Clustering Based Large Margin Classification: A Scalable Approach using SOCP Formulation

This paper presents a novel Second Order Cone Programming (SOCP) formulation for large scale binary classification tasks. Assuming that the class conditional densities are mixture distributions, where each component of the mixture has a spherical covariance, the second order statistics of the components can be estimated efficiently using clustering algorithms like BIRCH. For each cluster, the second order moments are used to derive a second order cone constraint via a Chebyshev-Cantelli inequality. This constraint ensures that any data point in the cluster is classified correctly with a high probability. This leads to a large margin SOCP formulation whose size depends on the number of clusters rather than the number of training data points. Hence, the proposed formulation scales well for large datasets when compared to the state-of-the-art classifiers, Support Vector Machines (SVMs). Experiments on real world and synthetic datasets show that the proposed algorithm outperforms SVM solvers in terms of training time and achieves similar accuracies.

A model investigation of vegetation-atmosphere interactions on a millennial timescale

A terrestrial biosphere model with dynamic vegetation capability, Integrated Biosphere Simulator (IBIS2), coupled to the NCAR Community Atmosphere Model (CAM2) is used to investigate the multiple climate-forest equilibrium states of the climate system. A 1000-year control simulation and another 1000-year land cover change simulation that consisted of global deforestation for 100 years followed by re-growth of forests for the subsequent 900 years were performed. After several centuries of interactive climate-vegetation dynamics, the land cover change simulation converged to essentially the same climate state as the control simulation. However, the climate system takes about a millennium to reach the control forest state. In the absence of deep ocean feedbacks in our model, the millennial time scale for converging to the original climate state is dictated by long time scales of the vegetation dynamics in the northern high latitudes. Our idealized modeling study suggests that the equilibrium state reached after complete global deforestation followed by re-growth of forests is unlikely to be distinguishable from the control climate. The real world, however, could have multiple climate-forest states since our modeling study is unlikely to have represented all the essential ecological processes (e. g. altered fire regimes, seed sources and seedling establishment dynamics) for the reestablishment of major biomes.

Learning from positive and unlabelled examples using maximum margin clustering

Learning from Positive and Unlabelled examples (LPU) has emerged as an important problem in data mining and information retrieval applications. Existing techniques are not ideally suited for real world scenarios where the datasets are linearly inseparable, as they either build linear classifiers or the non-linear classifiers fail to achieve the desired performance. In this work, we propose to extend maximum margin clustering ideas and present an iterative procedure to design a non-linear classifier for LPU. In particular, we build a least squares support vector classifier, suitable for handling this problem due to symmetry of its loss function. Further, we present techniques for appropriately initializing the labels of unlabelled examples and for enforcing the ratio of positive to negative examples while obtaining these labels. Experiments on real-world datasets demonstrate that the non-linear classifier designed using the proposed approach gives significantly better generalization performance than the existing relevant approaches for LPU.

Vegetation impact on stream chemical fluxes: Mule Hole watershed (South India)

The proportion of chemical elements passing through vegetation prior to being exported in a stream was quantified for a forested tropical watershed(Mule Hole, South India) using an extensive hydrological and geochemical monitoring at several scales. First, a solute annual mass balance was established at the scale of the soil-plant profile for assessing the contribution of canopy interaction and litter decay to the solute fluxes of soil inputs (overland flow) and soil outputs (pore water flow as seepages). Second, based on the respective contributions of overland flow and seepages to the stream flow as estimated by a hydrological lumped model, we assigned the proportion of chemical elements in the stream that transited through the vegetation at both flood event (End Member Mixing Analysis) and seasonal scales. At the scale of the 1D soil-plant profile, leaching from the canopy constituted the main source of K above the ground surface. Litter decay was the main source of Si, whereas alkalinity, Ca and Mg originated in the same proportions from both sources. The contribution of vegetation was negligible for Na. Within the soil, all elements but Na were removed from the pore water in proportions varying from 20% for Cl to 95% for K: The soil output fluxes corresponded to a residual fraction of the infiltration fluxes. The behavior of K, Cl, Ca and Mg in the soil-plant profile can be explained by internal cycling, as their soil output fluxes were similar to the atmospheric inputs. Na was released from soils as a result of Na-plagioclase weathering and accompanied by additional release of Si. Concentration of soil pore water by evapotranspiration might limit the chemical weathering in the soil. Overall, the solute K, Ca, Mg, alkalinity and Si fluxes associated with the vegetation turnover within the small experimental watershed represented 10-15 times the solute fluxes exported by the stream, of which 83-97% transited through the vegetation. One important finding is that alkalinity and Si fluxes at the outlet were not linked to the ``current weathering'' of silicates in this watershed. These results highlight the dual effect of the vegetation cover on the solute fluxes exported from the watershed: On one hand the runoff was limited by evapotranspiration and represented only 10% of the annual rainfall, while on the other hand, 80-90% of the overall solute flux exported by the stream transited through the vegetation. The approach combining geochemical monitoring and accurate knowledge of the watershed hydrological budget provided detailed understanding of several effects of vegetation on stream fluxes: (1) evapotranspiration (limiting), (2) vertical transfer through vegetation from vadose zone to ground surface (enhancing) and (3) redistribution by throughfalls and litter decay. It provides a good basis for calibrating geochemical models and more precisely assessing the role of vegetation on soil processes. (C) 2014 Elsevier Ltd. All rights reserved.

Projected climate change impacts on vegetation distribution over Kashmir Himalayas

Despite high vulnerability, the impact of climate change on Himalayan ecosystem has not been properly investigated, primarily due to the inadequacy of observed data and the complex topography. In this study, we mapped the current vegetation distribution in Kashmir Himalayas from NOAA AVHRR and projected it under A1B SRES, RCP-4.5 and RCP-8.5 climate scenarios using the vegetation dynamics model-IBIS at a spatial resolution of 0.5A degrees. The distribution of vegetation under the changing climate was simulated for the 21st century. Climate change projections from the PRECIS experiment using the HADRM3 model, for the Kashmir region, were validated using the observed climate data from two observatories. Both the observed as well as the projected climate data showed statistically significant trends. IBIS was validated for Kashmir Himalayas by comparing the simulated vegetation distribution with the observed distribution. The baseline simulated scenario of vegetation (1960-1990), showed 87.15 % agreement with the observed vegetation distribution, thereby increasing the credibility of the projected vegetation distribution under the changing climate over the region. According to the model projections, grasslands and tropical deciduous forests in the region would be severely affected while as savannah, shrubland, temperate evergreen broadleaf forest, boreal evergreen forest and mixed forest types would colonize the area currently under the cold desert/rock/ice land cover types. The model predicted that a substantial area of land, presently under the permanent snow and ice cover, would disappear by the end of the century which might severely impact stream flows, agriculture productivity and biodiversity in the region.

Vegetation structure and composition of tropical evergreen and deciduous forests in Uttara Kannada District, Western Ghats under different disturbance regimes

In the present paper, we present the structure and composition of tropical evergreen and deciduous forests in the Western Ghats monitored under a long-term programme involving Indian Institute of Science, Earthwatch and volunteer investigators from HSBC. Currently, there is limited evidence on the status and dynamics of tropical forests in the context of human disturbance and climate change. Observations made in this study show that the `more disturbed' evergreen and one of the deciduous plots have low species diversity compared to the less-disturbed forests. There are also variations in the size class structure in the more and `less disturbed' forests of all the locations. The variation is particularly noticeable in the DBH size class 10 - 15 cm category. When biomass stock estimates are considered, there was no significant difference between evergreen and deciduous forests. The difference in biomass stocks between `less disturbed' and `more disturbed' forests within a forest type is also low. Thus, the biomass and carbon stock has not been impacted despite the dependence of communities on the forests. Periodic and long-term monitoring of the status and dynamics of the forests is necessary in the context of potential increased human pressure and climate change. There is, therefore, a need to inform the communities of the impact of extraction and its effect on regeneration so as to motivate them to adopt what may be termed as ``adaptive resource management'', so as to sustain the flow of forest products.

Influence of the upwelling events on the delta C-13 and delta O-18 of the benthic bivalve shells of the South Western Continental Margin of India

In this study, the influence of the spatial and temporal variability of upwelling intensity and the associated biological productivity observed during different phases of summer monsoon along the southwestern continental margin of India (SWCMI) on the delta C-13 and delta O-18 of the inorganic biogenic carbonate shells was investigated. Multispecies benthic bivalve shells (1-5 mm) separated from ten surface sediment samples of SWCMI (off 12 degrees N, 10 degrees N and 9 degrees N) collected during the onset (OSM) and peak (PSM) phase of the summer monsoon of 2009 were analysed for delta C-13 and delta O-18. Sea surface temperature along the study region indicates prominent upwelling in PSM than in OSM. A comparison of analytical and predicted values for delta O-18 in the bivalve shells confirmed their in situ origin during both the sampling periods. During PSM, the delta C-13 values in the benthic bivalve shells were more depleted in C-13 than during OSM which recorded lower values of delta C-13 in dissolved inorganic carbon of bottom waters expected in the study region in PSM due to the upwelled waters, high surface productivity and the associated high degradation of the organic matter in the subsurface and bottom waters. However, this depletion of delta C-13 was not observed in benthic bivalve shells obtained from 10 degrees N, since it is influenced by high export fluxes of carbon from the Cochin estuary since early monsoon months.

Disaggregation of LST over India: comparative analysis of different vegetation indices

The non-availability of high-spatial-resolution thermal data from satellites on a consistent basis led to the development of different models for sharpening coarse-spatial-resolution thermal data. Thermal sharpening models that are based on the relationship between land-surface temperature (LST) and a vegetation index (VI) such as the normalized difference vegetation index (NDVI) or fraction vegetation cover (FVC) have gained much attention due to their simplicity, physical basis, and operational capability. However, there are hardly any studies in the literature examining comprehensively various VIs apart from NDVI and FVC, which may be better suited for thermal sharpening over agricultural and natural landscapes. The aim of this study is to compare the relative performance of five different VIs, namely NDVI, FVC, the normalized difference water index (NDWI), soil adjusted vegetation index (SAVI), and modified soil adjusted vegetation index (MSAVI), for thermal sharpening using the DisTrad thermal sharpening model over agricultural and natural landscapes in India. Multi-temporal LST data from Landsat-7 Enhanced Thematic Mapper Plus (ETM+) and Moderate Resolution Imaging Spectroradiometer (MODIS) sensors obtained over two different agro-climatic grids in India were disaggregated from 960 m to 120 m spatial resolution. The sharpened LST was compared with the reference LST estimated from the Landsat data at 120 m spatial resolution. In addition to this, MODIS LST was disaggregated from 960 m to 480 m and compared with ground measurements at five sites in India. It was found that NDVI and FVC performed better only under wet conditions, whereas under drier conditions, the performance of NDWI was superior to other indices and produced accurate results. SAVI and MSAVI always produced poorer results compared with NDVI/FVC and NDWI for wet and dry cases, respectively.