Alexei N. Skurikhin

Polygonal tundra geomorphological change in response to warming alters future CO2 and CH4 flux on the Barrow Peninsula.

The landscape of the Barrow Peninsula in northern Alaska is thought to have formed over centuries to millennia, and is now dominated by ice-wedge polygonal tundra that spans drained thaw-lake basins and interstitial tundra. In nearby tundra regions, studies have identified a rapid increase in thermokarst formation (i.e., pits) over recent decades in response to climate warming, facilitating changes in polygonal tundra geomorphology. We assessed the future impact of 100 years of tundra geomorphic change on peak growing season carbon exchange in response to: (i) landscape succession associated with the thaw-lake cycle; and (ii) low, moderate, and extreme scenarios of thermokarst pit formation (10%, 30%, and 50%) reported for Alaskan arctic tundra sites. We developed a 30 × 30 m resolution tundra geomorphology map (overall accuracy:75%; Kappa:0.69) for our ~1800 km² study area composed of ten classes; drained slope, high center polygon, flat-center polygon, low center polygon, coalescent low center polygon, polygon trough, meadow, ponds, rivers, and lakes, to determine their spatial distribution across the Barrow Peninsula. Land-atmosphere CO2 and CH4 flux data were collected for the summers of 2006-2010 at eighty-two sites near Barrow, across the mapped classes. The developed geomorphic map was used for the regional assessment of carbon flux. Results indicate (i) at present during peak growing season on the Barrow Peninsula, CO2 uptake occurs at -902.3 10(6) gC-CO2 day(-1) (uncertainty using 95% CI is between -438.3 and -1366 10(6) gC-CO2 day(-1)) and CH4 flux at 28.9 10(6) gC-CH4 day(-1) (uncertainty using 95% CI is between 12.9 and 44.9 10(6) gC-CH4 day(-1)), (ii) one century of future landscape change associated with the thaw-lake cycle only slightly alter CO2 and CH4 exchange, while (iii) moderate increases in thermokarst pits would strengthen both CO2 uptake (-166.9 10(6) gC-CO2 day(-1)) and CH4 flux (2.8 10(6) gC-CH4 day(-1)) with geomorphic change from low to high center polygons, cumulatively resulting in an estimated negative feedback to warming during peak growing season.

Automated tree crown detection and size estimation using multi-scale analysis of high-resolution satellite imagery

We tested an automated multi-scale approach for detecting individual trees and estimating tree crown geometry using high spatial resolution satellite imagery. Individual tree crowns are identified as local extrema points in the Laplacian-of-Gaussian scale-space pyramid that is constructed based on linear scale-space theory. The approach simultaneously detects tree crown centres and estimates tree crown sizes (radiuses). We evaluated our method using two 0.6-m resolution QuickBird images of a forest site that underwent a large shift in tree density between image captures due to drought-associated mortality. The automated multi-scale approach produced tree count estimates with an accuracy of 54% and 73% corresponding to the dense and sparse forests, respectively. Estimated crown diameters were linearly correlated with field-measured crown diameters (r = 0.73–0.86). Tree count accuracies and size estimates were comparable with alternative methods. Future use of the presented approach is merited based on the results of our study, but requires further investigation in a broader range of forest types.

Arctic tundra ice-wedge landscape characterization by active contours without edges and structural analysis using high-resolution satellite imagery

In this letter, we present a semi-automated approach to identify and classify Arctic polygonal tundra landscape components, such as troughs, ponds, rivers and lakes, using high spatial resolution satellite imagery. The approach starts by segmenting water bodies from an image, which are then categorized using shape-based classification. Segmentation uses combination of multispectral bands and is based on the active contours without edges technique. The segmentation is robust to noise and can detect objects with weak boundaries, which is important for the extraction of troughs. Classification of the regions is accomplished by utilizing distance transform and regional structural characteristics. The approach is evaluated using 0.6 m resolution WorldView-2 satellite image of ice-wedge polygonal tundra. The segmentation user’s and producer’s accuracies are approximately 92% and 97%, respectively. Visual inspection of the classification results has demonstrated qualitatively accurate object categorization.

Selecting Summary Statistics in Approximate Bayesian Computation for Calibrating Stochastic Models

Approximate Bayesian computation (ABC) is an approach for using measurement data to calibrate stochastic computer models, which are common in biology applications. ABC is becoming the “go-to” option when the data and/or parameter dimension is large because it relies on user-chosen summary statistics rather than the full data and is therefore computationally feasible. One technical challenge with ABC is that the quality of the approximation to the posterior distribution of model parameters depends on the user-chosen summary statistics. In this paper, the user requirement to choose effective summary statistics in order to accurately estimate the posterior distribution of model parameters is investigated and illustrated by example, using a model and corresponding real data of mitochondrial DNA population dynamics. We show that for some choices of summary statistics, the posterior distribution of model parameters is closely approximated and for other choices of summary statistics, the posterior distribution is not closely approximated. A strategy to choose effective summary statistics is suggested in cases where the stochastic computer model can be run at many trial parameter settings, as in the example.

Proximity Graphs Based Multi-scale Image Segmentation

We present a novel multi-scale image segmentation approach based on irregular triangular and polygonal tessellations produced by proximity graphs. Our approach consists of two separate stages: polygonal seeds generation followed by an iterative bottom-up polygon agglomeration. We employ constrained Delaunay triangulation combined with the principles known from visual perception to extract an initial irregular polygonal tessellation of the image. These initial polygons are built upon a triangular mesh composed of irregular sized triangles, whose spatial arrangement is adapted to the image content. We represent the image as a graph with vertices corresponding to the built polygons and edges reflecting polygon relations. The segmentation problem is then formulated as Minimum Spanning Tree (MST) construction. We build a successive fine-to-coarse hierarchy of irregular polygonal partitions by an iterative graph contraction. It uses local information and merges the polygons bottom-up based on local region- and edge- based characteristics.

Visual attention based detection of signs of anthropogenic activities in satellite imagery

With increasing deployment of satellite imaging systems, only a small fraction of collected data can be subject to expert scrutiny. We present and evaluate a two-tier approach to broad area search for signs of anthropogenic activities in highresolution commercial satellite imagery. The method filters image information using semantically oriented interest points by combining Harris corner detection and spatial pyramid matching. The idea is that anthropogenic structures, such as rooftop outlines, fence corners, road junctions, are locally arranged in specific angular relations to each other. They are often oriented at approximately right angles to each other (which is known as rectilinearity relation). Detecting rectilinear structures provides an opportunity to highlight regions most likely to contain anthropogenic activity. This is followed by supervised classification of regions surrounding the detected corner points as anthropogenic vs. natural scenes. We consider, in particular, a search for signs of anthropogenic activities in uncluttered areas.

Learning tree-structured approximations for conditional random fields

Exact probabilistic inference is computationally intractable in general probabilistic graph-based models, such as Markov Random Fields and Conditional Random Fields (CRFs). We investigate spanning tree approximations for the discriminative CRF model. We decompose the original computationally intractable grid-structured CRF model containing many cycles into a set of tractable sub-models using a set of spanning trees. The structure of spanning trees is generated uniformly at random among all spanning trees of the original graph. These trees are learned independently to address the classification problem and Maximum Posterior Marginal estimation is performed on each individual tree. Classification labels are produced via voting strategy over the marginals obtained on the sampled spanning trees. The learning is computationally efficient because the inference on trees is exact and efficient. Our objective is to investigate the capability of approximation of the original loopy graph model with loopy belief propagation inference via learning a pool of randomly sampled acyclic graphs. We focus on the impact of memorizing the structure of sampled trees. We compare two approaches to create an ensemble of spanning trees, whose parameters are optimized during learning: (1) memorizing the structure of the sampled spanning trees used during learning and, (2) not storing the structure of the sampled spanning trees after learning and regenerating trees anew. Experiments are done on two image datasets consisting of synthetic and real-world images. These datasets were designed for the tasks of binary image denoising and man-made structure recognition.

An Automatic Framework for Figure-Ground Segmentation in Cluttered Backgrounds

This paper proposes an automatic framework for figure-ground segmentation of edged images in the presence of cluttered background. Our work employs perceptual grouping concepts to characterize image segments by means of their saliency, which is computed via tensor voting. The main innovation of our work is a case-based thresholding scheme which iteratively eliminates edge segments with low-saliency in multiple scales, preserving those that are more likely to belong to foreground. The key idea is classifying saliency histograms in several cases by considering the relative position of the modes of the figure/ground distributions and applying specific actions in each case. We have performed extensive experiments in order to evaluate our framework both quantitatively and qualitatively, including real images from the Berkeley dataset.

Hierarchical Spanning Tree-Structured Approximation for Conditional Random Fields: An Empirical Study

We present a learning algorithm to construct a discriminative Conditional Random Fields cascade model. We decompose the original grid-structured graph model using a set of spanning trees which are learned and added into the cascade architecture iteratively one after another. A spanning tree at each cascade layer takes both outputs from the previous layer nodes as well as the observed variables, which are processed by all the layers. The structure of spanning trees is generated uniformly at random among all spanning trees of the original graph. The result of the learning is the number of cascade layers, the structure of the generated spanning trees, and the set of optimized parameters corresponding to the spanning trees. We performed the experimental validation on synthetic and real-world imagery datasets and demonstrated better performance of the cascade tree-based model over the original grid-structured CRF model with loopy belief propagation inference.

Recursive active contours for hierarchical segmentation of wetlands in high-resolution satellite imagery of Arctic landscapes

We present a semi-automated approach to recognize and hierarchically partition water-body regions of Arctic tundra landscape, such as streams, inundated drained thaw lake basins and ice wedge polygon ponds, in high resolution satellite imagery. The approach integrates the active contours without edges (ACWE) technique and shape-based recognition, and introduces a recursive mode of ACWE application. We build a successive coarse-to-fine hierarchy of image partitions corresponding to the low-gradient Arctic wetlands by recursively partitioning them at the coarser scale into constituent parts. The approach is evaluated using 0.6 m resolution WorldView-2 satellite image of Arctic tundra landscape. The water-body regions segmentation producers accuracy is 97.7 %, and the users accuracy is 92.9 %. Visual inspection of the classification and hierarchical partitioning of the segmented water-body regions has demonstrated their qualitatively accurate recognition and hierarchical partitioning.