An evaluation of Guided Regularized Random Forest for classification and regression tasks in remote sensing

New Earth observation missions and technologies are delivering large amounts of data. Processing this data requires developing and evaluating novel dimensionality reduction approaches to identify the most informative features for classification and regression tasks. Here we present an exhaustive evaluation of Guided Regularized Random Forest (GRRF), a feature selection method based on Random Forest. GRRF does not require fixing a priori the number of features to be selected or setting a threshold of the feature importance. Moreover, the use of regularization ensures that features selected by GRRF are non-redundant and representative. Our experiments based on various kinds of remote sensing images, show that GRRF selected features provides similar results to those obtained when using all the available features. However, the comparison between GRRF and standard random forest features shows substantial differences: in classification, the mean overall accuracy increases by almost 6% and, in regression, the decrease in RMSE almost reaches 2%. These results demonstrate the potential of GRRF for remote sensing image classification and regression. Especially in the context of increasingly large geodatabases that challenge the application of traditional methods.     

Mezen Bay—a key area for understanding Weichselian glaciations in northern Russia

Sediment successions in coastal cliffs around Mezen Bay, southeastern White Sea, record an unusually detailed history of former glaciations, interstadial marine and fluvial events from the Weichselian. A regional glaciation model for the Weichselian is based on new data from the Mezen Bay area and previously published data from adjacent areas. Following the Mikulinian (Eemian) interglacial a shelf‐centred glaciation in the Kara Sea is reflected in proglacial conditions at 100–90 ka. A local ice‐cap over the Timan ridge existed between 75 and 65 ka. Renewed glaciation in the Kara Sea spread southwestwards around 60 ka only, interrupted by a marine inundation, before it advanced to its maximum position at about 55–50 ka. After a prolonged ice‐free period, the Scandinavian ice‐sheet invaded the area from the west and terminated east of Mezen Bay about 17 ka. The previously published evidence of a large ice‐dammed lake in the central Arkhangelsk region, Lake Komi, finds no support in this study. Copyright © 2003 John Wiley & Sons, Ltd.     

Evidence for a mid‐Pleistocene shift of ice‐drift pattern in the Nordic seas

Sediment proxy records from a continuous, 1.5 million year long deep‐sea sediment core from a site in the western Norwegian Sea were used to obtain new insights into the nature of palaeoceanographic change in the northern North Atlantic (Nordic seas) during the climatic shift of the Mid‐Pleistocene Revolution (MPR). Red‐green sediment colour and magnetic susceptibility records both reveal significant differences in their mean values when comparing the intervals older than 700 000 yr (700 ka) with those from the past 500 kyr. The timing and duration of these changes indicates that the MPR in the Nordic seas is characterised by a gradual transition lasting about 200 kyr. Together with further sedimentological evidence this suggests that the mid‐Pleistocene climate shift was accompanied by a general change in ice‐drift pattern. It is further proposed that prior to the onset of the major late Pleistocene glaciations in the Northern Hemisphere a significant proportion of the ice in the eastern Nordic seas originated from a southern provenance, whereas later it dominantly came from the surrounding landmasses. Copyright © 2003 John Wiley & Sons, Ltd.     

Investigating the mechanisms leading to the deglaciation of past continental northern hemisphere ice sheets with the CLIMBER–GREMLINS coupled model

A coupling procedure between a climate model of intermediate complexity (CLIMBER-2.3) and a 3-dimensional thermo-mechanical ice-sheet model (GREMLINS) has been elaborated. The resulting coupled model describes the evolution of atmosphere, ocean, biosphere, cryosphere and their mutual interactions. It is used to perform several simulations of the Last Deglaciation period to identify the physical mechanisms at the origin of the deglaciation process. Our baseline experiment, forced by insolation and atmospheric CO₂, produces almost complete deglaciation of past northern hemisphere continental ice sheets, although ice remains over the Cordilleran region at the end of the simulation and also in Alaska and Eastern Siberia. Results clearly demonstrate that, in this study, the melting of the North American ice sheet is critically dependent on the deglaciation of Fennoscandia through processes involving switches of the thermohaline circulation from a glacial mode to a modern one and associated warming of the northern hemisphere. A set of sensitivity experiments has been carried out to test the relative importance of both forcing factors and internal processes in the deglaciation mechanism. It appears that the deglaciation is primarily driven by insolation. However, the atmospheric CO₂ modulates the timing of the melting of the Fennoscandian ice sheet, and results relative to Laurentide illustrate the existence of threshold CO₂ values, that can be translated in terms of critical temperature, below which the deglaciation is impeded. Finally, we show that the beginning of the deglaciation process of the Laurentide ice sheet may be influenced by the time at which the shift of the thermohaline circulation from one mode to the other occurs.     

Late Quaternary submarine bedforms and ice‐sheet flow in Gerlache Strait and on the adjacent continental shelf,Antarctic Peninsula

Geophysical data from Gerlache Strait, Croker Passage, Bismarck Strait and the adjacent continental shelf reveal streamlined subglacial bedforms that were produced at the bed of the Antarctic Peninsula Ice Sheet (APIS) during the last glaciation. The spatial arrangement and orientation of these bedforms record the former drainage pattern and flow dynamics of an APIS outlet up‐flow, and feeding into, a palaeo‐ice stream in the Western Bransfield Basin. Evidence suggests that together, they represent a single ice‐flow system that drained the APIS during the last glaciation. The ice‐sheet outlet flowed north/northeastwards through Gerlache Strait and Croker Passage and converged with a second, more easterly ice‐flow tributary on the middle shelf to form the main palaeo‐ice stream. The dominance of drumlins with low elongation ratios suggests that ice‐sheet outlet draining through Gerlache Strait was comparatively slower than the main palaeo‐ice stream in the Western Bransfield Basin, although the low elongation ratios may also partly reflect the lack of sediment. Progressive elongation of drumlins further down‐flow indicates that the ice sheet accelerated through Croker Passage and the western tributary trough, and fed into the main zone of streaming flow in the Western Bransfield Basin. Topography would have exerted a strong control on the development of the palaeo‐ice stream system but subglacial geology may also have been significant given the transition from crystalline bedrock to sedimentary strata on the inner–mid‐shelf. In the broader context, the APIS was drained by a number of major fast‐flowing outlets through cross‐shelf troughs to the outer continental shelf during the last glaciation. Copyright © 2004 John Wiley & Sons, Ltd.     

Palaeotemperature reconstructions of the European permafrost zone during marine oxygen isotope Stage 3 compared with climate model results

A palaeotemperature reconstruction based on periglacial phenomena in Europe north of approximately 51 °N, is compared with high‐resolution regional climate model simulations of the marine oxygen isotope Stage 3 (Stage 3) palaeoclimate. The experiments represent Stage 3 warm (interstadial), Stage 3 cold (stadial) and Last Glacial Maximum climatic conditions. The palaeotemperature reconstruction deviates considerably for the Stage 3 cold climate experiments, with mismatches up to 11 °C for the mean annual air temperature and up to 15 °C for the winter temperature. However, in this reconstruction various factors linking climate and permafrost have not been taken into account. In particular a relatively thin snow cover and high climatic variability of the glacial climate could have influenced temperature limits for ice‐wedge growth. Based on modelling the 0 °C mean annual ground temperature proves to be an appropriate upper temperature limit. Using this limit, mismatches with the Stage 3 cold climate experiments have been reduced but still remain. We therefore assume that the Stage 3 ice wedges were generated during short (decadal time‐scale) intervals of extreme cold climate, below the mean temperatures indicated by the Stage 3 cold climate model simulations. Copyright © 2003 John Wiley & Sons, Ltd.     

Estimation of snow water equivalent using microwave radiometry over Arctic first‐year sea ice

The magnitude and spatial distribution of snow on sea ice are both integral components of the ocean–sea‐ice–atmosphere system. Although there exists a number of algorithms to estimate the snow water equivalent (SWE) on terrestrial surfaces, to date there is no precise method to estimate SWE on sea ice. Physical snow properties and in situ microwave radiometry at 19, 37 and 85 GHz, V and H polarization were collected for a 10‐day period over 20 first‐year sea ice sites. We present and compare the in situ physical, electrical and microwave emission properties of snow over smooth Arctic first‐year sea ice for 19 of the 20 sites sampled. Physical processes creating the observed vertical patterns in the physical and electrical properties are discussed. An algorithm is then developed from the relationship between the SWE and the brightness temperature measured at 37 GHz (55°) H polarization and the air temperature. The multiple regression between these variables is able to account for over 90% of the variability in the measured SWE. This algorithm is validated with a small in situ data set collected during the 1999 field experiment. We then compare our data against the NASA snow thickness algorithm, designed as part of the NASA Earth Enterprise Program. The results indicated a lack of agreement between the NASA algorithm and the algorithm developed here. This lack of agreement is attributed to differences in scale between the Special Sensor Microwave/Imager and surface radiometers and to differences in the Antarctic versus Arctic snow physical and electrical properties. Copyright © 2003 John Wiley & Sons, Ltd.     

Ice‐cover variability on shallow lakes at high latitudes: model simulations and observations

A one‐dimensional thermodynamic model for simulating lake‐ice phenology is presented and evaluated. The model can be driven with observed daily or hourly atmospheric forcing of air temperature, relative humidity, wind speed, cloud amount and snowfall. In addition to computing the energy balance components, key model output includes the temperature profile at an arbitrary number of levels within the ice/snow (or the water temperature if there is no ice) and ice thickness (clear ice and snow‐ice) on a daily basis, as well as freeze‐up and break‐up dates. The lake‐ice model is used to simulate ice‐growth processes on shallow lakes in arctic, sub‐arctic, and high‐boreal forest environments. Model output is compared with field and remote sensing observations gathered over several ice seasons. Simulated ice thickness, including snow‐ice formation, compares favourably with field measurements. Ice‐on and ice‐off dates are also well simulated when compared with field and satellite observations, with a mean absolute difference of 2 days. Model simulations and observations illustrate the key role that snow cover plays on the seasonal evolution of ice thickness and the timing of spring break‐up. It is also shown that lake morphometry, depth in particular, is a determinant of ice‐off dates for shallow lakes at high latitudes. Copyright © 2003 John Wiley & Sons, Ltd.     

The importance of oxygen isotope provenance in relation to solute content of bulk meltwaters at Imersuaq Glacier,West Greenland

Stable oxygen isotope analysis and measurement of several dissolved cations and anions of bulk meltwater samples have provided information about the hydrochemical environment of the glacial hydrological system at Imersuaq Glacier, an outlet tongue from the Greenland ice‐sheet, West Greenland. The samples were collected at frequent intervals during the period 20–28 July 2000 in a small (<20 L s^?1) englacial meltwater outlet at the glacier margin. The results document the following findings: (i) a marked diurnal variation of δ¹⁸O is related to the composition of oxygen isotope provenances, mainly near‐marginal local superimposed ice and basal up‐sheared ice further up‐glacier; (ii) a relationship is seen between all base cations (Na⁺, K⁺, Ca²⁺, Mg²⁺), SO₄^2? and δ¹⁸O, indicating that solute acquisition is provided by solid–solution contact with the up‐sheared ice—as the relationship with Cl^? is weak the influence of seasalt‐derived solutes is small in the area; (iii) when the melt rate is high, two diurnal maxima of δ¹⁸O values and solute concentrations are measured, and it is suggested that a snow meltwater component is responsible for the second maximum of δ¹⁸O—a short residence time leads to a delayed decrease in ion concentrations. Copyright © 2003 John Wiley & Sons, Ltd.     

Pervasive wetland flooding in the glacial drift prairie of North Dakota (USA)

This article describes a unique flood hazard, produced by the dramatic expansion of wetlands in Nelson County, located within the North American Prairie Pothole Region of North Dakota, USA. There has been an unprecedented increase in the number, average size, and permanence of prairie wetlands, and a significant increase in the size of a closed lake (Stump Lake) due to a decade-long wet spell that began in 1993 following a prolonged drying trend. Base-line land cover information from the 1992 USGS National Land Cover Characterization dataset, and a Landsat TM scene acquired 9 July 2001 are used to assess the growth of the closed lake and wetland pond surface areas, and to analyze the type and area of various land cover classes inundated between 1992 and 2001. The open water profile in Nelson County changed from one marked by relatively comparable coverage of closed lake and wetland pond areas in 1992, to one in which wetland open water accounted for the vast majority of total open water in 2001. The bulk of the wetland pond area expansion occurred by displacing existing wetland vegetation and agricultural cropland. Producers responded to the flood hazard by filing Federal Crop Insurance Corporation (FCIC) claims and enrolling cropland in the Conservation Reserve Program (CRP), a federal land retirement program. Land taken out of agricultural production has had an enormous impact upon the agricultural sector that forms the economic base of the rural economy. In 2001 the land taken out of production due to CRP enrollment and preventive planting claims represented nearly 42% of Nelson County’s 205.2 K ha base agricultural land. The patterns obtained from this detailed study of Nelson County are likely to be the representative of the more publicized flood disaster occurring within the Devils Lake Basin of North Dakota.