Statistics for Spatial Data.

1. Introduction and Overview. 2. Detecting Influential Observations and Outliers. 3. Detecting and Assessing Collinearity. 4. Applications and Remedies. 5. Research Issues and Directions for Extensions. Bibliography. Author Index. Subject Index.

Regression Diagnostics: Identifying Influential Data and Sources of Collinearity

Climate change 2014 - Synthesis Report

We evaluate 179 classifiers arising from 17 families (discriminant analysis, Bayesian, neural networks, support vector machines, decision trees, rule-based classifiers, boosting, bagging, stacking, random forests and other ensembles, generalized linear models, nearest-neighbors, partial least squares and principal component regression, logistic and multinomial regression, multiple adaptive regression splines and other methods), implemented in Weka, R (with and without the caret package), C and Matlab, including all the relevant classifiers available today. We use 121 data sets, which represent the whole UCI data base (excluding the large-scale problems) and other own real problems, in order to achieve significant conclusions about the classifier behavior, not dependent on the data set collection. The classifiers most likely to be the bests are the random forest (RF) versions, the best of which (implemented in R and accessed via caret) achieves 94.1% of the maximum accuracy overcoming 90% in the 84.3% of the data sets. However, the difference is not statistically significant with the second best, the SVM with Gaussian kernel implemented in C using LibSVM, which achieves 92.3% of the maximum accuracy. A few models are clearly better than the remaining ones: random forest, SVM with Gaussian and polynomial kernels, extreme learning machine with Gaussian kernel, C5.0 and avNNet (a committee of multi-layer perceptrons implemented in R with the caret package). The random forest is clearly the best family of classifiers (3 out of 5 bests classifiers are RF), followed by SVM (4 classifiers in the top-10), neural networks and boosting ensembles (5 and 3 members in the top-20, respectively).

/pdf/do-we-need-hundreds-of-classifiers-to-solve-real-world-1igtahh9y9.pdf

Do we need hundreds of classifiers to solve real world classification problems

A wide range of methods for analysis of airborne- and satellite-derived imagery continues to be proposed and assessed. In this paper, we review remote sensing implementations of support vector machines (SVMs), a promising machine learning methodology. This review is timely due to the exponentially increasing number of works published in recent years. SVMs are particularly appealing in the remote sensing field due to their ability to generalize well even with limited training samples, a common limitation for remote sensing applications. However, they also suffer from parameter assignment issues that can significantly affect obtained results. A summary of empirical results is provided for various applications of over one hundred published works (as of April, 2010). It is our hope that this survey will provide guidelines for future applications of SVMs and possible areas of algorithm enhancement.

https://www.aboutgis.com/Publications/Mountrakis_SVM_review_in_remote_sensing_ISPRS2010.pdf

Support vector machines in remote sensing: A review

Airborne Laser Scanning (ALS) data hold a great deal of promise in monitoring the reduction of single trees and forests with high accuracy. In the literature, the canopy height model (CHM) is the main input used frequently for forest change detection. ALS also has the key capability of delivering 3D point clouds, not only from the top canopy surface, but also from the entire canopy profile and also from the terrain. We investigated the use of two additional parameters, which exploit these capabilities for assessing the reduction of wooded area: Slope-adapted echo ratio (sER) and Sigma0. In this study, two ALS point cloud data sets (2005 and 2011) were used to calculate Digital Surface Model (DSM), sER, and Sigma0 in 1.5 km2 forest area in Vorarlberg, Austria. Image differencing was applied to indicate the change in the three difference models individually and in their combinations. Decision trees were used to classify the area of removed trees with the minimum mapping unit of 13 m2. The final results were evaluated by a knowledge-based manual digitization using completeness and correctness measures. The best result is achieved using the combination of sER and DSM, namely a correctness of 92% and a completeness of 85%.

Assessment of Wooded Area Reduction by Airborne Laser Scanning

Airborne laser scanning (ALS) and image matching are the two main techniques for generating point clouds for large areas. While the classification of ALS point clouds has been well investigated, th...

Classification of image matching point clouds over an urban area

. The amount of dead wood is a significant parameter for the description and assessment of forest habitat quality under the terms of the Habitats directive and Natura 2000 guidelines. EU member states are obliged by the Natura 2000 regulations to report on habitat quality in a regular interval of six years. To fulfil this task, the areas should be surveyed in the field, which requires an enormous amount of workload if done only by conventional field work. In this study the applicability of airborne laser scanning data as the single data source for the detection of downed trees in forest habitats is investigated. A focus is laid on the comparison of point clouds with only discrete (XYZ) and full-waveform (including echo width) information as input data. In our paper we present an automatic workflow which is able to detect downed trees with high completeness for both data sets (77.8% for discrete and 75.6% for full-waveform data). Due to large amount of false positives, the correctness using discrete ALS data is poorer (63.1%) than for full-waveform data (89.9%). It was found that the quality of the result is also influenced by factors such as dimension, state of decay, vegetation density and penetration of the foliage by the laser.

https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/II-5-W2/199/2013/isprsannals-II-5-W2-199-2013.pdf

Comparison of discrete and full-waveform ALS for dead wood detection

We propose a flexible framework for automated forest patch delineations that exploits a set of canopy structure features computed from airborne laser scanning (ALS) point clouds. The approach is based on an iterative subdivision of the point cloud using k-means clustering followed by an iterative merging step to tackle oversegmentation. The framework can be adapted for different applications by selecting relevant input features that best measure the intended homogeneity. In our study, the performance of the segmentation framework was tested for the delineation of forest patches with a homogeneous canopy height structure on the one hand and with similar water cycle conditions on the other. For the latter delineation, canopy components that impact interception and evapotranspiration were used, and the delineation was mainly driven by leaf area, tree functional type, and foliage density. The framework was further tested on two scenes covering a variety of forest conditions and topographies. We demonstrate that the delineated patches capture well the spatial distributions of relevant canopy features that are used for defining the homogeneity. The consistencies range from R 2 = 0.84 to R 2 = 0.86 and from R 2 = 0.80 to R 2 = 0.91 for the most relevant features in the delineation of patches with similar height structure and water cycle conditions, respectively.

Adaptive Framework for the Delineation of Homogeneous Forest Areas Based on LiDAR Points

The convergence diameter of shield tunnels is detected by ellipse fitting or local curve fitting to cross-section points. However, the tunnel section, which is extruded by an external force, has an irregular elliptical shape, and the waist of the tunnel is often blocked by accessories, resulting in data loss. This study proposes a convergence diameter and radial dislocation detection method based on block-level fitting. The proposed method solves the accuracy degradation caused by the model error and point cloud incompletion. First, the noise points in the tunnel section point cloud are removed using the least trimmed squares method. Second, the tunnel transverse seam is then located using the image edge detection algorithm. Third, the endpoint of the convergence diameter is determined by making a specific segment the center and shifting the detector from the center to the pinpoint. Finally, the convergence diameter and radial dislocation are detected by the endpoints of the segments. The experimental results showed that the absolute detection accuracy of this method was better than 3 mm, and the repeated detection accuracy was better than 2 mm. This result is consistent with prior total station measurements, which are more suitable for practical engineering applications.

/pdf/shield-tunnel-convergence-diameter-detection-based-on-self-f6duxvcu.pdf

Norbert Pfeifer

Papers

Assessment of Wooded Area Reduction by Airborne Laser Scanning

Classification of image matching point clouds over an urban area

Comparison of discrete and full-waveform ALS for dead wood detection

Adaptive Framework for the Delineation of Homogeneous Forest Areas Based on LiDAR Points

Shield Tunnel Convergence Diameter Detection Based on Self-Driven Mobile Laser Scanning