stakes of such decisions and considerations spanning entire water resource zones.

Evaluation of Pollen Apps Forecasts: The Need for Quality Control in an eHealth Service.

BackgroundPollen forecasts are highly valuable for allergen avoidance and thus raising the quality of life of persons concerned by pollen allergies. They are considered as valuable free services for the public. Careful scientific evaluation of pollen forecasts in terms of accurateness and reliability has not been available till date.ObjectiveThe aim of this study was to analyze 9 mobile apps, which deliver pollen information and pollen forecasts, with a focus on their accurateness regarding the prediction of the pollen load in the grass pollen season 2016 to assess their usefulness for pollen allergy sufferers.MethodsThe following number of apps was evaluated for each location: 3 apps for Vienna (Austria), 4 apps for Berlin (Germany), and 1 app each for Basel (Switzerland) and London (United Kingdom). All mobile apps were freely available. Today’s grass pollen forecast was compared throughout the defined grass pollen season at each respective location with measured grass pollen concentrations. Hit rates were calculated for the exact performance and for a tolerance in a range of ±2 and ±4 pollen per cubic meter.ResultsIn general, for most apps, hit rates score around 50% (6 apps). It was found that 1 app showed better results, whereas 3 apps performed less well. Hit rates increased when calculated with tolerances for most apps. In contrast, the forecast for the “readiness to flower” for grasses was performed at a sufficiently accurate level, although only two apps provided such a forecast. The last of those forecasts coincided with the first moderate grass pollen load on the predicted day or 3 days after and performed even from about a month before well within the range of 3 days. Advertisement was present in 3 of the 9 analyzed apps, whereas an imprint mentioning institutions with experience in pollen forecasting was present in only three other apps.ConclusionsThe quality of pollen forecasts is in need of improvement, and quality control for pollen forecasts is recommended to avoid potential harm to pollen allergy sufferers due to inadequate forecasts. The inclusion of information on reliability of provided forecasts and a similar handling regarding probabilistic weather forecasts should be considered.

Proactive energy management of solar greenhouses with risk assessment to enhance smart specialisation in China

For better time-allocation of stored energy, the solar greenhouse (SGH) is equipped with some storage devices designed economically for local weather: wall storage actively managed with energy-store/retrieve fans and Safety Energy (SE which is a solar collector and fully thermally isolated heat tank) designed for non-regular extreme weather. A proactive energy management process, addressing the optimal energy utilisation through dynamic cooperation of the wall and the SE, is presented in this paper. Based on probabilistic weather forecast and a SGH thermal model, found by system identification, the operation set-points are optimised proactively by minimising the plant probable thermal “cost” and weather-related risk in a scheduling period to take pre-emptory action against potential emergencies. The optimisation is formulated in a two-level control scheme. A master problem optimises the primary (wall-soil) storage operation against the expected weather, and a sub-problem operates the SE as a supplement to the limited wall storage in order to create a better indoor environment. The main task of the slave problem manager is to find the optimal SE operation under probable extreme weather to keep reserves to minimise any risk of severe crop loss. The overall optimisation is solved by a hybrid evolutionary algorithm based on a genetic algorithm. The tests show good potential for energy saving and crop cold stress minimisation, as well as great tolerance to forecast errors for most of the cases in Monte-Carlo simulation. The capacity of the proposed real-world system to implement the tested risk management scheme over web “recommendations” satisfies the need to close the loop of an effective Internet of Things (IoT) system, based on the MACQU (Management And Control for QUality) technological platform.

Resolution of Probabilistic Weather Forecasts with Application in Disease Management.

Predictive systems in disease management often incorporate weather data among the disease risk factors, and sometimes this comes in the form of forecast weather data rather than observed weather data. In such cases, it is useful to have an evaluation of the operational weather forecast, in addition to the evaluation of the disease forecasts provided by the predictive system. Typically, weather forecasts and disease forecasts are evaluated using different methodologies. However, the information theoretic quantity expected mutual information provides a basis for evaluating both kinds of forecast. Expected mutual information is an appropriate metric for the average performance of a predictive system over a set of forecasts. Both relative entropy (a divergence, measuring information gain) and specific information (an entropy difference, measuring change in uncertainty) provide a basis for the assessment of individual forecasts.