Kernel classifier construction using orthogonal forward selection and boosting with Fisher ratio class separability measure

A greedy technique is proposed to construct parsimonious kernel classifiers using the orthogonal forward selection method and boosting based on Fisher ratio for class separability measure. Unlike most kernel classification methods, which restrict kernel means to the training input data and use a fixed common variance for all the kernel terms, the proposed technique can tune both the mean vector and diagonal covariance matrix of individual kernel by incrementally maximizing Fisher ratio for class separability measure. An efficient weighted optimization method is developed based on boosting to append kernels one by one in an orthogonal forward selection procedure. Experimental results obtained using this construction technique demonstrate that it offers a viable alternative to the existing state-of-the-art kernel modeling methods for constructing sparse Gaussian radial basis function network classifiers. that generalize well.

Fast kernel classifier construction using orthogonal forward selection to minimise leave-one-out misclassification rate

We propose a simple yet computationally efficient construction algorithm for two-class kernel classifiers. In order to optimise classifier's generalisation capability, an orthogonal forward selection procedure is used to select kernels one by one by minimising the leave-one-out (LOO) misclassification rate directly. It is shown that the computation of the LOO misclassification rate is very efficient owing to orthogonalisation. Examples are used to demonstrate that the proposed algorithm is a viable alternative to construct sparse two-class kernel classifiers in terms of performance and computational efficiency.

A kernel-based two-class classifier for imbalanced data sets

Many kernel classifier construction algorithms adopt classification accuracy as performance metrics in model evaluation. Moreover, equal weighting is often applied to each data sample in parameter estimation. These modeling practices often become problematic if the data sets are imbalanced. We present a kernel classifier construction algorithm using orthogonal forward selection (OFS) in order to optimize the model generalization for imbalanced two-class data sets. This kernel classifier identification algorithm is based on a new regularized orthogonal weighted least squares (ROWLS) estimator and the model selection criterion of maximal leave-one-out area under curve (LOO-AUC) of the receiver operating characteristics (ROCs). It is shown that, owing to the orthogonalization procedure, the LOO-AUC can be calculated via an analytic formula based on the new regularized orthogonal weighted least squares parameter estimator, without actually splitting the estimation data set. The proposed algorithm can achieve minimal computational expense via a set of forward recursive updating formula in searching model terms with maximal incremental LOO-AUC value. Numerical examples are used to demonstrate the efficacy of the algorithm.

Probability density estimation with tunable kernels using orthogonal forward regression

A generalized or tunable-kernel model is proposed for probability density function estimation based on an orthogonal forward regression procedure. Each stage of the density estimation process determines a tunable kernel, namely, its center vector and diagonal covariance matrix, by minimizing a leave-one-out test criterion. The kernel mixing weights of the constructed sparse density estimate are finally updated using the multiplicative nonnegative quadratic programming algorithm to ensure the nonnegative and unity constraints, and this weight-updating process additionally has the desired ability to further reduce the model size. The proposed tunable-kernel model has advantages, in terms of model generalization capability and model sparsity, over the standard fixed-kernel model that restricts kernel centers to the training data points and employs a single common kernel variance for every kernel. On the other hand, it does not optimize all the model parameters together and thus avoids the problems of high-dimensional ill-conditioned nonlinear optimization associated with the conventional finite mixture model. Several examples are included to demonstrate the ability of the proposed novel tunable-kernel model to effectively construct a very compact density estimate accurately.

Can shells of freshwater mussels (Unionidae) be used to estimate low summer discharge of rivers and associated droughts?

This paper examines if shell oxygen isotope ratios (d18Oar) of Unio sp. can be used as a proxy of past discharge of the river Meuse. The proxy was developed from a modern dataset for the reference time interval 1997–2007, which showed a logarithmic relationship between discharge and measured water oxygen isotope ratios(d18Ow). To test this relationship for past time intervals,d18Oar values were measured in the aragonite of the growth increments of four Unio sp. shells; two from a relatively wet period and two from a very dry time interval (1910–1918 and 1969–1977, respectively). Shell d18Oar records were converted into d18Ow values using existing water temperature records. Summer d18Ow values, reconstructed from d18Oar of 1910–1918, showed a similar range as the summer d18Ow values for the reference time interval 1997–2007, whilst summer reconstructed d18Ow values for the time interval 1969–1977 were anomalously high. These high d18Ow values suggest that the river Meuse experienced severe summer droughts during the latter time interval. d18Ow values were then applied to calculate discharge values. It was attempted to estimate discharge from the reconstructed d18Ow values using the logarithmic relationship between d18Ow and discharge. A comparison of the calculated summer discharge results with observed discharge data showed that Meuse low-discharge events below a threshold value of 6 m3/s can be detected in the reconstructed d18Ow records, but true quantification remains problematic.

Using zero-norm constraint for sparse probability density function estimation

A new sparse kernel probability density function (pdf) estimator based on zero-norm constraint is constructed using the classical Parzen window (PW) estimate as the target function. The so-called zero-norm of the parameters is used in order to achieve enhanced model sparsity, and it is suggested to minimize an approximate function of the zero-norm. It is shown that under certain condition, the kernel weights of the proposed pdf estimator based on the zero-norm approximation can be updated using the multiplicative nonnegative quadratic programming algorithm. Numerical examples are employed to demonstrate the efficacy of the proposed approach.

Combining satellite data and models to estimate cloud radiative effect at the surface and in the atmosphere

Satellite measurements and numerical forecast model reanalysis data are used to compute an updated estimate of the cloud radiative effect on the global multi-annual mean radiative energy budget of the atmosphere and surface. The cloud radiative cooling effect through reflection of shortwave radiation dominates over the longwave heating effect, resulting in a net cooling of the climate system of –21 Wm-2. The shortwave radiative effect of cloud is primarily manifest as a reduction in the solar radiation absorbed at the surface of -53 Wm-2. Clouds impact longwave radiation by heating the moist tropical atmosphere (up to around 40 Wm-2 for global annual means) while enhancing the radiative cooling of the atmosphere over other regions, in particular higher latitudes and sub-tropical marine stratocumulus regimes. While clouds act to cool the climate system during the daytime, the cloud greenhouse effect heats the climate system at night. The influence of cloud radiative effect on determining cloud feedbacks and changes in the water cycle are discussed.

Multimodel estimate of the global terrestrial water balance: setup and first results

Six land surface models and five global hydrological models participate in a model intercomparison project (WaterMIP), which for the first time compares simulation results of these different classes of models in a consistent way. In this paper the simulation setup is described and aspects of the multi-model global terrestrial water balance are presented. All models were run at 0.5 degree spatial resolution for the global land areas for a 15-year period (1985-1999) using a newly-developed global meteorological dataset. Simulated global terrestrial evapotranspiration, excluding Greenland and Antarctica, ranges from 415 to 586 mm year-1 (60,000 to 85,000 km3 year-1) and simulated runoff ranges from 290 to 457 mm year-1 (42,000 to 66,000 km3 year-1). Both the mean and median runoff fractions for the land surface models are lower than those of the global hydrological models, although the range is wider. Significant simulation differences between land surface and global hydrological models are found to be caused by the snow scheme employed. The physically-based energy balance approach used by land surface models generally results in lower snow water equivalent values than the conceptual degree-day approach used by global hydrological models. Some differences in simulated runoff and evapotranspiration are explained by model parameterizations, although the processes included and parameterizations used are not distinct to either land surface models or global hydrological models. The results show that differences between model are major sources of uncertainty. Climate change impact studies thus need to use not only multiple climate models, but also some other measure of uncertainty, (e.g. multiple impact models).

Attribution of climate change: a methodology to estimate the potential contribution to increases in potato yield in Scotland since 1960

Maincrop potato yields in Scotland have increased by 3035 similar to t similar to ha-1 since 1960 as a result of many changes, but has changing climate contributed anything to this? The purpose of this work was to answer this question. Daily weather data for the period 19602006 were analysed for five locations covering the zones of potato growing on the east coast of Scotland (between 55.213 and 57.646 similar to N) to determine trends in temperature, rainfall and solar radiation. A physiologically based potato yield model was validated using data obtained from a long-term field trial in eastern Scotland and then employed to simulate crop development and potential yield at each of the five sites. Over the 47 similar to years, there were significant increases in annual air and 30 similar to cm soil temperatures (0.27 and 0.30 similar to K similar to decade-1, respectively), but no significant changes in annual precipitation or in the timing of the last frost in spring and the first frost of autumn. There was no evidence of any north to south gradient of warming. Simulated emergence and canopy closure became earlier at all five sites over the period with the advance being greater in the north (3.7 and 3.6 similar to days similar to decade-1, respectively) than the south (0.5 and 0.8 similar to days similar to decade-1, respectively). Potential yield increased with time, generally reflecting the increased duration of the green canopy, at average rates of 2.8 similar to t similar to ha-1 decade-1 for chitted seed (sprouted prior to planting) and 2.5 similar to t similar to ha-1 decade-1 for unchitted seed. The measured warming could contribute potential yield increases of up to 13.2 similar to t similar to ha-1 for chitted potato (range 7.119.3 similar to t similar to ha-1) and 11.5 similar to t similar to ha-1 for unchitted potato (range 7.115.5 similar to t similar to ha-1) equivalent to 3439% of the increased potential yield over the period or 2326% of the increase in actual measured yields.

Using meteorological wind data to estimate turbine generation output: a sensitivity analysis

Various studies investigating the future impacts of integrating high levels of renewable energy make use of historical meteorological (met) station data to produce estimates of future generation. Hourly means of 10m horizontal wind are extrapolated to a standard turbine hub height using the wind profile power or log law and used to simulate the hypothetical power output of a turbine at that location; repeating this procedure using many viable locations can produce a picture of future electricity generation. However, the estimate of hub height wind speed is dependent on the choice of the wind shear exponent a or the roughness length z0, and requires a number of simplifying assumptions. This paper investigates the sensitivity of this estimation on generation output using a case study of a met station in West Freugh, Scotland. The results show that the choice of wind shear exponent is a particularly sensitive parameter which can lead to significant variation of estimated hub height wind speed and hence estimated future generation potential of a region.

A study into the accuracy of using meteorological wind data to estimate turbine generation output

Meteorological (met) station data is used as the basis for a number of influential studies into the impacts of the variability of renewable resources. Real turbine output data is not often easy to acquire, whereas meteorological wind data, supplied at a standardised height of 10 m, is widely available. This data can be extrapolated to a standard turbine height using the wind profile power law and used to simulate the hypothetical power output of a turbine. Utilising a number of met sites in such a manner can develop a model of future wind generation output. However, the accuracy of this extrapolation is strongly dependent on the choice of the wind shear exponent alpha. This paper investigates the accuracy of the simulated generation output compared to reality using a wind farm in North Rhins, Scotland and a nearby met station in West Freugh. The results show that while a single annual average value for alpha may be selected to accurately represent the long term energy generation from a simulated wind farm, there are significant differences between simulation and reality on an hourly power generation basis, with implications for understanding the impact of variability of renewables on short timescales, particularly system balancing and the way that conventional generation may be asked to respond to a high level of variable renewable generation on the grid in the future.

Updated estimate of aerosol direct radiative forcing from satellite observations and comparison against the Hadley Centre climate model

The fourth assessment report of the Intergovernmental Panel on Climate Change (IPCC) includes a comparison of observation-based and modeling-based estimates of the aerosol direct radiative forcing. In this comparison, satellite-based studies suggest a more negative aerosol direct radiative forcing than modeling studies. A previous satellite-based study, part of the IPCC comparison, uses aerosol optical depths and accumulation-mode fractions retrieved by the Moderate Resolution Imaging Spectroradiometer (MODIS) at collection 4. The latest version of MODIS products, named collection 5, improves aerosol retrievals. Using these products, the direct forcing in the shortwave spectrum defined with respect to present-day natural aerosols is now estimated at −1.30 and −0.65 Wm−2 on a global clear-sky and all-sky average, respectively, for 2002. These values are still significantly more negative than the numbers reported by modeling studies. By accounting for differences between present-day natural and preindustrial aerosol concentrations, sampling biases, and investigating the impact of differences in the zonal distribution of anthropogenic aerosols, good agreement is reached between the direct forcing derived from MODIS and the Hadley Centre climate model HadGEM2-A over clear-sky oceans. Results also suggest that satellite estimates of anthropogenic aerosol optical depth over land should be coupled with a robust validation strategy in order to refine the observation-based estimate of aerosol direct radiative forcing. In addition, the complex problem of deriving the aerosol direct radiative forcing when aerosols are located above cloud still needs to be addressed.