Ashlin Richardson

Mapping fire scars using Radarsat-2 polarimetric SAR data

Abstract Climate change is increasing the frequency and size of wildfire events in Canadas forests. The size, distribution, and remoteness of boreal forest fire events make them a challenge to accurately monitor. Radarsat-2 is all-weather radar technology and offers high spatial resolution, cross polarization, polarimetric capabilities, and a wide swath width. In this paper, Radarsat-2 fine quad-pol data were analyzed utilizing the polarimetric phase information to map historical fire scars over two main study sites. The study demonstrated that historical fire scars, less than 10 years old and without strong topographic variation, had distinct polarimetric signatures with relatively higher quad-pol probabilities, leading to detection with low false alarm rates. By combining polarimetric decomposition and new classification approaches, fire scars were extracted; the resulting burned areas matched the true burned areas according to GIS polygons from the provincial forest fire database. Our new K-Nearest Neighbors Graph Clustering classifier, unlike the classical Wishart scheme, does not depend on backscatter intensity; it relies more on polarization information, and is more tolerant of topographic variations. These new approaches have revealed an exciting new application, mapping historical fire scars with polarimetric radar.

A deceleration model for bicycle peloton dynamics and group sorting

Extending earlier computer models of bicycle peloton dynamics, we add a deceleration parameter by which deceleration magnitude varies as a function of cyclist strength. This model is validated by applying speed data from a mass-start race composed of 14 cyclists, and running simulation trials using 14 simulated cyclists that generated positional profiles which compare well with the positional profiles observed in the actual mass-start race data. Keeping constant the speed variation profile from the mass-start race as introduced into the simulation, a set of simulation experiments were run, including: varying the number of cyclists; varying the duration of a single near-threshold output event; and varying the course elevation. The results consistently show sorting of pelotons into smaller groups whose mean fitness corresponds with relative group position, i.e. fitter groups are closer to the front. Sorting of pelotons into fitness-related groups provides insight into the mechanics of similar group divisions within biological collectives in which members present heterogeneous physiological fitness capacities.

Data Fusion Study Between Polarimetric SAR, Hyperspectral and Lidar Data for Forest Information

ALOS PALSAR L-band quad-pol data were acquired over our study area on Vancouver Island in British Columbia in the summer of 2007. The site has significant topographic relief and high biomass in this temperate coastal rainforest. Our emphasis was on integration and fusion techniques of polarimetric SAR, hyperspectral and LIDAR data for useful forest information extraction. The polarimetric SAR techniques and analysis methods studied in this project drew on the work of other researchers. The Jong-Sen Lee algorithm for polarization compensation for terrain azimuth slope variations was implemented and tested. The Shane Cloude decomposition method, with basic types of scattering analysis for reducing sensitivity to topography effects was examined and applied. In this study, the hyperspectral data was used for providing high spectral resolution information, such as major forest species and land-cover characterization, and the LIDAR data were utilized to generate information related to vertical structure of both the underlying topography and the forest structure. The combination of these data sources and techniques provided an opportunity to examine the potential capabilities of polarimetric SAR and the synergy of the fused data for forest classification.

Unsupervised nonparametric classification of polarimetric SAR data using the K-nearest neighbor graph

Polarimetric SAR classifications are often based on assumptions about the shape of clusters in the data space. Such a scheme will fail for nonlinear structures in the feature space, unless the classification algorithm has the capacity to describe cluster shapes in sufficient generality. Existing polarimetric SAR classification methods are faced by this exact problem: typically they initialize clusters in the Cloude-Pottier parameter space [1], further optimizing them in the coherency matrix space [2, 3]. Methods using K-means [2] or agglomeration [3] require clusters that are spherical, or compact and well separated, respectively. In the Cloude-Pottier space, these requirements are not met, so initialization in the Cloude-Pottier space cannot be consistent with optimization by K-means or agglomeration. This paper sets out to address this problem, by implementing a new data-driven clustering approach, for arbitrarily shaped clusters. It is applied to quad-polarisation data, demonstrating the new methodologys potential for forest land-cover type discrimination.

A framework for efficiently parallelizing nonlinear noise reduction algorithm

In hyperspectral imagery, noise reduction is a vital and common pre-processing step that needs to be executed accurately and efficiently. Until recently, hyperspectral data was modeled using linear stochastic processes and the noise was assumed to manifest itself in a narrow spatial frequency band. The signal and noise are thus considered independent and most of the proposed noise reduction algorithms transform the hyperspectral data linearly from one space to another for noise and signal separation. Hyperspectral data, however, exhibits nonlinear characteristics making the noise frequency and signal dependent [1, 2]. Therefore, to accurately reduce the noise in hyperspectral data, a nonlinear noise reduction algorithm, such as the one we propose in this paper, must be considered. The algorithm, however, is computationally expensive and requires parallelization. To this end, we offer a framework which we have implemented and evaluated.

Forest stand level correlation analysis of ALOS-1 PALSAR signatures

In this paper we analyze the effects of polarization, environmental conditions and forest structure upon the backscatter response of forested stands. This analysis is based upon a time series of ALOS-1 PALSAR images acquired over our study site in Xunke County, Heilongjiang Province, China. Based on six scenes, we analyzed the polarization and environment conditions on the forest stands. Backscatter coefficients of HV channel had a greater dynamic range than HH channel. HV channel was less influenced by weather and wind speed conditions. Our observations found canopy density greatly influenced the forest stand backscatter. Backscatter coefficient showed weak correlations to canopy density, mean tree height and mean diameter at breast height (DBH). Correlations were much stronger when the forest stands were grouped with canopy density or mean tree height. Before grouping the highest correlation coefficient between backscatter coefficients and tree height was 0.377, the value for HV image acquired on August 07, 2007. After grouping these forest stands by canopy density, the correlation was 0.95 to tree height. We also analyzed the correlations with two different tree species, and obtained similar results.

Hierarchical unsupervised nonparametric classification of polarimetric SAR time series data

Clustering (and classification) among other approaches of land-cover type discrimination for Polarimetric SAR (Pol-SAR) data often explicitly or implicitly assume a lot about the shape of the clusters (or the classes, in the case of classification). For example, this is an issue for Pol-SAR classification methods [1,2,3] that initialize clusters in decomposition parameter feature spaces [4], subsequently refining the clusters by Wishart moving-means iterations in coherency matrix (T3) space. Indeed, using the means as cluster (or class) representatives can be successful, provided that clusters in the data are compact, well separated, and convex. However, highly nonlinear features and unusually shaped clusters are often obtained when dealing with PolSAR data. To address this issue we present a data-driven hierarchical clustering technique. This we demonstrate for forest-type discrimination purposes with a multi-temporal Radarsat-2 sandwich.

Eigen decomposition parameter based forest mapping using Radarsat-2 PolSAR data

In this paper, a set of polarimetric eigenvalue and eigenvector based parameters, e.g. entropy and anisotropy, are investigated for forest application. The correlation terms of the eigenvectors, μ1 and μ2, are found to be better for forest mapping in both summer and winter using Radarsat-2 quad-polarimetric space borne SAR data. These are used to automatically identify forest class pixels from the volume scattering category of a Freeman-Durden Wishart unsupervised segmentation map. The algorithm scheme was developed and implemented using fully polarimetric Radarsat-2 SAR (PolSAR) data acquired in July and October and the validity was evaluated using the ground reference data created from SPOT5 K-clustering classification map.

Topographic relief compensation on spaceborne polarimetric SAR for forest applications

In order to use the advanced capabilities of polarimetric SAR data for forest applications, analysis methods must address topographic relief effects in mountainous regions. ALOS PALSAR L-band and Radarsat-2 C-band polarimetric SAR data over study sites in Hinton, Alberta, and the Greater Victoria Watershed District, BC, were collected and used to investigate the effectiveness of polarization orientation shifts correction, decomposition filtering techniques and local incident angle compensation. We found that the polarization orientation shifts estimated from PALSAR winter data corresponded to the topographic relief. The shifts from the PALSAR summer data were noisy. The polarization orientation shifts were not seen in the Radarsat-2 data. Various polarimetric parameterizations were useful for identifying terrain features and land cover types. The Cloude entropy, dominant scattering alpha and eigenvalues (λ2+λ3) were used to create HSV coding images for better differentiation of forest, vegetation and water surfaces with minimized topographic effects. The local incident angle compensation on single polarization backscatters was applied to Radarsat-2 data. Due to the high biomass volume in the study area and the higher radar frequency, the correlations between the polarization backscatter and the ground measured volumes at the plot level were poor.