Treeline vegetation composition and change in Canada's western Subarctic from AVHRR and canopy reflectance modeling

Climate change is expected to have significant impacts on northern vegetation, particularly along transition zones such as the treeline. Studies of vegetation composition and change in this ecotone have largely focussed on local analysis of individual trees using labour intensive stand reconstruction techniques, which are spatially limited and do not consider vegetation types other than trees. Remote sensing may be well suited to monitoring recent changes across the treeline because it captures integrated changes of all vegetation life forms over large spatial extents. This research examines treeline vegetation composition and change along the western subarctic treeline mapped by Timoney et al. (1992) using a 1 km resolution, 22-year AVHRR archive from 1985-2006. While most remote sensing studies on vegetation change in arctic and subarctic regions only exploit information contained in the Normalized Difference Vegetation Index (NDVI), we examine long-term reflectance trends in AVHRR bands 1 and 2 in addition to NDVI. The GeoSail canopy reflectance model is used to map treeline composition by combining information from 22-year summertime and early springtime composite images. A set of spectral change vectors are then generated from GeoSail simulations and used to classify trends in AVHRR along the treeline to estimate vegetation change. Evaluation of vegetation composition against the MODIS Vegetation Continuous Fields (VCF) product that has been recently validated along the treeline reveals good spatial correspondence. Temporal trends are shown to agree with literature on tundra-taiga vegetation dynamics in recent decades. Evidence is presented that suggests replacement of bare surfaces with herb, conifer decline along the southern treeline, increased shrubiness, and increased conifer recruitment and growth in the north.     

Sample complexity for Markov chain self-tuner

For self-tuning control of a finite state Markov chain whose parametrized transition probabilities satisfy an ‘identifiability condition’, we establish a bound on the number of samples required to attain a prescribed measure of near-optimality with a prescribed probability.     

A computer program for an illness-death process with time-dependent covariates

This paper presents a computer program for estimating transition probabilities between states in a stochastic model for an illness-death process which incorporates time-dependent covariates. Parameters are estimated by the method of maximum likelihood using the Newton-Raphson iterative procedure. The program provides the standard normal deviate statistics as well as the value of the maximum of the likelihood function which can be used on repeated applications to test hypotheses concerning coefficients associated with covariates. Although this program is demonstrated by using a model with two ‘illness’ states and two ‘death’ states, it is also suitable for analyzing data with models involving fewer states, such as the analysis of survival time with covariates assuming a proportional hazard model.     

A ‘No Panacea Theorem’ for classifier combination

We introduce the ‘No Panacea Theorem’ (NPT) for multiple classifier combination, previously proved only in the case of two classifiers and two classes. In this paper, we extend the NPT to cases of multiple classifiers and multiple classes. We prove that if the combination function is continuous and diverse, there exists a situation in which the combination algorithm will give very bad performance. The proof relies on constructing ‘pathological’ probability density distributions that have high densities in particular areas such that the combination functions give incorrect classification. Thus, there is no optimal combination algorithm that is suitable in all situations. It can be seen from this theorem that the probability density functions (pdfs) play an important role in the performance of combination algorithms, so studying the pdfs becomes the first step of finding a good combination algorithm. Although devised for classifier combination, the NPT is also relevant to all supervised classification problems.     

Bayesian change detection based on spatial sampling and Gaussian mixture model

A Gaussian mixture model (GMM) and Bayesian inferencing based unsupervised change detection algorithm is proposed to achieve change detection on the difference image computed from satellite images of the same scene acquired at different time instances. Each pixel of the difference image is represented by a feature vector constructed from the difference image values of the neighbouring pixels to consider the contextual information. The feature vectors of the difference image are modelled as a GMM. The conditional posterior probabilities of changed and unchanged pixel classes are automatically estimated by partitioning GMM into two distributions by minimizing an objective function. Bayesian inferencing is then employed to segment the difference image into changed and unchanged classes by using the conditional posterior probability of each class. Change detection results are shown on real datasets.     

Assessment of multispectral, -temporal and -angular MODIS data for tree cover mapping in the tundra–taiga transition zone

The latitudinal tree cover gradient is an important characteristic of the tundra–taiga transition zone stretching around the northern hemisphere. Accurately mapped continuous tree cover fields would enable the depiction of forest extent over this ecotone, which is sensitive to climate change, natural disturbances and human activities. The objective of this study was to assess the explanatory power of multispectral, -temporal and -angular MODIS data to estimate tree cover at the regional scale in northernmost Finland. The standard MODIS BRDF/Albedo (MOD43B) data products at approximately 1 km resolution were used. The tree cover was estimated using generalized linear models (GLM), which were calibrated and evaluated by high resolution biotope inventory data. The benefit of coupling the multispectral, -temporal and -angular variables was assessed by variation partitioning. The predicted tree cover fields were also used for the forest–non-forest classification over a larger region and compared with the forest extent of Finnish CORINE land cover 2000 data set. The results show that multitemporal and -angular variables can increase the accuracy of the tree cover estimates and mapping of the forest extent in comparison to the peak of the growing season nadir-view multispectral data. The season of the data acquisition also affect the model performance, the late-spring and early-summer data being superior to mid- and late-summer data. Although the pure effect of the multiangular variables i.e. the parameters of the MODIS BRDF model and selected multiangular indices were relatively small in the models, the inclusion of these data increased the accuracy of the tree cover estimates in the mires in comparison to the peak of the growing season nadir-view multispectral data and multitemporal variables.     

Thermal comfort and clothing insulation of resting tent occupants at high altitude

Thirty-nine males and 18 females, in six groups, participated in six high altitude treks (each lasting 3–4 weeks and climbing up to 5500 m) in the Himalaya and Karakoram. Inverse relationships between mean overnight total insulation (sleeping bag plus clothing) and air temperature in tents were recorded for all treks. Average overnight thermal sensations varied little with air temperature as the subjects modified their clothing insulation to maintain thermal sensations warmer than ‘neutral’ for all treks. For combined treks, subjects adjusted their mean overnight total insulation up to 7 clo for thermal sensations of between 0 (‘neutral’) and +1 (‘slightly warm’) on average, measured on the standard seven-point thermal sensation scale developed for everyday low-altitude conditions. Very few subjects (3% of all daily responses, on average) reported ‘cool’ or ‘cold’ sensations. General tent discomfort increased with altitude suggesting that subjects interpreted tent comfort predominantly in terms of thermal outdoor conditions.     

The living in a lattice rule language

This paper presents a rule-based query language for an object-oriented database model. The database model supports complex objects, object identity, classes and types, and a class/type hierarchy. The instances are described by ‘object relations’ which are functions from a set of objects to value sets and other object sets. The rule language is based on object-terms which provide access to objects via the class hierarchy. Rules are divided into two classes: object-preserving rules manipulating existing objects (yielding a new ‘view’ on objects available in the object base) and object-generating rules creating new objects with properties derived from existing objects. The derived object sets are included in a class lattice. We give conditions for whether the instances of the ‘rules’ heads are ‘consistent’, i.e. represent object relations where the properties of the derived objects are functionally determined by the objects.     

Experiments with an adaptive Bayesian restoration method

This paper describes a Bayesian restoration method applied to two-dimensional measured images, whose detector response function is not completely known. The response function is assumed Gaussian with standard deviation depending on the estimate of the local density of the image. The convex hull of the K-nearest neighbours (KNN) of each ‘on’ pixel is used to compute the local density. The method has been tested on ‘sparse’ images, with and without noise background.     

Bioequivalence revisited: non-parametric analysis of two-period cross-over studies

Hauschke et al.'s non-parametric bioequivalence procedure for treatment effects and some aspects of computer implementation, among them Meineke and De Hey's algorithm, are explored. For studies with up to sixty subjects, a table of indices of the ranked intersubject-intergroup mean ratios or differences is given, to establish non-parametric 90% confidence intervals. It is shown that non-parametric analysis is not limited to treatment effects: it can also be applied to period and sequence effects. This extended procedure can be seen as the non-parametric analogue of analysis of variance on two-period cross-over studies. A FORTRAN program (BIOQNEW) incorporating Meineke and De Mey's algorithm is presented. This program provides non-parametric point estimates for treatment and period effects, 90% and 95% confidence intervals for test-versus-reference treatments, the 95% confidence interval for periods and a test on sequence effects, so that it can also be used for other than bioequivalence studies. BIOEQNEW can handle ratios (‘multiplicative model’) as well as differences (‘additive model’). It optionally provides the complete non-parametric posterior probability distribution for treatment ratios or differences, so that Schuirmann's ‘two one-sided tests procedure’ can also be performed in a non-parametric way.     

Detection of vegetation abundance change in the alpine tree line using multitemporal Landsat Thematic Mapper imagery

Detection of alpine tree line change using pixel-based approaches on medium spatial resolution imagery is challenging because of very slow tree sprawl without obvious boundaries. However, vegetation abundance or density in the tree line zones may change over time and such a change may be detected using subpixel-based approaches. In this research, a linear spectral mixture analysis (LSMA)-based approach was used to examine alpine tree line change in the Northern Tianshan Mountains located in Northwestern China. Landsat Thematic Mapper (TM) imagery was unmixed into three fraction images (i.e. green vegetation – GV, shade, and soil) using the LSMA approach. The GV and soil fractions at different years were used to examine vegetation abundance change based on samples in the alpine tree line. The results show that Picea schrenkiana abundance around the top of the forested area increased approximately by 18.6% between 1990 and 2010, but remained stable in the central forest region over this period. Juniperus sabina abundance around the top of the forested area, in the central scrub region, and at the top of the scrub region increased approximately by 19.3%, 8.2%, and 15.6%, respectively. The increased vegetation abundance and decreased soil abundance of both P. schrenkiana and J. sabina indicate vegetation sprawl in the alpine tree line between 1990 and 2010. This research will be valuable for better understanding the impacts of climate change on vegetation change in the alpine tree line of central Asia.     

The use of prior probabilities in maximum likelihood classification of remotely sensed data

The expected distribution of classes in a final classification map can be used to improve classification accuracies. Prior information is incorporated through the use of prior probabilities—that is, probabilities of occurrence of classes which are based on separate, independent knowledge concerning the area to be classified. The use of prior probabilities in a classification system is sufficiently versatile to allow (1) prior weighting of output classes based on their anticipated sizes; (2) the merging of continuously varying measurements (multispectral signatures) with discrete collateral information datasets (e.g., rock type, soil type); and (3) the construction of time-sequential classification systems in which an earlier classification modifies the outcome of a later one. The prior probabilities are incorporated by modifying the maximum likelihood decision rule employed in a Bayesian-type classifier to calculate a posteriori probabilities of class membership which are based not only on the resemblance of a pixel to the class signature, but also on the weight of the class which is estimated for the final output classification. In the merging of discrete collateral information with continuous spectral values into a single classification, a set of prior probabilities (weights) is estimated for each value which the discrete collateral variable may assume (e.g., each rock type or soil type). When maximum likelihood calculations are performed, the prior probabilities appropriate to the particular pixel are used in classification. For time-sequential classification, the prior classification of a pixel indexes a set of appropriate conditional probabilities reflecting either the confidence of the investigator in the prior classification or the extent to which the prior class identified is likely to change during the time period of interest.     

Learning hierarchical clustering from examples — application to the adaptive construction of dissimilarity indices

A classical problem of Pattern Recognition consists in looking for an operator of classification (a ‘classifier’) induced from a learning set on which classes are known. A problem frequently encountered in practice is that of looking for an operator of clustering (a ‘clusterfier’, in opposition to ‘classifier’) from a learning set on which clusters are also known. In the first case, we have to find an operator which allocates each new object to one of the classes defined by the learning set. In the second case, we have to find an operator which detects classes in the complete population, taking in account as well as possible the information given by the classes on the learning set. We propose a new approach permitting to induce and aggregation index from knowledge acquiring on the learning set; the aggregation index thus obtained permits to induce a hierarchy which infers the desired classes on the whole population.

A nearest neighbours algorith with validity constraints has been realized to induce the final hierarchy. We obtain a CPU time clearly shorter than with the classical hierarchical ascending classification algorithm which does not use inference.

This program has permitted to find aggregation indices adapted to particular learning sets (elongated classes, spherical class with central kernel, half spherical class with central kernel, noising elongated classes…), and some of new indices permit to recognize more specific classes than the usual indices.     

Using local transition probability models in Markov random fields for forest change detection

Change detection based on the comparison of independently classified images (i.e. post-classification comparison) is well-known to be negatively affected by classification errors of individual maps. Incorporating spatial-temporal contextual information in the classification helps to reduce the classification errors, thus improving change detection results. In this paper, spatial-temporal Markov Random Fields (MRF) models were used to integrate spatial-temporal information with spectral information for multi-temporal classification in an attempt to mitigate the impacts of classification errors on change detection. One important component in spatial-temporal MRF models is the specification of transition probabilities. Traditionally, a global transition probability model is used that assumes spatial stationarity of transition probabilities across an image scene, which may be invalid if areas have varying transition probabilities. By relaxing the stationarity assumption, we developed two local transition probability models to make the transition model locally adaptive to spatially varying transition probabilities. The first model called locally adjusted global transition model adapts to the local variation by multiplying a pixel-wise probability of change with the global transition model. The second model called pixel-wise transition model was developed as a fully local model based on the estimation of the pixel-wise joint probabilities. When applied to the forest change detection in Paraguay, the two local models showed significant improvements in the accuracy of identifying the change from forest to non-forest compared with traditional models. This indicates that the local transition probability models can present temporal information more accurately in change detection algorithms based on spatial-temporal classification of multi-temporal images. The comparison between the two local transition models showed that the fully local model better captured the spatial heterogeneity of the transition probabilities and achieved more stable and consistent results over different regions of a large image scene.     

Addition and removal of information for a knowledge base with incomplete information

A knowledge base containing incomplete information in the form of disjunctions and negative information shows difficulties regarding the update operators. In this paper simple and straightforward definitions are given for an ‘adding’ operator (‘+’) and a ‘removing’ operator (‘−’) using Hebrand models.     

CIPP — a versatile analytical model for VBR traffic

Correlated interarrival time Poisson process (CIPP) has been proposed in Proceedings of the Fifth Biennial Conference on Signal Processing and Communications (SPCOM’99), IISc, Bangalore, July 1999, pp. 43–50; J. Indian Inst. Sci. 79 (3) (1999) 233–249] for modeling both the composite arrival process of packets in broadband networks and the individual video source modeling. The CIPP — a generalization of the Poisson process — is a stationary counting process and is parameterized by correlation parameter ‘ρ’, the degree of correlation in adjacent interarrivals and ‘λ’, the intensity of the process. In this paper, we develop the theory for CIPP/M/1 queue and undertake the performance modeling of statistical multiplexer with VBR video traffic in broadband networks using the CIPP/M/1 queue. We first derive the expressions for stationary distributions for queue length and waiting time in a CIPP/M/1 queue. Then, we derive the queuing performance measures of interest. For reasons of feasibility of theoretical performance modeling and realistic compulsions, we propose a deterministic smoothing with random (geometrically distributed) packet sizes. We simulate a queue with (thus smoothed) VBR video trace data as input to compare with the theoretical performance measures derived above. Experimental results show that the CIPP/M/1 queue models well the statistical multiplexer performance with the real-world MPEG-1 VBR video traffic input.     

On the complexity of ω-type Turing acceptors

Turing machines are considered as recognizers of sets of infinite (ω-type) sequences, so called ω-languages. The basic results on such ω-type Turing acceptors were presented in a preceding paper. This paper focuses on the theory of deterministic ω-type Turing acceptors (ω-DTA's) which turns out to be crucially different from the ‘classical’ theory of Turing machines. It is shown that there exists no ω-DTA which is universal for all ω-DTA's. Two infinite complexity hierarchies for ω-DTA's are established, the ‘states hierarchy’, corresponding to the number of states in the machine, and the ‘designated sets hierarchy’, corresponding to the number of designated sets of states used in the recognition. Concrete examples of ω-languages characterizing each of the complexity classes are exhibited. Two additional examples of interesting ω-languages are presented:

1. (i) An ω-language which is ‘inherently non-deterministic’, i.e. can be recognized by a non-deterministic Turing acceptor but by no deterministic acceptor.

2. (ii) An ω-language which cannot be recognized even by a non-deterministic Turing acceptor.

The above examples are constructed without using diagonalization. Oscillating ω-DTA's, i.e. ω-DTA's which are allowed to oscillate on ω-inputs, are also considered and are shown to be strictly more powerful than non-oscillating ω-DTA's, yet strictly less powerful than non-deterministic ω-Turing acceptors.     

Improving global scale land cover classifications with multi-directional POLDER data and a decision tree classifier

Several investigations indicate that the Bidirectional Reflectance Distribution Function (BRDF) contains information that can be used to complement spectral information for improved land cover classification accuracies. Prior studies on the addition of BRDF information to improve land cover classifications have been conducted primarily at local or regional scales. Thus, the potential benefits of adding BRDF information to improve global to continental scale land cover classification have not yet been explored. Here we examine the impact of multidirectional global scale data from the first Polarization and Directionality of Earth Reflectances (POLDER) spacecraft instrument flown on the Advanced Earth Observing Satellite (ADEOS-1) platform on overall classification accuracy and per-class accuracies for 15 land cover categories specified by the International Geosphere Biosphere Programme (IGBP).

A set of 36,648 global training pixels (7 × 6 km spatial resolution) was used with a decision tree classifier to evaluate the performance of classifying POLDER data with and without the inclusion of BRDF information. BRDF ‘metrics’ for the eight-month POLDER on ADEOS-1 archive (10/1996–06/1997) were developed that describe the temporal evolution of the BRDF as captured by a semi-empirical BRDF model. The concept of BRDF ‘feature space’ is introduced and used to explore and exploit the bidirectional information content. The C5.0 decision tree classifier was applied with a boosting option, with the temporal metrics for spectral albedo as input for a first test, and with spectral albedo and BRDF metrics for a second test. Results were evaluated against 20 random subsets of the training data.

Examination of the BRDF feature space indicates that coarse scale BRDF coefficients from POLDER provide information on land cover that is different from the spectral and temporal information of the imagery. The contribution of BRDF information to reducing classification errors is also demonstrated: the addition of BRDF metrics reduces the mean, overall classification error rates by 3.15% (from 18.1% to 14.95% error) with larger improvements for producer's accuracies of individual classes such as Grasslands (+ 8.71%), Urban areas (+ 8.02%), and Wetlands (+ 7.82%). User's accuracies for the Urban (+ 7.42%) and Evergreen Broadleaf Forest (+ 6.70%) classes are also increased. The methodology and results are widely applicable to current multidirectional satellite data from the Multi-angle Imaging Spectroradiometer (MISR), and to the next generation of POLDER-like multi-directional instruments.     

A level set method based on the Bayesian risk for medical image segmentation

This paper proposes an alternative criterion derived from the Bayesian risk classification error for image segmentation. The proposed model introduces a region-based force determined through the difference of the posterior image densities for the different classes, a term based on the prior probability derived from Kullback-Leibler information number, and a regularity term adopted to avoid the generation of excessively irregular and small segmented regions. Compared with other level set methods, the proposed approach relies on the optimum decision of pixel classification and the estimates of prior probabilities; thus the approach has more reliability in theory and practice. Experiments show that the proposed approach is able to extract the complicated shapes of targets and robust for various types of medical images. Moreover, the algorithm can be easily extendable for multiphase segmentation.     

Global image feature correspondence under a multi-image network

This paper discusses a global image feature correspondence strategy under a multi-image network. The term multi-image network is used to describe such geometries, where every object point ‘node’ has more than two corresponding imaging rays associated with it. The term global image feature correspondence refers to all those image features that are matched simultaneously. The global image feature correspondence requires a global matching strategy without employing any one image as a fixed reference image. The primary characteristics of the work includes application of epipolar curve constraints, use of multi-ray triangulation residuals in object space, adoption of least-squares network optimisation and application of global quality control measures. The matching speed for object point determination in the reported multi-image network reconstruction implementation, in the case of four images, reaches 120 points per second using a Pentium-200 processor. A three-dimensional triangulation accuracy of close to 0.1 pixel is achieved.