MEGA: Predicting the best classifier combination using meta-learning and a genetic algorithm

Classifier combination through ensemble systems is one of the most effective approaches to improve the accuracy of classification systems. Ensemble systems are generally used to combine classifiers; However, selecting the best combination of individual classifiers is a challenging task. In this paper, we propose an efficient assembling method that employs both meta-learning and a genetic algorithm for the selection of the best classifiers. Our method is called MEGA, standing for using MEta-learning and a Genetic Algorithm for algorithm recommendation. MEGA has three main components: Training, Model Interpretation and Testing. The Training component extracts meta-features of each training dataset and uses a genetic algorithm to discover the best classifier combination. The Model Interpretation component interprets the relationships between meta-features and classifiers using a priori and multi-label decision tree algorithms. Finally, the Testing component uses a weighted k-nearest-neighbors algorithm to predict the best combination of classifiers for unseen datasets. We present extensive experimental results that demonstrate the performance of MEGA. MEGA achieves superior results in a comparison of three other methods and, most importantly, is able to find novel interpretable rules that can be used to select the best combination of classifiers for an unseen dataset.

$$\mu PT$$ statistical ensemble: systems with fluctuating energy, particle number, and volume

Within the theory of statistical ensemble, the so-called $$\mu PT$$ μ P T ensemble describes equilibrium systems that exchange energy, particles, and volume with the surrounding. General, model-independent features of volume and particle number statistics are derived. Non-analytic points of the partition function are discussed in connection with divergent fluctuations and ensemble equivalence. Quantum and classical ideal gases, and a model of Bose gas with mean-field interactions are discussed as examples of the above considerations.

Accounting for land model error in numerical weather prediction ensemble systems: toward ensemble-based coupled land/atmosphere data assimilation

The ensembles used in the NOAA National Centers for Environmental Prediction (NCEP) global data assimilation and numerical weather prediction (NWP) system are under-dispersed at and near the land surface, preventing their use in ensemble-based land data assimilation. Comparison to offline (land-only) data assimilation ensemble systems suggests that while the relevant atmospheric fields are under-dispersed in NCEP’s system, this alone cannot explain the under-dispersed land component, and an additional scheme is required to explicitly account for land model error. This study then investigates several schemes for perturbing the soil (moisture and temperature) states in NCEP’s system, qualitatively comparing the induced ensemble spread to independent estimates of the forecast error standard deviation in soil moisture, soil temperature, 2m temperature, and 2m humidity. Directly adding perturbations to the soil states, as is commonly done in offline systems, generated unrealistic spatial patterns in the soil moisture ensemble spread. Application of a Stochastically Perturbed Physics Tendencies scheme to the soil states is inherently limited in the amount of soil moisture spread that it can induce. Perturbing the land model parameters, in this case vegetation fraction, generated a realistic distribution in the ensemble spread, while also inducing perturbations in the land (soil states) and atmosphere (2m states) that are consistent with errors in the land/atmosphere fluxes. The parameter perturbation method is then recommended for NCEP’s ensemble system, and it is currently being refined within the development of an ensemble-based coupled land/atmosphere data assimilation for NCEP’s NWP system.

EUMETNET SRNWPEPS-EPS project and convection-permitting LAM-EPS database

<p>SRNWP-EPS module/project into EUMETNET NWP Cooperation Programme has as main goals facilitating and coordinating the cooperation on developing reliable mesoscale convection-permitting ensemble systems (LAM-EPS) in Europe, and, at the same time, grouping efforts developing tools which can be smoothly applied to any LAM-EPS. This is motivated by the fact that the development of LAM-EPS capabilities in Europe is crucial for forecasting a range of weather phenomena and in particular for improving HIW (High Impact Weather) prediction. Due   to the latter, the current SRNWP-EPS 2019-2023 phase is focused on extreme events.</p><p>The project results as a survey on products for high-impact weather forecasting and the R2O (Research to Operations) LAM-EPS applications will be presented. The three main R2O forecasting tools developed as project requirements are: calibration of daily and  12 hours extremes for variables such as 10 metres maximum wind gusts, maximum accumulated precipitation, maximum and minimum2m temperatures; the forecasting post-processing LAM-EPS products devoted to HIW forecasting and focused on aeronautics such as icing, thunderstorms’ diagnostic and classification, clear-air turbulence and fog; and tools to apply in an affordable way an Extreme Forecast Index (EFI) and Shift of Tales Index (SOT) on LAM-EPSs.</p><p>Moreover, an off-line database of European convection-permitting LAM-EPS ensembles has been established at ECMWF, which archives convection related parameters close to the surface. The aim of LAM-EPS database is to foster coordinate research and collaborations around LAM-EPSs in order to improve HIW events bringing together all European LAM-NWP consortia (ALADIN, HIRLAM, COSMO, LACE, MetOffice partners, etc.). At the time of writing, nine participants are currently archiving since 1<sup>st</sup> of June of 2020: MOGREPS-UK (MetOffice), MEPS (MetCoOp), <em>γ</em>SREPS (AEMET), IT-EPS (ItAF-REMET), IREPS (Met Éireann), COMEPS (DMI), MF-AromeEps (MétéoFrance), RMI-EPS (RMI) and ICON-D2-EPS (DWD). The SRNWP-EPS convection-permitting LAM-EPS database is currently being used by project research sub-groups, for example to check multi-ensemble performance or comparing two LAM-EPSs in their common overlapping area.</p>

Validation of Ensemble-Based Probabilistic Tropical Cyclone Intensity Change

Although there has been substantial improvement to numerical weather prediction models, accurate predictions of tropical cyclone rapid intensification (RI) remain elusive. The processes that govern RI, such as convection, may be inherently less predictable; therefore a probabilistic approach should be adopted. Although there have been numerous studies that have evaluated probabilistic intensity (i.e., maximum wind speed) forecasts from high resolution models, or statistical RI predictions, there has not been a comprehensive analysis of high-resolution ensemble predictions of various intensity change thresholds. Here, ensemble-based probabilities of various intensity changes are computed from experimental Hurricane Weather Research and Forecasting (HWRF) and Hurricanes in a Multi-scale Ocean-coupled Non-hydrostatic (HMON) models that were run for select cases during the 2017–2019 seasons and verified against best track data. Both the HWRF and HMON ensemble systems simulate intensity changes consistent with RI (30 knots 24 h−1; 15.4 m s−1 24 h−1) less frequent than observed, do not provide reliable probabilistic predictions, and are less skillful probabilistic forecasts relative to the Statistical Hurricane Intensity Prediction System Rapid Intensification Index (SHIPS-RII) and Deterministic to Probabilistic Statistical (DTOPS) statistical-dynamical systems. This issue is partly alleviated by applying a quantile-based bias correction scheme that preferentially adjusts the model-based intensity change at the upper-end of intensity changes. While such an approach works well for high-resolution models, this bias correction strategy does not substantially improve ECMWF ensemble-based probabilistic predictions. By contrast, both the HWRF and HMON systems provide generally reliable predictions of intensity changes for cases where RI does not take place. Combining the members from the HWRF and HMON ensemble systems into a large multi-model ensemble does not improve upon HMON probablistic forecasts.

Benefits of stochastic parameterizations in subseasonal to seasonal (S2S) forecasts

<p>Recently, there has been much interest in issuing subseasonal to seasonal (S2S) forecasts, although their skill is often debated. In addition to large systematic errors, ensemble systems are often overconfident, i.e. have incorrect information about the uncertainty of a particular forecast. Stochastic parameterization schemes are used routinely to remedy the problem of overconfidence, but also have the potential to reduce systematic model errors. </p><p>Here, we study the impact of adding a stochastic parameterization scheme in coupled simulations with the climate model CESM.  Physical processes associated with S2S-predictability, like the Madden-Julian  Oscillation (MJO) and Northern Hemispheric blocking are analyzed. In the simulations with a stochastic parameterization scheme, the northward propagation of the MJO is captured better, leading to an improved MJO lifecycle. The impact on other atmospheric fields like precipitation and winds will be discussed. </p>

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 $$4.5\%$$ 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 $$1.7 \%$$ 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.