Automated classification of stellar spectra — II. Two-dimensional classification with neural networks and principal components analysis

We investigate the application of neural networks to the automation of MK spectral classification. The data set for this project consists of a set of over 5000 optical (3800–5200 Å) spectra obtained from objective prism plates from the Michigan Spectral Survey. These spectra, along with their two-dimensional MK classifications listed in the Michigan Henry Draper Catalogue, were used to develop supervised neural network classifiers. We show that neural networks can give accurate spectral type classifications (σ₆₈= 0.82 subtypes, σ_rms= 1.09 subtypes) across the full range of spectral types present in the data set (B2–M7). We show also that the networks yield correct luminosity classes for over 95 per cent of both dwarfs and giants with a high degree of confidence.   Stellar spectra generally contain a large amount of redundant information. We investigate the application of principal components analysis (PCA) to the optimal compression of spectra. We show that PCA can compress the spectra by a factor of over 30 while retaining essentially all of the useful information in the data set. Furthermore, it is shown that this compression optimally removes noise and can be used to identify unusual spectra.   This paper is a continuation of the work carried out by von Hippel et al. (Paper I).     

Iapetus surface variability revealed from statistical clustering of a VIMS mosaic: The distribution of CO2

We present a detailed study of an Iapetus mosaic of VIMS data with high spatial resolution (0.5 × 0.5° or ∼6.4 km/pixel). The spectra were taken in August 2007 and provide the highest VIMS spatial resolution data for this object during Cassini’s primary mission. We analyze this set of data using a statistical clustering approach to reduce the analysis of a large number of data (∼10⁴ spectra from 0.35 to 5.10 μm) to the study of seven representative groups accounting for 99.6% of the surface covered by the original sample. We analyze the spectral absorption bands in the spectra of the different clusters indicative of different composition over the observed surface. We find coherence between the distribution of the clusters and the geographical features on the surface. We give special attention to the study of the water ice and CO₂ bands. We find that CO₂ is widespread over the entire surface being studied, including the bright and dark areas on Iapetus’ surface, and is probably trapped at the molecular level with other materials. The strength of the CO₂ band in the areas where both, H₂O- and carbon-bearing materials exist, gives support to the hypothesis that this volatile is formed on the surface of Iapetus as a product of irradiation of these two components. Finally, we also compare the Iapetus CO₂ with that on other satellites confirming, that there are evident differences on the center, depth and width of the band on Iapetus and Phoebe, where CO₂ has been suggested to be endogenous.     

A preliminary methods comparison for measurement of dissolved organic nitrogen in seawater

Routine determination of dissolved organic nitrogen (DON) is performed in numerous laboratories around the world using one of three families of methods: UV oxidation (UV), persulfate oxidation (PO), or high temperature combustion (HTC). Essentially all routine methods measure total dissolved nitrogen (TDN) and calculate DON by subtracting the dissolved inorganic nitrogen (DIN). While there is currently no strong suggestion that any of these methods is inadequate, there are continuing suspicions of slight inaccuracy by UV methods.This is a report of a broad community methods comparison where 29 sets (7 UV, 13 PO, and 9 HTC) of TDN analyses were performed on five samples with varying TDN and DIN concentrations. Analyses were done in a “blind” procedure with results sent to the first author. With editing out one set of extreme outliers (representing 5 out of 145 ampoules analyzed), the community comparability for analyzing the TDN samples was in the 8–28% range (coefficient of variation representing one standard deviation for the five individual samples by 28 analyses). When DIN concentrations were subtracted uniformly (single DIN value for each sample), the comparability was obviously worse (19–46% cv). This comparison represents a larger and more diverse set of analyses, but the overall comparability is only marginally better than that of the Seattle workshop of a decade ago. Grouping methods, little difference was seen other than inconclusive evidence that the UV methods gave TDN values for several of the samples higher than HTC methods. Since there was much scatter for each of the groups of methods and for all analyses when grouped, it is thought that more uniformity in procedures is probably needed. An important unplanned observation is that variability in DIN analyses (used in determining the final analyte in most UV and PO methods) is essentially as large as the variability in the TDN analyses.This exercise should not be viewed as a qualification exercise for the analysts, but should instead be considered a broad preliminary test of the comparison of the families of methods being used in various laboratories around the world. Based on many independent analyses here, none of the routinely used methods appears to be grossly inaccurate, thus, most routine TDN analyses being reported in the literature are apparently accurate. However, it is not reassuring that the ability of the international community to determine DON in deep oceanic waters continues to be poor. It is suggested that as an outgrowth of this paper, analysts using UV and PO methods experiment and look more carefully at the completeness of DIN conversion to the final analyte and also at the accuracy of their analysis of the final analyte. HTC methods appear to be relatively easy and convenient and have potential for routine adoption. Several of the authors of this paper are currently working together on an interlaboratory comparison on HTC methodology.     

The Near Infrared Spectrum of Miranda: Evidence of Crystalline Water Ice

A spectrum from 1.2 to 2.5 μm of Uranus' small satellite Miranda obtained in June 1999 reveals strong water-ice signatures. It confirms the existence of a 2.0-μm water feature previously detected on Miranda and shows a strong second broad 1.5-μm water-ice absorption feature. The spectra also reveal a weak absorption band at 1.65 μm that is indicative of crystalline water ice. Reflectance models which combine the new spectra with new photometry indicate that the spectra are characteristic of a mostly water-ice surface, with a large fraction of carbonaceous or silicate contaminates, and the possible presence of ammonia hydrate, as implied by an apparent weak feature near 2.2 μm. The possible presence of other volatiles is also investigated.     

Reviews

GEOMORPHOLOGY by R. J. Chorley, S. A. Schumm and D. E. Sugden.’ 19 × 25 cm, xxiii and 605 pages. Methuen: London 1984 (ISBN 0 416 32590 4) $A39.95 (cloth).

THEMES IN GEOMORPHOLOGY edited by A. F. Pitty. 13 × 21 cm, 280 pages. Croom Helm: London 1985 (ISBN 0 7099 2066 0) $A49.50 (cloth).

DESERTS AND ARID LANDS edited by F. El‐Baz. 16 × 25 cm, vi and 222 pages. Nijhoff: The Hague 1984 (ISBN 90 247 2850 9) $US46.00 (cloth).

COASTAL GEOMORPHOLOGY IN AUSTRALIA edited by B. G. Thorn. 16 × 24 cm, xv and 349 pages. Academic Press: Sydney 1984 (ISBN 0 12 687880 3) $A49.50 (cloth).

KWONGAN: Plant Life on the Sandplain edited by J. S. Pate and J. S. Beard. University of Western Australia Press: Nedlands 1984 (ISBN 0 85564 230 0) $A25.00 (limp).

THEMES IN BIOGEOGRAPHY edited by J. A. Taylor. 22 × 14 cm, xxvii and 404 pages. Croom Helm: London 1984 (ISBN 0 7099 2428 3) $A25.95 (limp).

ENVIRONMENTAL IMPACT ASSESSMENT: A Practical Guide by C. F. Porter. 22 × 15 cm, xiii and 269 pages. University of Queensland Press: St Lucia 1985 (ISBN 0 7022 1699 2) $A35.00 (cloth).

THE CLIMATIC SCENE edited by M. J. Tooley and G. M. Sheail. 16 × 24 cm, xx and 306 pages. George Allen and Unwin: London 1985 (ISBN 0 04 551089 X) $A59.95 (cloth).

VOLCANIC HAZARDS: A Sourcebook on the Effects of Eruptions by R. J. Blong. 24 × 15 cm, xvi and 424 pages. Academic Press: Sydney 1984 (ISBN 0 12 107180 4) $A68.00 (cloth).

THE ECONOMICS OF BUSHFIRES: The South Australian Experience edited by D. T. Healey, F. G. Jarrett and J. M. McKay. 22 × 15 cm, x and 152 pages. Oxford University Press: Melbourne 1985 (ISBN 0 19 554669 5) $A19.99 (cloth).     

A compositional interpretation of trans-neptunian objects taxonomies

Trans-neptunian objects (TNOs) are a population of small objects orbiting the Sun beyond Neptune. Because of their distance they are difficult to observe spectroscopically, but a large body of photometric observations is available and growing. TNOs are important tracers of the evolution of the outer Solar System and key when testing current dynamical evolution theories. Previous statistical studies of the colors of TNOs have yielded useful but limited results regarding the chemical history and evolution of these bodies.With the aim at obtaining compositional information on the small and distant TNOs we introduce a statistical cluster analysis (labelled albedo) based on colors and published albedos of TNOs. We compare it to a previous taxonomy, to illustrate the significance of including the albedo information when determining the composition of the objects. When the albedo contribution is removed from the data, the new taxonomy (now labelled classical) is in general agreement with the published ones, supporting the applicability of our approach. Making use of modeled reflectance spectra of a variety of plausible mixtures found on the surface of TNOs, we extract the average surface composition of each taxon, for both the classical and the albedo taxonomy, in a statistically consistent fashion.Differently from previous and classical, the albedo taxonomy establishes a direct link between the colors and albedos of the objects and their surface composition, allowing, for the first time, a quick assessment of the chemical history of TNOs. In fact, under closer examination the taxa show trends in composition that might be evolutionary in nature. If a simple ‘snow lines’ model is adopted, we can infer that albedo taxa relate the current objects’ locations to their original ones, prior to the migration of the outer planets. We regard the large population that characterizes the darkest classes spread at a variety of semi-major axis distances as one of the intriguing results of this work.     

Temperatures and thermophysical properties of the lunar outermost layer

Comparisons of calculated diurnal and eclipse temperatures of the lunar outermost layer are made with Earth-based infrared and millimeter data. The thermophysical model upon which the calculations are based incorporates variable physical properties. The thermal conductivity is a function of both density (depth) and temperature; the specific heat is a function of temperature; the density is a function of depth; and the dielectric constant and loss tangent are functions of density (depth). Laboratory measurements and Apollo sample results are incorporated in the property data. Calculational cases are based largely upon different density profiles. The model is consistent with the data, and the comparisons of theoretical and observational temperatures are very favorable. For such comparisons, further sophistication of the thermophysical model of the outermost layer is probably not justified.     

Saturn’s F ring grains: Aggregates made of crystalline water ice

We present models of the near-infrared (1-5 μm) spectra of Saturn’s F ring obtained by Cassini’s Visual and Infrared Mapping Spectrometer (VIMS) at ultra-high phase angles (177.4-178.5°). Modeling this spectrum constrains the size distribution, composition, and structure of F ring particles in the 0.1-100 μm size range. These spectra are very different from those obtained at lower phase angles; they lack the familiar 1.5 and 2 μm absorption bands, and the expected 3 μm water ice primary absorption appears as an unusually narrow dip at 2.87 μm. We have modeled these data using multiple approaches. First, we use a simple Mie scattering model to constrain the size distribution and composition of the particles. The Mie model allows us to understand the overall shapes of the spectra in terms of dominance by diffraction at these ultra-high phase angles, and also to demonstrate that the 2.87 μm dip is associated with the Christiansen frequency of water ice (where the real refractive index passes unity). Second, we use a combination of Mie scattering with Effective Medium Theory to probe the effect of porous (but structureless) particles on the overall shape of the spectrum and depth of the 2.87 μm band. Such simple models are not able to capture the shape of this absorption feature well. Finally, we model each particle as an aggregate of discrete monomers, using the Discrete Dipole Approximation (DDA) model, and find a better fit for the depth of the 2.87 μm feature. The DDA models imply a slightly different overall size distribution. We present a simple heuristic model which explains the differences between the Mie and DDA model results. We conclude that the F ring contains aggregate particles with a size distribution that is distinctly narrower than a typical power law, and that the particles are predominantly crystalline water ice.