Deterministic 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.

Comparison of deterministic and ensemble weather forecasts on ship sailing speed optimization

Ship sailing speed optimization models are constructed based on prediction of ship fuel consumption, whose accuracy is highly influenced by the quality of sea and weather information. In this study, we develop two fusion methods for combining external meteorological data with ship noon report data, including the rhumb line based fusion method and the direct fusion method, and compare them in terms of accuracy in providing meteorological data. Next, we propose a framework based on the better data fusion strategy for comparing the impacts of deterministic and ensemble weather forecasts on ship speed optimization performance, enabling the evaluation of ship fuel consumptions under different speed plans based on weather forecast data available before departure. Results show that speed optimization based on ensemble weather forecasts has greater potential than that based on deterministic weather forecasts to diminish ship fuel consumption and thus to reduce greenhouse gas emissions.

Verification of a global weather forecasting system for decision-making in farming over Africa

In the framework of AfriCultuReS project, operational deterministic weather forecasts provide valuable information on the expected weather conditions over the African continent as a part of federation of services within the project. In this study, we investigate the performance of the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) over pilot regions in Africa by utilizing available surface observations, satellite and reanalysis data. The verification period covers two consecutive years (June 2018–May 2020). In addition, we assess the ability of the model to provide skillful forecasts through three high-impact precipitation events that occurred during this period. The results show that the model presents both positive and negative biases with respect to its predicted near surface air temperature, underestimates the near surface relative humidity and the mean sea-level pressure, while overestimates the wind speed at 10 m. The neighborhood-based statistical verification of the 24-h accumulated precipitation reveals that the model forecasts the precipitation events more accurately as the verification area is increasing but at higher precipitation thresholds its performance deteriorates. Different variability, errors and correlation between simulated and observed precipitation exist in each forecast lead day and region. A range of model behavior and forecast skill is found with respect to the examined three precipitation events. Skillful forecasts up to four days ahead were provided in the cases of the Tropical Cyclone IDAI and the flash flooding events in northern Tunisia, while the lowest performance was found in the region of the West African Monsoon.

Towards real-time probabilistic ash deposition forecasting for New Zealand

Volcanic ashfall forecasts are highly dependent on eruption source parameters (ESPs) and synoptic weather conditions at the time and location of the eruption. In New Zealand, MetService and GNS Science have been jointly developing an ashfall forecast system that incorporates four-dimensional high-resolution numerical weather prediction (NWP) and ESPs into the HYSPLIT model, a state-of-the art hybrid Eulerian and Lagrangian dispersion model widely used for volcanic ash. However, these forecasts are based on discrete ESPs combined with a deterministic weather forecast and thus provide no information on output uncertainty. This shortcoming hinders stakeholder decision making, particularly near the geographical margin of forecasted ashfall and in areas with large gradients in forecasted ash deposition. Our study presents a new approach that incorporates uncertainty from both eruptive and meteorological inputs to deliver uncertainty in the model output. To this end, we developed probability density functions (PDFs) for three key ESPs (plume height, mass eruption rate, eruption duration) tailored to New Zealand’s volcanoes and combine them with NWP ensemble datasets to generate probabilistic ashfall forecasts using the HYSPLIT model. We show that the Latin Hypercube Sampling (LHS) technique can be used to representatively span this four-dimensional parameter space and allow us to add uncertainty quantification to rapid response forecast systems. For a case study of a hypothetical eruption at Tongariro, New Zealand we suggest that large parts of New Zealand’s North Island would not receive adequate warning for potential ashfall if uncertainties were not included in the forecasts. We also propose new probabilistic summary products to support public information and emergency responders decision making.

Committor Functions for Climate Phenomena at the Predictability Margin: The Example of El Niño–Southern Oscillation in the Jin and Timmermann Model

Abstract Many atmosphere and climate phenomena lie in the gray zone between weather and climate: they are not amenable to deterministic forecast, but they still depend on the initial condition. A natural example is medium-range forecasting, which is inherently probabilistic because it lies beyond the deterministic predictability time of the atmosphere, but for which statistically significant prediction can be made, which depends on the current state of the system. Similarly, one may ask the probability of occurrence of an El Niño event several months ahead of time. We introduce a quantity that corresponds precisely to this type of prediction problem: the committor function is the probability that an event takes place within a given time window, as a function of the initial condition. We compute it in the case of a low-dimensional stochastic model for El Niño, the Jin and Timmermann model. In this context, we show that the ability to predict the probability of occurrence of the event of interest may differ strongly depending on the initial state. The main result is the new distinction between probabilistic predictability (when the committor function is smooth and probability can be computed, which does not depend sensitively on the initial condition) and probabilistic unpredictability (when the committor function depends sensitively on the initial condition). We also demonstrate that the Jin and Timmermann model might be the first example of a stochastic differential equation with weak noise for which transition between attractors does not follow the Arrhenius law, which is expected based on large deviation theory and generic hypothesis. Significance Statement A key problem for atmospheric and climate phenomena is to predict events beyond the time scale over which deterministic weather forecast is possible. In a simple model of El Niño, we demonstrate the existence of two regimes, depending on initial conditions. For initial conditions in the “probabilistic predictability” regime, the system is unpredictable deterministically because of chaos, but the probability of occurrence of the event can still be predicted because it depends only weakly on the initial condition. In the “probabilistic unpredictability” regime, even predicting probabilities is difficult, because the probability depends strongly on initial conditions. These new concepts of probabilistic predictability and unpredictability should be key in understanding the predictability potential for rare events in climate problems, as well as in other complex dynamics.

Reply to “Comments on ‘What Is the Predictability Limit of Midlatitude Weather?’”

Abstract In their comment, Žagar and Szunyogh raised concerns about a recent study by Zhang et al. that examined the predictability limit of midlatitude weather using two up-to-date global models. Zhang et al. showed that deterministic weather forecast may, at best, be extended by 5 days, assuming we could achieve minimal initial-condition uncertainty (e.g., 10% of current operational value) with a nearly perfect model. Žagar and Szunyogh questioned the methodology and the experiments of Zhang et al. Specifically, Žagar and Szunyogh raised issues regarding the effects of model error on the growth of the forecast uncertainty. They also suggested that estimates of the predictability limit could be obtained using a simple parametric model. This reply clarifies the misunderstandings in Žagar and Szunyogh and demonstrates that experiments conducted by Zhang et al. are reasonable. In our view, the model error concern in Žagar and Szunyogh does not apply to the intrinsic predictability limit, which is the key focus of Zhang et al. and the simple parametric model described in Žagar and Szunyogh does not serve the purpose of Zhang et al.

Developing an Automated Medium-Range Flood Awareness System for Switzerland Based on Probabilistic Forecasts of Integrated Water Vapor Fluxes

Abstract Floods in the Alpine region can be destructive and cause large economic losses. Many rivers and lakes in Switzerland are regulated and flood damage can be mitigated through an optimal management of lake levels and runoff. This requires high-quality forecasts of atmospheric flood precursors extending beyond short-range (forecast days 1–5) predictions. In several places around the world atmospheric rivers or extreme integrated vapor transport (IVT) are causally related to flood events. Also in Switzerland, extreme IVT oriented perpendicular to the main orography heralds extreme flood events. This relationship is exploited in an operational flood warning system on the medium-range (here forecast days 6–10) time scale based on probabilistic medium-range forecasts of IVT and precipitation over Switzerland provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS). This entails first a comprehensive probabilistic verification of the direction and magnitude of (extreme) IVT and second the development of compact visualizations for the operational use by hydrologists. Based on 20 years of probabilistic reforecasts, we show that both regular and extreme IVT has a better predictability than precipitation and IVT is predictable out to day 8. As the direction of IVT is of central importance for flood risk in Switzerland, we develop a visualization that summarizes probabilistic information on both the direction and magnitude of the IVT together with users of the product. The result is an operational flood warning system based solely on atmospheric flood precursors to extend flood warning information beyond the range of high-resolution deterministic weather forecasts.

Communication and interpretation of regional weather forecasts: a survey of the Italian public

The aim of the present study is to contribute to the correctness and effectiveness of weather forecast communication, the importance of which has been steadily growing along with the improvement in numerical weather prediction models and methods as well as the general awareness about the increase of extreme events within a context of global climate change. An extensive survey was conducted among the general users of the weather forecasts issued by the regional meteorological service of Tuscany, Italy (LaMMA Consortium), which resulted in 2388 volunteers responding to the questions aimed at better understanding of how people access, interpret and use weather forecasts. The survey also includes some items investigated in previous research, allowing comparison with similar findings in other countries. The most critical issue concerns the uncertainty information, investigated with the main aim of verifying the existence and relevance of inferential mechanisms in the interpretation of weather icons and maps used in LaMMA forecasts to assess uncertainty. The present study also discusses users' interpretations of the probability of precipitation forecasts and their preferences on how forecast uncertainty is conveyed. Results show that, even if the Italian public is accustomed to strictly deterministic weather forecasts, people attribute uncertainty to them on their own even if lacking any explicit indication, thus suggesting the need to supplement the existing forecasts with both graphical and textual information about uncertainty, particularly in the case of precipitation forecasts.

Highway Accident Modeling and Forecasting in Winter

Environmental attributes are critical risk factors that have proven to affect collision rates. Associated driving risks can be reduced by better maintenance of roadway infrastructure, enforcement of speed limits or other traffic laws. Given the preventive nature of these policies and regulations, accurate predictions of environmental attributes are needed. Currently, most of road safety prediction models are based on deterministic weather forecasts which are not able to capture changes in the likelihood of collision occurrence. As a result, probabilistic forecast is required to improve decision making, mainly in winter. In this paper, a stochastic approach to modeling highway collisions in winter time is considered which enables better assessment of driving conditions and a more accurate prediction. A logistic regression model with covariates is applied to crash data where environmental characteristics are modeled as a finite state space homogeneous multivariate discrete time Markov chain. After fitting the model, weather prediction as well as the conditional predictive probability distribution of collision occurrence are obtained. As the application, the ability of the proposed model to predict hourly environmental attributes and collision occurrence is examined using real highway crash data. The performance of the developed stochastic model is compared with several existing models in the literature using actual collision data. The results demonstrate that the proposed stochastic model outperforms existing models and it accurately predicts collision occurrence in the presence of stochastically changeable winter weather conditions. As a result, the proposed probabilistic forecast model can be used as a valuable tool in a decision support system.

Translation of Ensemble Weather Forecasts into Probabilistic Air Traffic Capacity Impact

Today’s trend in probabilistic weather forecasting is toward utilizing ensemble prediction systems. In the Next Generation Air Transportation System (NextGen), ensemble-based weather forecasting will be a common practice. Therefore, this paper explores a novel approach of using high-resolution, ensemble-based numerical weather prediction model data for weather-related, probabilistic aviation impact forecasting. The concept represents a paradigm shift from “creating ensembles of weather information” (e.g., maps of predicted weather hazard intensity) to “developing ensembles of aviation-relevant information” (maps of potential throughput as measured by the available flow capacity ratio), which entails a translation of weather forecasts into predictions of reduced airspace capacity. The proof-of-concept is exemplified by focusing on convective storms; however, in principal, the approach may be applicable to other aviation hazards, like turbulence, icing, or ceiling and visibility. The concept is most pertinent to strategic en route traffic flow management, but it also applies to terminal area applications. A probabilistic approach is appropriate for strategic planning horizons, for which deterministic weather forecasts are significantly less accurate and an ensemble of forecasts may provide guidance about the weather (and impact) uncertainty.

Probabilistic Weather Forecasting for Winter Road Maintenance

Winter road maintenance is one of the main tasks for the Washington State Department of Transportation. Anti-icing, that is, the preemptive application of chemicals, is often used to keep the roadways free of ice. Given the preventive nature of anti-icing, accurate predictions of road ice are needed. Currently, anti-icing decisions are usually based on deterministic weather forecasts. However, the costs of the two kinds of errors are highly asymmetric because the cost of a road closure due to ice is much greater than that of taking anti-icing measures. As a result, probabilistic forecasts are needed to optimize decision making.We propose two methods for forecasting the probability of ice formation. Starting with deterministic numerical weather predictions, we model temperature and precipitation using distributions centered around the bias-corrected forecasts. This produces a joint predictive probability distribution of temperature and precipitation, which then yields the probability of ice formation, defined here as the occurrence of precipitation when the temperature is below freezing. The first method assumes that temperatures, as well as precipitation, at different spatial locations are conditionally independent given the numerical weather predictions. The second method models the spatial dependence between forecast errors at different locations. The model parameters are estimated using a Bayesian approach via Markov chain Monte Carlo.We evaluate both methods by comparing their probabilistic forecasts with observations of ice formation for Interstate Highway 90 in Washington State for the 2003–2004 and 2004–2005 winter seasons. The use of the probabilistic forecasts reduces costs by about 50% when compared to deterministic forecasts. The spatial method improves the reliability of the forecasts, but does not result in further cost reduction when compared to the first method.

Uncertainty assessment via Bayesian revision of ensemble streamflow predictions in the operational river Rhine forecasting system

Ensemble streamflow forecasts obtained by using hydrological models with ensemble weather products are becoming more frequent in operational flow forecasting. The uncertainty of the ensemble forecast needs to be assessed for these products to become useful in forecasting operations. A comprehensive framework for Bayesian revision has been recently developed and applied to operational flood forecasting with deterministic weather forecasts. The Bayesian revision yields a posterior density, conditional on all information available to the forecaster at the onset of a forecast run. This conditional density objectively quantifies the uncertainty. Here the Bayesian approach is generalized for use with ensemble weather predictions. An end‐to‐end application of a Bayesian postprocessor for ensemble streamflow forecasts in the river Rhine forecasting system is presented. A verification of the postprocessor shows good performance when compared in terms of the ranked probability skill score to non‐Bayesian uncertainty assessment, such as ranking threshold exceedance probabilities for members of a streamflow ensemble prediction. In this context it is also addressed how the proposed Bayesian processor can serve in supporting rational decision making for flood warning under conditions of uncertainty.

Automated Route Generation for Avoiding Deterministic Weather in Transition Airspace

We investigate the problem of synthesizing weather avoidance routes in the transition airspace, from the 200 n mile range to the metering fixes of an airport, given a deterministic weather forecast. The investigation of the problem is motivated by the desire to maximize airspace capacity while ensuring safe separation between aircraft and between aircraft and hazardous weather. Three solution methods are compared with current day operations. Emphasis is placed on a comparison of the arrival traffic weather avoidance routing method and metrics associated with such routes. Historical weather avoidance paths (baseline) are compared with three alternatives: variations of the standard arrival routes, a geometric optimization solution synthesizing multiple nonintersecting routes, and two free-flight approaches in which aircraft fly weather avoidance routes using a greedy prioritization method. Results indicate that increases in capacity over today's system are achievable, while maintaining safety. However, such increases are limited by the method, the required supporting infrastructure, and weather severity and location.

Predictability Mysteries in Cloud-Resolving Models

Abstract The rapid amplification of small-amplitude perturbations by the chaotic nature of the atmospheric dynamics intrinsically limits the skill of deterministic weather forecasts. In this study, limited-area cloud-resolving numerical weather prediction (NWP) experiments are conducted to investigate the role of mesoscale processes in determining predictability. The focus is set on domain-internal error growth by integrating an ensemble of simulations using slightly modified initial conditions but identical lateral boundary conditions. It is found that the predictability of the three investigated cases taken from the Mesoscale Alpine Programme (MAP) differs tremendously. In terms of normalized precipitation spread, values between 0.05 (highly predictable) and 1 (virtually unpredictable) are obtained. Analysis of the derived ensemble spread demonstrates that the diabatic forcing associated with moist dynamics is the prime source of rapid error growth. However, in agreement with an earlier study it is found that the differentiation between convective and stratiform rain is unable to account for the distinctive precipitation spreads of the three cases. In particular, instability indices are demonstrated to be poor predictors of the predictability level. An alternate hypothesis is proposed and tested. It is inspired by the dynamical instability theory and states that significant loss of predictability only occurs over moist convectively unstable regions that are able to sustain propagation of energy against the mean flow. Using a linear analysis of gravity wave propagation, this hypothesis is shown to provide successful estimates of the predictability level for the three cases under consideration.