Time filter

Source Type

Beauchemin M.,Canada Center For Remote Sensing

In the first part of this paper, we present a method to build affinity matrices for spectral clustering from a density estimator relying on K-means with subbagging procedure. The approach is anchored in the theoretical works of Wong (1980, 1982a, b) [13-15] on the asymptotic properties of K-means as a density estimation method. The subbagging procedure is introduced to improve the density estimate accuracy. The behavior of the proposed method is analyzed on diverse data sets and two new mechanisms are suggested to improve clustering results on non-convex data. In the second part of the paper, we establish a link between the presented method and the evidence accumulation clustering (EAC) approach by showing that a normalized version of the density-based similarity matrix is approximately equal to a normalized version of the co-association matrix. The co-association matrix provides the co-occurrence probability of data pairs assigned to a same cluster over multiple K-means clustering partitions. Experimental results on artificial and real data demonstrate the effectiveness of the method and provide empirical support for the established link. © 2014. Source

Although studies agree that climate warming will cause permafrost thaw, projected permafrost conditions differ widely, and most projections use half degree latitude/longitude or coarser spatial resolution. Using a process-based model, this study projected changes of permafrost from 2010 to 2200 at 30 m by 30 m resolution for a region in the northwest of the Hudson Bay Lowlands in Canada. This long-term spatially detailed modeling revealed some general features of permafrost dynamics with climate warming. Temporally, permafrost degradation at a site can be divided into five stages: gradual-thawing stage, increased-thawing stage, frequent-talik stage, isothermal-permafrost stage, and permafrost-free stage. This study determined the beginning or ends of the stages for each grid cell and mapped the degradation stages in this region. Spatially, permafrost was predicted to become increasingly discontinuous with climate warming. By the end of the 22nd century, only 20% to 65% of the land area in this region will be underlain by permafrost. With the formation of taliks, the maximum summer thaw depth will increase significantly, and near-surface permafrost will disappear in many areas while permafrost at depth can persist for decades. Thus, the spatial distribution of near-surface permafrost and permafrost at depth can be very different. This study also shows that climate scenarios, the depth of permafrost considered, spatial resolution and associated ground conditions used for modeling could cause significant differences in permafrost projections. Key Points Permafrost degradation at a site can be divided into five stages Permafrost is predicted to become increasingly discontinuous The distributions differ for permafrost in near-surface and at depth ©2013. Her Majesty the Queen in Right of Canada. Source

Toutin T.,Canada Center For Remote Sensing
IEEE Transactions on Geoscience and Remote Sensing

The new high-resolution mode of Radarsat-2 was evaluated for digital elevation model (DEM) generation using stereo-radargrammetry with Toutin's 3-D physical model as part of the Canadian Space Agency's programs. In addition, the impact of radar parameters on stereo-extracted DEMs was evaluated. Three stereo pairs using different radar parameters (resolutions; HH and VV polarizations; slant and ground range; image spacing) from ultrafine-mode Radarsat-2 data acquired over a test site north of Qubec City, Canada, were formed to generate DEMs, which were thus compared to 15-cm-accurate lidar elevation data. Results showed a good accuracy for the three stereo-extracted DEMs: less than 0.8 m for the horizontal positioning and around 3 m (1σ) for the elevation on bare soils, with small 3-D biases (2-3 m). The HH stereo pair in slant-range format slightly but consistently achieved the best results. © 2006 IEEE. Source

The impact of water stress on plant stomatal conductance (g) has been widely studied but with little consensus as to the processes governing its responses. The photosynthesis-driven stomatal conductance models usually employ constant model parameters and attribute the decrease of g from water stress to the reduction of leaf photosynthesis. This has been challenged by studies showing that the model parameter values decrease when the plant is under water stress. In this study, the impact of plant water stress on the parameter values in stomatal conductance models is evaluated using the approach recently developed by S. Wang et al. and the tower flux measurements at a Canadian boreal aspen forest. Results show that the slope parameter (a) in the stomatal conductance models decreases substantially with the development of plant water stress. The magnitude of this reduction is dependent on how plant water stress is represented. Overall, the relative reduction of α from its maximum value is 28% when soil water content decreases from 0.38 to 0.18 m 3 m -3, and is 38% when Bowen ratio increases from 0.25 to 3.5. Equations for α correction to account for water stress impacts are proposed. Further studies on different ecosystems are necessary to quantify the parameter variations with water stress among different climate regions and plant species. © 2012 American Meteorological Society. Source

Toutin T.,Canada Center For Remote Sensing
IEEE Transactions on Geoscience and Remote Sensing

Digital surface models (DSMs) extracted from high-resolution Radarsat-2 stereo-images using different geometric modeling (deterministic, new hybrid, and empirical) are evaluated. The 3-D deterministic models are Toutin's and hybrid Toutin's models (TM and HTM) developed at the Canada Centre for Remote Sensing, and the empirical model is the rational function model (RFM). TM is computed with one and eight ground control points (GCPs), HTM without GCP and RFM supplied by MacDonald, Dettwiler and Associates Ltd. is postprocessed with 3-9 GCPs depending of degrees of 2-D polynomial functions. The DSMs are then generated and compared to 0.2-m accurate lidar elevation data. Because DSMs included the height of land covers, elevation linear errors with 68% and 90% confidence level (LE68 and LE90) are computed and compared over bare surfaces only. LE90 results are: TM with eight GCPs achieves the best results (6.3 m), then HTM with no GCP (7 m), TM with one GCP (8.6 m), and finally RFM the worst (9.7 m) whatever the polynomial degree and GCP number. HTM is the only modeling not using any GCP, which offers a strong advantage in operational environments. © 2012 IEEE. Source

Discover hidden collaborations