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

Recent developments in sensor technology yielded a major progress in airborne laser bathymetry for capturing shallow water bodies. Modern topo-bathymetric small foot print laser scanners do no longer use the primary near infrared (NIR) signal (=1064 nm) but only emit and receive the frequency doubled green signal (λ = 532 nm). For calculating correct water depths accurate knowledge of the water surface (air-water-interface) is mandatory for obtaining accurate spot positions and water depths. Due to the ability of the green signal to penetrate water the first reflections do not exactly represent the water surface but, depending on environmental parameters like turbidity, a certain penetration into the water column can be observed. This raises the question if it is even feasible to determine correct water level heights from the green laser echoes only. In this article, therefore, the near water surface penetration properties of the green laser signal are analyzed based on a test flight of the River Pielach (Austria) carried out with Riegl's VQ-820-G (532 nm) and VQ-580 (1064 nm) scanners mounted on the same airborne platform. It is shown that within the study area the mean penetration into the water column is in the range of 10–25 cm compared to the NIR signal as reference. However, as the upper hull of the green water surface echoes coincides with the NIR signal with cm-precision, it is still possible to derive water surface models from the green laser echoes only via statistical analysis of aggregated neighboring echoes and robustly keep the underestimation of the water level below 6 cm. This especially holds for still and stationary flowing water bodies.

/pdf/analyzing-near-water-surface-penetration-in-laser-bathymetry-52nn9bqdo9.pdf

Analyzing near water surface penetration in laser bathymetry – A case study at the River Pielach

This paper suggests a new approach for change detection (CD) in 3D point clouds. It combines classification and CD in one step using machine learning. The point cloud data of both epochs are merged for computing features of four types: features describing the point distribution, a feature relating to relative terrain elevation, features specific for the multi-target capability of laser scanning, and features combining the point clouds of both epochs to identify the change. All these features are merged in the points and then training samples are acquired to create the model for supervised classification, which is then applied to the whole study area. The final results reach an overall accuracy of over 90% for both epochs of eight classes: lost tree, new tree, lost building, new building, changed ground, unchanged building, unchanged tree, and unchanged ground.

Integrated Change Detection and Classification in Urban Areas Based on Airborne Laser Scanning Point Clouds.

Since the mid-1990s, the Institute of Photogrammetry and Remote Sensing (I.P.F.) is engaged in Airborne Laser Scanning (ALS) in research and development. Scientific contributions have been made in a wide field of related topics like full waveform signal analysis, georeferencing and filtering of ALS point clouds, automatic breakline modelling, DTM generation, quality control, etc. Apart from that, converting research ideas into software solutions is an enduring tradition at the I.P.F. for which the DTM program SCOP++ is an example. Partial solutions of ALS-related issues have been implemented in SCOP++, but a complete processing chain is missing, as the development cycles for this highly interactive program are long. Thus, the objectives of the new OPALS program system are to provide a complete processing chain for large ALS projects and to shorten development cycles significantly. OPALS is designed as a collection of small well-defined modules which can be accessed in three different ways: (i) from DOS/Unix shells as executables, (ii) from Python shells as full-featured, platform-independent Python modules or (iii) from custom C++ programs by dynamic linkage (DLL) for fastest module calls. Sophisticated custom processing chains can be established by freely combining the OPALS modules using shell or Python scripts. To reduce development times, a lightweight framework is introduced. It allows non-expert programmers to implement their own modules, concentrating on the implementation of their latest research outcomes, whereas the framework deals with general issues like validation of user inputs, error handling, logging, etc. In this way, new research outcomes get available more rapidly for the scientific community. OPALS does not only target researchers, but also ALS service providers dealing with large ALS projects. Efficient data handling is a precondition for this purpose. Thus, the OPALS data manager (ODM) is one of the core units, allowing administration of data volumes in the order of 10 9 points. The ODM acts as spatial cache and provides high-performance spatial queries. Currently, a quality control package (opalsQC) is in progress and first results (point density maps, strip difference maps, 3D-strip shifts) are presented in the paper.

https://publik.tuwien.ac.at/files/PubDat_177262.pdf

Orientation and processing of airborne laser scanning data (opals) - concept and first results of a comprehensive als software

Diameter at breast height (DBH) is one of the most important parameter in forestry. With increasing use of terrestrial and airborne laser scanning in forestry, new exceeding possibilities to directly derive DBH emerge. In particular, high resolution point clouds from laser scanners on board unmanned aerial systems (UAS) are becoming available over forest areas. In this case study, DBH estimation from a UAS point cloud based on modeling the relevant part of the tree stem with a cylinder, is analyzed with respect to accuracy and completeness. As reference, manually measured DBHs and DBHs from terrestrial laser scanning point clouds are used for comparison. We demonstrate that accuracy and completeness of the cylinder fit are depending on the stem diameter. Stems with DBH > 20 cm feature almost 100% successful reconstruction with relative differences to the reference DBH of 9% (DBH 20–30 cm) down to 1.8% for DBH > 40 cm.

A Case Study of UAS Borne Laser Scanning for Measurement of Tree Stem Diameter

Outlining patches dominated by different plants in wetland vegetation provides information on species succession, microhabitat patterns, wetland health and ecosystem services. Aerial photogrammetry and hyperspectral imaging are the usual data acquisition methods but the application of airborne laser scanning (ALS) as a standalone tool also holds promises for this field since it can be used to quantify 3-dimensional vegetation structure. Lake Balaton is a large shallow lake in western Hungary with shore wetlands that have been in decline since the 1970s. In August 2010, an ALS survey of the shores of Lake Balaton was completed with 1 pt/m2 discrete echo recording. The resulting ALS dataset was processed to several output rasters describing vegetation and terrain properties, creating a sufficient number of independent variables for each raster cell to allow for basic multivariate classification. An expert-generated decision tree algorithm was applied to outline wetland areas, and within these, patches dominated by Typha sp. Carex sp., and Phragmites australis. Reed health was mapped into four categories: healthy, stressed, ruderal and die-back. The output map was tested against a set of 775 geo-tagged ground photographs and had a user’s accuracy of > 97% for detecting non-wetland features (trees, artificial surfaces and low density Scirpus stands), > 72% for dominant genus detection and > 80% for most reed health categories (with 62% for one category). Overall classification accuracy was 82.5%, Cohen’s Kappa 0.80, which is similar to some hyperspectral or multispectral-ALS fusion studies. Compared to hyperspectral imaging, the processing chain of ALS can be automated in a similar way but relies directly on differences in vegetation structure and actively sensed reflectance and is thus probably more robust. The data acquisition parameters are similar to the national surveys of several European countries, suggesting that these existing datasets could be used for vegetation mapping and monitoring.

Norbert Pfeifer

Papers

Analyzing near water surface penetration in laser bathymetry – A case study at the River Pielach

Integrated Change Detection and Classification in Urban Areas Based on Airborne Laser Scanning Point Clouds.

Orientation and processing of airborne laser scanning data (opals) - concept and first results of a comprehensive als software

A Case Study of UAS Borne Laser Scanning for Measurement of Tree Stem Diameter

Categorizing Wetland Vegetation by Airborne Laser Scanning on Lake Balaton and Kis-Balaton, Hungary