Distributional semantics and machine learning for statistical machine translation

[EU]Lan honetan semantika distribuzionalaren eta ikasketa automatikoaren erabilera aztertzen dugu itzulpen automatiko estatistikoa hobetzeko. Bide horretan, erregresio logistikoan oinarritutako ikasketa automatikoko eredu bat proposatzen dugu hitz-segiden itzulpen- probabilitatea modu dinamikoan modelatzeko. Proposatutako eredua itzulpen automatiko estatistikoko ohiko itzulpen-probabilitateen orokortze bat dela frogatzen dugu, eta testuinguruko nahiz semantika distribuzionaleko informazioa barneratzeko baliatu ezaugarri lexiko, hitz-cluster eta hitzen errepresentazio bektorialen bidez. Horretaz gain, semantika distribuzionaleko ezagutza itzulpen automatiko estatistikoan txertatzeko beste hurbilpen bat lantzen dugu: hitzen errepresentazio bektorial elebidunak erabiltzea hitz-segiden itzulpenen antzekotasuna modelatzeko. Gure esperimentuek proposatutako ereduen baliagarritasuna erakusten dute, emaitza itxaropentsuak eskuratuz oinarrizko sistema sendo baten gainean. Era berean, gure lanak ekarpen garrantzitsuak egiten ditu errepresentazio bektorialen mapaketa elebidunei eta hitzen errepresentazio bektorialetan oinarritutako hitz-segiden antzekotasun neurriei dagokienean, itzulpen automatikoaz haratago balio propio bat dutenak semantika distribuzionalaren arloan.

Spherical nonparametric estimators applied to the UGAMP model integration for AMIP

The aim of this paper is essentially twofold: first, to describe the use of spherical nonparametric estimators for determining statistical diagnostic fields from ensembles of feature tracks on a global domain, and second, to report the application of these techniques to data derived from a modern general circulation model. New spherical kernel functions are introduced that are more efficiently computed than the traditional exponential kernels. The data-driven techniques of cross-validation to determine the amount elf smoothing objectively, and adaptive smoothing to vary the smoothing locally, are also considered. Also introduced are techniques for combining seasonal statistical distributions to produce longer-term statistical distributions. Although all calculations are performed globally, only the results for the Northern Hemisphere winter (December, January, February) and Southern Hemisphere winter (June, July, August) cyclonic activity are presented, discussed, and compared with previous studies. Overall, results for the two hemispheric winters are in good agreement with previous studies, both for model-based studies and observational studies.

Statistical inference of OH concentrations and air mass dilution rates from successive observations of non-methane hydrocarbons in single air masses

Bayesian inference has been used to determine rigorous estimates of hydroxyl radical concentrations () and air mass dilution rates (K) averaged following air masses between linked observations of nonmethane hydrocarbons (NMHCs) spanning the North Atlantic during the Intercontinental Transport and Chemical Transformation (ITCT)-Lagrangian-2K4 experiment. The Bayesian technique obtains a refined (posterior) distribution of a parameter given data related to the parameter through a model and prior beliefs about the parameter distribution. Here, the model describes hydrocarbon loss through OH reaction and mixing with a background concentration at rate K. The Lagrangian experiment provides direct observations of hydrocarbons at two time points, removing assumptions regarding composition or sources upstream of a single observation. The estimates are sharpened by using many hydrocarbons with different reactivities and accounting for their variability and measurement uncertainty. A novel technique is used to construct prior background distributions of many species, described by variation of a single parameter . This exploits the high correlation of species, related by the first principal component of many NMHC samples. The Bayesian method obtains posterior estimates of , K and following each air mass. Median values are typically between 0.5 and 2.0 × 106 molecules cm−3, but are elevated to between 2.5 and 3.5 × 106 molecules cm−3, in low-level pollution. A comparison of estimates from absolute NMHC concentrations and NMHC ratios assuming zero background (the “photochemical clock” method) shows similar distributions but reveals systematic high bias in the estimates from ratios. Estimates of K are ∼0.1 day−1 but show more sensitivity to the prior distribution assumed.

Statistical properties of cloud lifecycles in cloud-resolving models

A new technique is described for the analysis of cloud-resolving model simulations, which allows one to investigate the statistics of the lifecycles of cumulus clouds. Clouds are tracked from timestep-to-timestep within the model run. This allows for a very simple method of tracking, but one which is both comprehensive and robust. An approach for handling cloud splits and mergers is described which allows clouds with simple and complicated time histories to be compared within a single framework. This is found to be important for the analysis of an idealized simulation of radiative-convective equilibrium, in which the moist, buoyant, updrafts (i.e., the convective cores) were tracked. Around half of all such cores were subject to splits and mergers during their lifecycles. For cores without any such events, the average lifetime is 30min, but events can lengthen the typical lifetime considerably.

Can knowledge of the state of the stratosphere be used to improve statistical forecasts of the troposphere?

Recent analysis of the Arctic Oscillation (AO) in the stratosphere and troposphere has suggested that predictability of the state of the tropospheric AO may be obtained from the state of the stratospheric AO. However, much of this research has been of a purely qualitative nature. We present a more thorough statistical analysis of a long AO amplitude dataset which seeks to establish the magnitude of such a link. A relationship between the AO in the lower stratosphere and on the 1000 hPa surface on a 10-45 day time-scale is revealed. The relationship accounts for 5% of the variance of the 1000 hPa time series at its peak value and is significant at the 5% level. Over a similar time-scale the 1000 hPa time series accounts for 1% of itself and is not significant at the 5% level. Further investigation of the relationship reveals that it is only present during the winter season and in particular during February and March. It is also demonstrated that using stratospheric AO amplitude data as a predictor in a simple statistical model results in a gain of skill of 5% over a troposphere-only statistical model. This gain in skill is not repeated if an unrelated time series is included as a predictor in the model. Copyright © 2003 Royal Meteorological Society

The potential to narrow uncertainty in regional climate predictions

Faced by the realities of a changing climate, decision makers in a wide variety of organisations are increasingly seeking quantitative predictions of regional and local climate. An important issue for these decision makers, and for organisations that fund climate research, is what is the potential for climate science to deliver improvements - especially reductions in uncertainty - in such predictions? Uncertainty in climate predictions arises from three distinct sources: internal variability, model uncertainty and scenario uncertainty. Using data from a suite of climate models we separate and quantify these sources. For predictions of changes in surface air temperature on decadal timescales and regional spatial scales, we show that uncertainty for the next few decades is dominated by sources (model uncertainty and internal variability) that are potentially reducible through progress in climate science. Furthermore, we find that model uncertainty is of greater importance than internal variability. Our findings have implications for managing adaptation to a changing climate. Because the costs of adaptation are very large, and greater uncertainty about future climate is likely to be associated with more expensive adaptation, reducing uncertainty in climate predictions is potentially of enormous economic value. We highlight the need for much more work to compare: a) the cost of various degrees of adaptation, given current levels of uncertainty; and b) the cost of new investments in climate science to reduce current levels of uncertainty. Our study also highlights the importance of targeting climate science investments on the most promising opportunities to reduce prediction uncertainty.

The potential to narrow uncertainty in projections of regional precipitation change

We separate and quantify the sources of uncertainty in projections of regional (*2,500 km) precipitation changes for the twenty-first century using the CMIP3 multi-model ensemble, allowing a direct comparison with a similar analysis for regional temperature changes. For decadal means of seasonal mean precipitation, internal variability is the dominant uncertainty for predictions of the first decade everywhere, and for many regions until the third decade ahead. Model uncertainty is generally the dominant source of uncertainty for longer lead times. Scenario uncertainty is found to be small or negligible for all regions and lead times, apart from close to the poles at the end of the century. For the global mean, model uncertainty dominates at all lead times. The signal-to-noise ratio (S/N) of the precipitation projections is highest at the poles but less than 1 almost everywhere else, and is far lower than for temperature projections. In particular, the tropics have the highest S/N for temperature, but the lowest for precipitation. We also estimate a ‘potential S/N’ by assuming that model uncertainty could be reduced to zero, and show that, for regional precipitation, the gains in S/N are fairly modest, especially for predictions of the next few decades. This finding suggests that adaptation decisions will need to be made in the context of high uncertainty concerning regional changes in precipitation. The potential to narrow uncertainty in regional temperature projections is far greater. These conclusions on S/N are for the current generation of models; the real signal may be larger or smaller than the CMIP3 multi-model mean. Also note that the S/N for extreme precipitation, which is more relevant for many climate impacts, may be larger than for the seasonal mean precipitation considered here.

A model study of oceanic mechanisms affecting Equatorial Pacific sea surface temperature during the 1997-98 El Niño

In this study, the processes affecting sea surface temperature variability over the 1992–98 period, encompassing the very strong 1997–98 El Niño event, are analyzed. A tropical Pacific Ocean general circulation model, forced by a combination of weekly ERS1–2 and TAO wind stresses, and climatological heat and freshwater fluxes, is first validated against observations. The model reproduces the main features of the tropical Pacific mean state, despite a weaker than observed thermal stratification, a 0.1 m s−1 too strong (weak) South Equatorial Current (North Equatorial Countercurrent), and a slight underestimate of the Equatorial Undercurrent. Good agreement is found between the model dynamic height and TOPEX/Poseidon sea level variability, with correlation/rms differences of 0.80/4.7 cm on average in the 10°N–10°S band. The model sea surface temperature variability is a bit weak, but reproduces the main features of interannual variability during the 1992–98 period. The model compares well with the TAO current variability at the equator, with correlation/rms differences of 0.81/0.23 m s−1 for surface currents. The model therefore reproduces well the observed interannual variability, with wind stress as the only interannually varying forcing. This good agreement with observations provides confidence in the comprehensive three-dimensional circulation and thermal structure of the model. A close examination of mixed layer heat balance is thus undertaken, contrasting the mean seasonal cycle of the 1993–96 period and the 1997–98 El Niño. In the eastern Pacific, cooling by exchanges with the subsurface (vertical advection, mixing, and entrainment), the atmospheric forcing, and the eddies (mainly the tropical instability waves) are the three main contributors to the heat budget. In the central–western Pacific, the zonal advection by low-frequency currents becomes the main contributor. Westerly wind bursts (in December 1996 and March and June 1997) were found to play a decisive role in the onset of the 1997–98 El Niño. They contributed to the early warming in the eastern Pacific because the downwelling Kelvin waves that they excited diminished subsurface cooling there. But it is mainly through eastward advection of the warm pool that they generated temperature anomalies in the central Pacific. The end of El Niño can be linked to the large-scale easterly anomalies that developed in the western Pacific and spread eastward, from the end of 1997 onward. In the far-western Pacific, because of the shallower than normal thermocline, these easterlies cooled the SST by vertical processes. In the central Pacific, easterlies pushed the warm pool back to the west. In the east, they led to a shallower thermocline, which ultimately allowed subsurface cooling to resume and to quickly cool the surface layer.

Climate change and heat-related mortality in six cities Part 1: model construction and validation

Heat waves are expected to increase in frequency and magnitude with climate change. The first part of a study to produce projections of the effect of future climate change on heat-related mortality is presented. Separate city-specific empirical statistical models that quantify significant relationships between summer daily maximum temperature (T max) and daily heat-related deaths are constructed from historical data for six cities: Boston, Budapest, Dallas, Lisbon, London, and Sydney. ‘Threshold temperatures’ above which heat-related deaths begin to occur are identified. The results demonstrate significantly lower thresholds in ‘cooler’ cities exhibiting lower mean summer temperatures than in ‘warmer’ cities exhibiting higher mean summer temperatures. Analysis of individual ‘heat waves’ illustrates that a greater proportion of mortality is due to mortality displacement in cities with less sensitive temperature–mortality relationships than in those with more sensitive relationships, and that mortality displacement is no longer a feature more than 12 days after the end of the heat wave. Validation techniques through residual and correlation analyses of modelled and observed values and comparisons with other studies indicate that the observed temperature–mortality relationships are represented well by each of the models. The models can therefore be used with confidence to examine future heat-related deaths under various climate change scenarios for the respective cities (presented in Part 2).

Flux comparison of Eulerian and Lagrangian estimates of Agulhas leakage: A case study using a numerical model

Estimating the magnitude of Agulhas leakage, the volume flux of water from the Indian to the Atlantic Ocean, is difficult because of the presence of other circulation systems in the Agulhas region. Indian Ocean water in the Atlantic Ocean is vigorously mixed and diluted in the Cape Basin. Eulerian integration methods, where the velocity field perpendicular to a section is integrated to yield a flux, have to be calibrated so that only the flux by Agulhas leakage is sampled. Two Eulerian methods for estimating the magnitude of Agulhas leakage are tested within a high-resolution two-way nested model with the goal to devise a mooring-based measurement strategy. At the GoodHope line, a section halfway through the Cape Basin, the integrated velocity perpendicular to that line is compared to the magnitude of Agulhas leakage as determined from the transport carried by numerical Lagrangian floats. In the first method, integration is limited to the flux of water warmer and more saline than specific threshold values. These threshold values are determined by maximizing the correlation with the float-determined time series. By using the threshold values, approximately half of the leakage can directly be measured. The total amount of Agulhas leakage can be estimated using a linear regression, within a 90% confidence band of 12 Sv. In the second method, a subregion of the GoodHope line is sought so that integration over that subregion yields an Eulerian flux as close to the float-determined leakage as possible. It appears that when integration is limited within the model to the upper 300 m of the water column within 900 km of the African coast the time series have the smallest root-mean-square difference. This method yields a root-mean-square error of only 5.2 Sv but the 90% confidence band of the estimate is 20 Sv. It is concluded that the optimum thermohaline threshold method leads to more accurate estimates even though the directly measured transport is a factor of two lower than the actual magnitude of Agulhas leakage in this model.

C---H local modes in cyclobutene, part I: FTIR studies, 600 to 10000 cm–1

The vibrational structure of C---H stretching states in gas-phase cyclobutene was studied using FTIR spectroscopy in the range 700–9000 cm−1. The structure was modelled using two effective vibrational Hamiltonians, one for each type of C---H bond present, consisting of local mode basis functions subject to coupling with symmetrically equivalent bonds and to Fermi resonances with suitable low frequency vibrations. Best-fit model parameters were determined using least-squares routines and the model predictions are compared to the observed band positions and intensities. Some discussion is given of the relevance of the observed couplings to intramolecular vibrational redistribution (IVR) which results in the observation of statistical behaviour in cyclobutene isomerization induced by excitation of C---H stretching overtones in the visible region.

The potential to narrow uncertainty in projections of stratospheric ozone over the 21st century

Future stratospheric ozone concentrations will be determined both by changes in the concentration of ozone depleting substances (ODSs) and by changes in stratospheric and tropospheric climate, including those caused by changes in anthropogenic greenhouse gases (GHGs). Since future economic development pathways and resultant emissions of GHGs are uncertain, anthropogenic climate change could be a significant source of uncertainty for future projections of stratospheric ozone. In this pilot study, using an "ensemble of opportunity" of chemistry-climate model (CCM) simulations, the contribution of scenario uncertainty from different plausible emissions pathways for ODSs and GHGs to future ozone projections is quantified relative to the contribution from model uncertainty and internal variability of the chemistry-climate system. For both the global, annual mean ozone concentration and for ozone in specific geographical regions, differences between CCMs are the dominant source of uncertainty for the first two-thirds of the 21st century, up-to and after the time when ozone concentrations return to 1980 values. In the last third of the 21st century, dependent upon the set of greenhouse gas scenarios used, scenario uncertainty can be the dominant contributor. This result suggests that investment in chemistry-climate modelling is likely to continue to refine projections of stratospheric ozone and estimates of the return of stratospheric ozone concentrations to pre-1980 levels.

Statistical downscaling of river flows

An extensive statistical ‘downscaling’ study is done to relate large-scale climate information from a general circulation model (GCM) to local-scale river flows in SW France for 51 gauging stations ranging from nival (snow-dominated) to pluvial (rainfall-dominated) river-systems. This study helps to select the appropriate statistical method at a given spatial and temporal scale to downscale hydrology for future climate change impact assessment of hydrological resources. The four proposed statistical downscaling models use large-scale predictors (derived from climate model outputs or reanalysis data) that characterize precipitation and evaporation processes in the hydrological cycle to estimate summary flow statistics. The four statistical models used are generalized linear (GLM) and additive (GAM) models, aggregated boosted trees (ABT) and multi-layer perceptron neural networks (ANN). These four models were each applied at two different spatial scales, namely at that of a single flow-gauging station (local downscaling) and that of a group of flow-gauging stations having the same hydrological behaviour (regional downscaling). For each statistical model and each spatial resolution, three temporal resolutions were considered, namely the daily mean flows, the summary statistics of fortnightly flows and a daily ‘integrated approach’. The results show that flow sensitivity to atmospheric factors is significantly different between nival and pluvial hydrological systems which are mainly influenced, respectively, by shortwave solar radiations and atmospheric temperature. The non-linear models (i.e. GAM, ABT and ANN) performed better than the linear GLM when simulating fortnightly flow percentiles. The aggregated boosted trees method showed higher and less variable R2 values to downscale the hydrological variability in both nival and pluvial regimes. Based on GCM cnrm-cm3 and scenarios A2 and A1B, future relative changes of fortnightly median flows were projected based on the regional downscaling approach. The results suggest a global decrease of flow in both pluvial and nival regimes, especially in spring, summer and autumn, whatever the considered scenario. The discussion considers the performance of each statistical method for downscaling flow at different spatial and temporal scales as well as the relationship between atmospheric processes and flow variability.

A hierarchical Bayesian model for predicting the functional consequences of amino-acid polymorphisms

Genetic polymorphisms in deoxyribonucleic acid coding regions may have a phenotypic effect on the carrier, e.g. by influencing susceptibility to disease. Detection of deleterious mutations via association studies is hampered by the large number of candidate sites; therefore methods are needed to narrow down the search to the most promising sites. For this, a possible approach is to use structural and sequence-based information of the encoded protein to predict whether a mutation at a particular site is likely to disrupt the functionality of the protein itself. We propose a hierarchical Bayesian multivariate adaptive regression spline (BMARS) model for supervised learning in this context and assess its predictive performance by using data from mutagenesis experiments on lac repressor and lysozyme proteins. In these experiments, about 12 amino-acid substitutions were performed at each native amino-acid position and the effect on protein functionality was assessed. The training data thus consist of repeated observations at each position, which the hierarchical framework is needed to account for. The model is trained on the lac repressor data and tested on the lysozyme mutations and vice versa. In particular, we show that the hierarchical BMARS model, by allowing for the clustered nature of the data, yields lower out-of-sample misclassification rates compared with both a BMARS and a frequen-tist MARS model, a support vector machine classifier and an optimally pruned classification tree.