Search Results

You are looking at 1 - 10 of 53 items for

  • Author or Editor: Paul J. Roebber x
  • Refine by Access: All Content x
Clear All Modify Search
Paul J. Roebber

Abstract

Evolutionary programming is applied to the postpocessing of ensemble forecasts of temperature on a spatial domain. These forecasts are obtained from the 11-member Reforecast V2 ensemble over the region from 24°–53°N to 125°–66°W for the period 1 January 1985–14 May 2011. The evolution is based upon a static ecosystem model that holds constant the number of individuals (algorithms), using a fixed rate of introduction of new algorithms and removal of existing algorithms. Each algorithm adheres to a specific underlying genetic architecture, and the selection pressure on the “species” is according to deterministic performance (root-mean-square error) on a training dataset. On a 2325-case, independent test dataset, the method improved root-mean-square error and ranked probability score relative to the Reforecast ensemble by 0.31°F (8.7%) and 3.3%, respectively, across the domain, with 96% of the grid points showing simultaneous improvements in both measures. The use of input information by the evolutionary programming algorithms varied by region; while the algorithm forecasts at all locations are fundamentally tied to the Reforecast ensemble forecast, northeastern locations found snow cover to be the next most useful input, whereas southwestern locations preferentially employed precipitable water. An adaptive form of the approach, developed to be readily implemented into operations, is tested in the absence of improving inputs but is found to only slightly degrade performance (1.2% in root-mean-square error and 0.6% in ranked probability skill score). A number of future extensions are discussed.

Full access
Paul J. Roebber

Abstract

Numerous observational and modeling studies have suggested the importance of cyclogenesis to the breakdown of the zonal flow and the maintenance of atmospheric blocks, whereas other studies have shown that low-frequency dynamics are sufficient to produce and maintain these patterns. Experiments with a simple model of the general circulation and empirical evidence obtained from reanalysis and cyclone tracking data are used to develop a conceptual understanding of this irregular response. The model’s results and observational data are consistent with the idea that the North Atlantic flow response to cyclone forcing is preconditioned by the state of the hemispheric circulation. A characteristic hemispheric flow configuration reminiscent of the shedding of a potential vorticity (PV) filament and tightening of the PV gradient is particularly responsive to cyclogenesis, with the likelihood of below 10th percentile North Atlantic zonal flow under these conditions increased by a factor of 2.37 for each cyclone event. A second characteristic pattern, reminiscent of the PV wave-breaking anticyclonic roll up, responds in an opposite way to cyclogenesis, with a decrease in the likelihood of below 10th percentile zonal flow by a factor of 0.67. This perspective is connected to the decades of research on blocking and opens an opportunity for the development of medium-range forecast model postprocessors that might provide probabilistic forecasts of North Atlantic flow transitions.

Full access
Paul J. Roebber

Abstract

A diagnostic study of two, successive operational model forecasts of a case of explosive cyclogenesis is presented, with the goal of understanding the rather substantial differences in the simulations. The rapid cyclogenesis, Which occurred to varying degree in both forecasts, can be explained as a moist baroclinic response to a strong 500-mb trough embedded within the polar airstream. The variability in the forecasts is related to differential growth of low-level cyclonic vorticity in association with amplification of the 500-mb vorticity gradient between the upstream trough and a locally forced downstream short-wave ridge, prior to the period of most rapid deepening. This antecedent vorticity growth was initiated by advection offshore of the east coast of North America of a tongue of stratospheric potential vorticity, identifiable in the conventional constant analysis as a weak short-wave trough at 500 mb. Once initiated, low-level development continued as a result of a self-development process involving an interaction between quasigeostrophic forcing of ascent and latent heat release; upon the arrival of the polar trough, rapid surface deepening ensued. The self-development process during the antecedent stage effectively lengthened the time scale of intensification, leading to greater increases in surface relative vorticity through vortex stretching. In addition, the upstream 500-mb trough was amplified during this period. The weaker development in the less successful simulation of this case occurred as a result of damped self-development and consequently reduced low-level vorticity and weaker cyclonic vorticity advection during the rapid deepening stage. The impact on predictability of such nonlinear interactions between process during cyclogenesis is discussed, with reference to short-and medium-range forecasts.

Full access
Paul J. Roebber

Difficulties associated with the teaching of complex subjects such as the atmospheric sciences create obstacles to learning and lead to relatively high rates of student attrition. An exploration of the role of mismatches between student learning styles and that implicit in curricular design was conducted at the University of Wisconsin—Milwaukee (UWM), with the objective of identifying methods for improving student learning and retention.

Open-ended interviews were used to elicit the opinions of past and present students and faculty. These data are analyzed to meet the study objectives. Key findings include the following: attrition rates in the program are high, but consistent with published rates across the United States in engineering; the predominant learning styles of students and faculty diverge substantially; curricular design is consistent with faculty rather than student learning styles; among students, undergraduates show the largest negative responses to existing modes of operation and the most interest in change; faculty also show considerable discomfort with existing modes and substantial support for change, although their rationale for this support may differ from that of students; support for a radical reorganization of the curriculum toward a case-study-driven learning process is weak, particularly among undergraduates; increased emphasis of physical examples and case studies within the existing curricular framework is supported, both for upper-level undergraduates and graduate students. Methods for addressing these limitations within atmospheric science curricula are presented.

Full access
Paul J. Roebber

Abstract

A new method for producing large member ensemble forecasts, using a variation of evolutionary programming (EP), is presented. A series of increasingly complex datasets are used to demonstrate the method and its potential utility. First, EP performance is considered using training and test “atmospheric” data derived from the Lorenz low-order dynamical system. Next, a modified form of the intermediate-order Lorenz model representing 500-hPa height is used. Finally, EP performance is evaluated using real 500-hPa data and day-3 forecasts of the reforecast model. As expected, short observational records limit the potential of the EP method by preventing proper training. A kind of perfect-prog approach, in which the EP ensemble is trained using a large “observational” sample constructed from the imperfect model, is shown to be a potentially viable means of counteracting limits to the observed record. The EP ensembles are shown to outperform dynamical model ensembles at the extremes, and to be competitive with dynamical models across a wide range of values of the variables of interest. In particular, the EP ensembles improve resolution compared to dynamical model ensembles, which suggests that further skill might be obtainable by also improving reliability using additional postprocessing. Future applications of the ensemble EP method are briefly discussed.

Full access
Paul J. Roebber

Abstract

A statistical analysis of 12 and 24 hour deepening rates for all surface lows analyzed on at least two successive NMC 12 hourly “front half” hemispheric surface charts was performed for one year of data. Both 12 and 24 hour deepening distributions showed statistically significant (at the 5% level) departures from normality, with the largest deviations occurring along the tail of the distribution associated with most rapid deepening. The sum of two normal curves of different means and standard deviations was successfully fitted to the deepening distributions, suggesting that most cases of explosive cyclogenesis are the result of some additional physical mechanism distinct from ordinary baroclinic instability.

The climatology of explosive cyclones (Sanders and Gyakum) was updated to include the 1979–82 cold seasons, and compared to the previous three-year sample. In addition, a climatology of formation positions, maximum deepening positions and dissipation positions for all cyclones in a one-year data sample was compiled. These studies indicate that the preferred regions of explosive cyclogenesis are primarily baroclinic zones, the climatological and statistical evidence therefore suggests that the explosive mechanism is a combination of the baroclinic process and some other mechanism or mechanisms.

Full access
Paul J. Roebber

Abstract

We introduce an adaptive form of postprocessor where algorithm structures are neural networks where the number of hidden nodes and the network training features evolve. Key potential advantages of this system are the flexible, nonlinear mapping capabilities of neural networks and, through backpropagation, the ability to rapidly establish capable predictors in an algorithm population. The system can be implemented after one initial training process and future changes to postprocessor inputs (new observations, new inputs, or model upgrades) are incorporated as they become available. As in prior work, the implementation in the form of a predator–prey ecosystem allows for the ready construction of ensembles. Computational requirements are minimal, and the use of a moving data window means that data storage requirements are constrained. The system adds predictive skill to a demonstration dynamical model representing the hemispheric circulation, with skill competitive with or exceeding that obtainable from multiple linear regression and standard artificial neural networks constructed under typical operational limitations. The system incorporates new information rapidly and the dependence of the approach on the training data size is similar to multiple linear regression. A loss of performance occurs relative to a fixed neural network architecture in which only the weights are adjusted after training, but this loss is compensated for by gains from the ensemble predictions. While the demonstration dynamical model is complex, current numerical weather prediction models are considerably more so, and thus a future step will be to apply this technique to operational weather forecast data.

Restricted access
Paul J. Roebber

Abstract

Statistical analysis of cyclone deepening rates has been used in the past to infer distinctions between physical processes operative in cases of explosive cyclogenesis and lesser storms. This note attempts to qualify the conclusions of the previous study by analyzing cyclone deepening data from a new prospective. The results suggest that the debate concerning the relative normality of these distributions is essentially irrelevant. Significant statistical evidence is provided to suggest that midlatitude maritime cyclogenesis exhibits a fundamentally different character from continental events, and that this distinction is evident across a wide spectrum of storm intensities.

Full access
Paul J. Roebber

Abstract

A structurally simple analytic quasi-geostrophic model is used to investigate the role of diabatic process resulting from surface fluxes of heat and moisture associated with ocean current meanders in midlatitude maritime cyclogenesis. The combined effect of sensible heat transfer and convective latent heating resulting both from surface water vapor fluxes and large-scale moisture convergence can contribute both directly and indirectly to cyclogenesis, the extent of the contribution being dependent on a number of cooperative processes interacting within the cyclone. Latent and sensible heating associated with shallow convection in a moist adiabatic environment in which the large scale convergence supplies ample moisture to the cumulus clouds represent the conditions under which the greatest enhancement of adiabatic model cyclogenesis occurs; but cyclogenesis induced by differential vorticity advection is 90° out of phase (with respect to the Large-scale tropospheric thermal wave) with direct diabatic intensification associated with sensible heating. The indirect effects of surface fluxes associated with the moistening and destabilization of the boundary layer and the amplification of the overall atmospheric baroclinity have greater potential to enhance model cyclogenesis than direct diabatic process under conditions of strong adiabatic forcing. Preliminary observational evidence suggests that the occurrence of explosive cyclogenesis over the western Atlantic Ocean is marginally more likely when cyclones propagate across the positions of mean warm sea surface temperature anomalies along the Gulf Stream boundary than in the absence of such features. The existence of such anomalies is neither a necessary nor a sufficient condition for explosive cyclogenesis to occur, but way enhance the development in some cases.

Full access
Paul J. Roebber

Abstract

Simulated evolution is used to generate consensus forecasts of next-day minimum temperature for a site in Ohio. The evolved forecast algorithm logic is interpretable in terms of physics that might be accounted for by experienced forecasters, but the logic of the individual algorithms that form the consensus is unique. As a result, evolved program consensus forecasts produce substantial increases in forecast accuracy relative to forecast benchmarks such as model output statistics (MOS) and those from the National Weather Service (NWS). The best consensus produces a mean absolute error (MAE) of 2.98°F on an independent test dataset, representing a 27% improvement relative to MOS. These results translate to potential annual cost savings for electricity production in the state of Ohio of the order of $2 million relative to the NWS forecasts. Perfect forecasts provide nearly $6 million in additional annual electricity production cost savings relative to the evolved program consensus.

The frequency of outlier events (forecast busts) falls from 24% using NWS to 16% using the evolved program consensus. Information on when busts are most likely can be provided through a logistic regression equation with two variables: forecast wind speed and the deviation of the NWS minimum temperature forecast from persistence. A forecast of a bust is 4 times more likely to be correct than wrong, suggesting some utility in anticipating the most egregious forecast errors. Discussion concerning the probabilistic applications of evolved programs, the application of this technique to other forecast problems, and the relevance of these findings to the future role of human forecasting is provided.

Full access