Ignoring data may be the only way to learn efficiently

Abstract

In designing learning algorithms it seems quite reasonable to construct them in a way such that all data the algorithm already has obtained are correctly and completely reflected in the hypothesis the algorithm outputs on these data. However, this approach may totally fail, i.e. it may lead to the unsolvability of the learning problem, or it may exclude any efficient solution of it. In particular, we present a natural learning problem and prove that it can be solved in polynomial time if and only if the algorithm is allowed to ignore data.     

A model for retrieving total sea ice concentration from a spaceborne dual-polarized passive microwave instrument operating near 90 GHz

Abstract

An algorithm has been developed for estimating total ice concentration from spaceborne high-frequency passive microwave instrumentation. The algorithm is intended for use with the coming Special Sensor Microwave/Imager (SSM/I) data giving a spatial resolution of 12 km. It is based on radiation physics and detailed millimetre wave surface signature measurements and can therefore be applied to other similar data. However, due to large effects on the signals caused by time varying atmospheric conditions and radiation properties of the ice, the algorithm is made self-adjusting. The atmospheric effects are implicitly treated as a smooth function of the ice concentration with tie points over open ocean and 100 per cent ice for each orbit. This means that the main errors are due to patches of heavy clouds and ice floes with atypical radiation properties. An error analysis indicates possible errors of the order of 5 percent for concentrations representative for the Arctic Basin, increasing with decreasing concentration.     

Atmospheric optical depth effects on angular anisotropy of plant canopy reflectance

Abstract

A field experiment was conducted to determine whether changes in atmospheric aerosol optical depth would effect changes in bi-directional reflectance distributions of vegetation canopies. Measurements were made of the directionally reflected radiance distributions of two pasture grass canopies (same species, different growth forms) and one soya bean plant canopy under different sky irradiance distributions, which resulted from a variation in aerosol optical depth. The reflected radiance data were analysed in the solar principal plane in two narrow spectral bands, one visible (662 nm) and one infrared (826 nm). The observed changes in reflectance for both wavelengths from irradiance distribution variation is interpreted to be due largely to changes in the percentage of shadowed area viewed by the sensor for the incomplete canopies (pasture grass). For the complete coverage vegetation canopy (soya bean) studied, the effects of specular reflection and the increased diffuse irradiance penetration into the canopy are concluded to be primary physical mechanisms responsible for reflectance changes. Observed reflectivities were found to be lower on a hazy day (higher optical depth with a greater diffuse fraction) than on a clear day, with solar zenith angles at about 58^° on both days, for full-coverage soya bean canopies. The reduced reflectance most likely results from a diminished specular reflection and a greater diffuse radiation penetration into the canopy, which effects an increased energy absorption at large solar zenith angles. The opposite was true for fractional coverage grass canopies at solar zenith angles of about 56^° since the shadowing was less on the hazy day and, therefore, the soil/litter background was more fully illuminated. In the near-infrared waveband the changes in reflectance are much less than in the visible and, therefore, normalized difference vegetation index values differ substantially under clear and hazy sky conditions for the same vegetation canopy conditions. Thus, the influence of atmospheric optical depth must be considered for accurate remote sensing and in situ data interpretation.     

Mesoscale variability of sea surface temperature in the North Atlantic

Abstract

Analyses of mesoscale horizontal distributions of temperature were performed for an area of the North Atlantic using data from the NOAA-7 and NOAA-6 Advanced Very High Resolution Radiometer (AVHRR). The zonal and meridianal variance spectra have.slopes between —1.4 and —2.5 with a clear maximum at -2.0. This is also true for the direction-dependent structure functions. The isotropic part of the variance spectra has a mean slope of —2.2±0.17 at scales of 10— 100km; this lies between the slopes of -1 and —3 predicted by the theories of two-dimensional and geostrophic turbulence. A comparison between measurements and theories is difficult because of the insufficient applicability of these theories to boundary layers. Moreover, in some cases there are significant maxima in the variance spectra at scales between 50 km and 250 km.     

High resolution observations of Weddell Sea surface currents using ERS-l SAR sea-ice motion vectors

Two areas of the Weddell Sea, one in the south and one in the west, were chosen for a preliminary investigation of sea-ice motion tracking from ERSt Synthetic Aperture Radar (SAR) images during the Austral summer. Only a small number of images were processed, so a manual tracking method was used. In the 3–day period between SAR images the atmosphere warmed near the surface, which led to significant changes in radar backscatter from, and thus in contrast between, ice floes and the areas between them. It was therefore not always possible to track features from one image to the next. The tracked features were clearly identified in images which were sub-sampled at onesixteenth of the full resolution available. In the southern Weddell Sea images, many large floes were present which allowed a quite detailed pattern of the surface water circulation to be mapped as the ice motion was predominantly forced by the ocean currents during a period of low surface wind speeds. The observed circulation pattern agreed well with previous observations from hydrographic surveys in this area north of the Filchner Ice Shelf. In the western Weddell Sea images good tracers were hard to find, but it was still possible to detect the edge of the western boundary current of the Weddell Gyre. Continuous monitoring of sea-ice motion in these two areas using SAR imagery could be a useful means of detecting changes in surface water flow which may be linked to the rate of formation of Antarctic bottom water.     

Special colour enhancement for three channels having similar radiances

A special enhancement algorithm is derived to colour separate ground-cover classes whose recorded wavelength distributions are similar but have significantly differing intensities. The special enhancement function is derived from considering the spectrum locus of a CIE 1931 (x,y) chromaticity diagram. By analogy with the relationship between the chromaticity co-ordinates and the channel radiance levels a function is derived that stretches the radiances to maximize the colour differences between such ground-cover classes. This Sinsusoidal Squeeze algorithm was applied to enhance the colour differentiation of Antarctic ice types as recorded by LANDSAT. A comparison between this special enhancement and the conventional colour composite results for Antarctic ice pack is presented. Field checks conducted during the 1980-81 Antarctic summer field season confirm our ability to delineate up to seven different types of ice by this technique. Additional applications of this special algorithm are believed to be forest class differentiation and bathymetric feature delineation.     

Fractoemission from Failure of Metal-GIass Interfaces

Electron and photon emissions accompanying the failure of a metal/glass interface were studied. Experimental evidence is given in support of a previously presented model involving microdischarges accompanying fracture which result from charge separation occurring between the metal and glass surfaces. The role of contact charging at the metal/glass interface in establishing this charge separation is discussed.     

Microbiology of brines used to mature feta cheese

The microfloras of brines from two commercial factories manufacturing traditional feta cheese were examined over a period of 60 days, which is the normal maturation time for Greek feta. Non-starter lactic acid bacteria were the dominant group, and Lactobacillus paracasei ssp. paracasei and Lactobacillus plantarum were the principal species identified. A number of strains of Debaryomyces hansenii were isolated from the brines, and both the yeasts and the lactobacilli possessed enzymes capable of degrading fat or protein. It is suggested that these microfloras play an important role in the maturation of feta cheese. No recognized pathogens were isolated from the brines of' either factory.