The multiple-forward-mode (MFM) inversion procedure is a set of methods for indirect canopy relectance model inversion using look-up tables (LUT). This thesis refines the MFM technique with regard to: 1) model parameterization for the MFM canopy reflectance model executions and 2) methods for limiting or describing multiple solutions. Forest stand structure estimates from the inversion were evaluated using 40 field validation sites in the Canadian Rocky Mountains. Estimates of horizontal and vertical crown radius were within 0.5m and 0.9m RMSE for both conifer and deciduous species. Density estimates were within 590 stems/ha RMSE for conifer and 310 stems/ha RMSE for deciduous. The most effective inversion method used a variable spectral domain with constrained, fine increment LUTs. A biomass estimation method was also developed using empirical relationships with crown area. Biomass density estimates using the MFM method were similar to estimates produced using other multispectral analysis methods (RMSE=50t/ha). / xvi, 156 leaves : ill. (some col.), maps ; 29 cm.

The mapping of forest age is important for effective forest inventory as age is indicative of a number of plant physiological processes. Field survey techniques have traditionally been used to collect forest inventory data, but these methods are costly and time-consuming. Remote sensing offers an alternative which is time-effective and cost-effective and can cover large areas. The aim of this research was to assess the capabilities of multispectral and hyperspectral remotely sensed image data and the statistical method, random forest, for Pinus patula age prediction. The first section of this study used spatial and spectral data derived from multispectral QuickBird imagery to predict forest age. Five co-occurrence texture measures (variance, contrast, correlation, homogeneity, and dissimilarity) were calculated on QuickBird panchromatic imagery (0.6 m spatial resolution) using 12 moving window sizes. The spectral data was extracted from visible and near infrared (NIR) QuickBird imagery (2.4 m spatial resolution). Using the random forest ensemble, various methods of combining the spectral and texture variables were evaluated. The best model was achieved using backward variable selection which aims to find the fewest number of input bands while maintaining the highest predictive accuracy. Only five of the original 64 variables were used in the final model (R2 = 0.68). The second part of this study examined the utility of the random forest ensemble and AISA Eagle hyperspectral image data to predict P. patula age. Random forest was used to determine the optimal subset of hyperspectral bands that could predict P. patula age. Two sequential variable selection methods were tested: forward and backward variable selection. Although both methods resulted in the same root mean square error (3.097), the backward variable selection method was unable to significantly reduce the large hyperspectral dataset and selected 206 variables for the model. The forward variable selection method successfully reduced the large dataset to only nine optimal bands while maintaining the highest predictive accuracy from the hyperspectral dataset (R2 = 0.6). Overall, we concluded that (i) remotely sensed data can produce accurate models for P. patula age prediction, (ii) random forest is an effective tool for the combination of spectral and spatial multispectral data, (iii) random forest is an effective tool for variable selection of a high dimensional hyperspectral dataset, and (iv), although random forest has mainly been used as a classifier, it is also a very effective tool for prediction. / Thesis (M.Sc.)-University of KwaZulu-Natal, Pietermaritzburg, 2010.

The goals of this project were to: (1) combine the ARC/INFO Geographic Information System (GIS) software with the TIGER data files and tree inventory data files, (2) demonstrate GIS as a tool in street tree inventory management, (3) answer a management related question, specifically the identification of dead and hazardous trees within the city, with the use of the GIS tool, and (4) prepare an article based on the project for submission to the Journal of Arboriculture.Dead and hazardous trees located along a city street are a major accident liability to a city. It is vital, for both safety and aesthetic purposes, that a community know the location of dead and hazardous trees. As a management tool the GIS can utilize the inventory data to aid the urban forester in interpreting the urban forest by identifying these tree locations. Through the use of point-in-polygon analysis and choropleth maps, these specific management concerns can be highlighted throughout the city.The power to visually demonstrate certain parts of town with higher concentrations of work needed is one way that GIS can provide the management tools necessary for better care of our urban forests. / Department of Landscape Architecture

Thesis (M.S.)--Mississippi State University. Department of Forestry. / Title from title screen. Includes bibliographical references.

Thesis (M.S.)--Brigham Young University. Dept. of Electrical and Computer Engineering, 2007. / Includes bibliographical references (p. 63-65).

<p dir="ltr">The distribution of people and infrastructure in North American cities is not random, nor is it equitable. Minoritized communities often have less access to trees and forests and to the benefits that they provide. This pattern has developed over centuries. North American cities were built on indigenous land, and the location of rivers, wetlands, forests, and prairies influenced where houses were built, factories erected, parks sited, and where natural areas were preserved. As people moved to cities, wealthier individuals could afford to buy homes in places that they found more desirable. These places often were near parks or lakes and often had more trees. Less wealthy people could only afford homes in places that had less green infrastructure and places that were nearer to environmental harms like polluting industries. Unfortunately, in the United States, wealth is associated with race, meaning that from an early time its cities were segregated. As cities expanded and migration from the Jim Crow South and immigration from Eastern Europe increased, these wealthier, whiter communities restricted non-white people from moving into them with racist covenants and through redlining, deepening segregation. In this dissertation, I explored how this ecological and social history affected the ecology of a metropolitan region’s forests, and how that related to who gets to experience them. I also showed how the associations between socio-demographic variables and tree canopy varied across an urban to exurban region. Finally, I identified barriers to requesting and planting trees that residents experience and strategies that may help governments and strategies that may help overcome them. I found that communities of color had less access to forests that have a higher capacity to provide ecosystem services; that while segregation existed in urban, suburban, and exurban communities, canopy inequity did not; and that managing existing trees and working with neighborhood associations is the best way to encourage planting and stewardship of trees that can reduce tree equity.</p><p><br></p>

Terrestrial laser scanning (TLS) may provide a way to increase timber value recovery by replacing manual timber cruising with a simple-to-use, cost-effective alternative. TLS has been studied in several trials worldwide. Past studies have not compared TLS based estimates with mill estimates of stem value and volume. Three differently stocked stands of hybrid poplar were selected for diameter, stem sinuosity and height measurement using manual cruising and TLS. Selected trees were harvested and transported to a mill where they were scanned and then processed into lumber and chips. Data gathered using both manual and TLS methods were used to obtain stem volume and value estimates to compare with mill estimates. Results indicated that TLS diameter measurements were more accurately matched to mill and manual measurements up to about 7.5 meters on the stem than above 7.5 meters on the stem in all three stands. Stem curvature comparisons indicated that the variation between TLS and mill centerline measurements was similar to the variation between repeat mill scan measurements of the same stems. Using TLS as a pre-harvest inventory tool showed that additional revenue could be obtained from the reallocation of saw-log and chip log volume to veneer logs of various sizes in all three stands. It was also shown that the sampling error required to estimate stand value was greater than was required to estimate stand volume within the same error limits. / Graduation date: 2012

Lidar is able to provide height and cover information which can be used to estimate selected forest attributes precisely. However, for users to evaluate whether the additional cost and complication associated with using Lidar merits adoption requires that the protocol to use lidar be thoroughly described and that a basis for selection of design parameters such as number of field plots and lidar pulse density be described. In our first analysis, we examine these issues by looking at the effects of pulse density and sample size on estimation when wall-to-wall lidar is used with a regression estimator. The effects were explored using resampling simulations. We examine both the effects on precision, and on the validity of inference. Pulse density had almost no effect on precision for the range examined, from 3 to .0625 pulses / m��. The effect of sample size on estimator precision was roughly in accordance with the behavior indicated by the variance estimator, except that for small samples the variance estimator had positive bias (the variance estimates were too small), compromising the validity of inference. In future analyses we plan to provide further context for wall-to-wall lidar-assisted estimation. While there is a lot of literature on modeling, there is limited information on how lidar-assisted approaches compare to existing methods, and what variables can or cannot be acquired, or may be acquired with reduced confidence. We expand our investigation of estimation in our second analysis by examining lidar obtained in a sampling mode in combination with Landsat. In this case we make inference about the feasibility of a lidar-assisted estimation strategy by contrasting its variance estimate with variance estimates from a variety of other sampling designs and estimators. Of key interest was how the precision of a two-stage estimator with lidar strips compared with a plot-only estimator from a simple random sampling design. We found that because the long and narrow lidar strips incorporate much of the landscape variability, if the number of lidar strips was increased from 7 to 15 strips, the precision of estimators with lidar can exceed that of estimators applied to plot-only SRS data for a much larger number of plots. Increasing the number of lidar strips is considered to be highly viable since the costs of field plots can be quite expensive in Alaska, often exceeding the cost of a lidar strip. A Landsat-assisted approach used for either an SRS or a two-stage sample was also found to perform well relative to estimators for plot-only SRS data. This proved beneficial when we combined lidar and Landsat-assisted regression estimators for two-stage designs using a composite estimator. The composite estimator yielded much better results than either estimator used alone. We did not assess the effects of changing the number of lidar strips in combination with using a composite estimator, but this is an important analysis we plan to perform in a future study. In our final analysis we leverage the synergy between lidar and Landsat to improve the explanatory power of auxiliary Landsat using a multilevel modeling strategy. We also incorporate a more sophisticated approach to processing Landsat which reflects temporal trends in individual pixels values. Our approach used lidar as an intermediary step to better match the spatial resolution of Landsat and increase the proportion of area overlapped between measurement units for the different sources of data. We developed two separate approaches for two different resolutions of data (30 m and 90 m) using multiple modeling alternatives including OLS and k nearest neighbors (KNN), and found that both resolution and the modeling approach affected estimates of residual variability, although there was no combination of model types which was a clear winner for all responses. The modeling strategies generally fared better for the 90 m approaches, and future analyses will examine a broader range of resolutions. Fortunately the approaches used are fairly flexible and there is nothing prohibiting a 1000 m implementation. In the future we also plan to look at using a more sophisticated Landsat time-series approach. The current approach essentially dampened the noise in the temporal trend for a pixel, but did not make use of information in the trend such as slope or indications of disturbance ��� which may provide additional explanatory power. In a future study we will also incorporate a multilevel modeling into estimation or mapping strategies and evaluate the contribution of the multilevel modeling strategy relative to alternate approaches. / Graduation date: 2013 / Access restricted to the OSU Community at author's request from June 21, 2012 - Dec. 21, 2012

Plantation forestry involves the management of man-made industrial forests for the purpose of producing raw materials for the pulp and paper, saw milling and other related wood products industries. Management of these forests is based on the cycle of planting, tending and felling of forest stands such that a sustainable operation is maintained. The monitoring and reporting of these forestry operations is critical to the successful management of the forestry industry. The aim of this study was to test whether the forestry operations of clear-felling, re-establishment and weed control could be qualitatively and quantitatively monitored through the application of classification and change detection techniques to multi-temporal medium (15-30 m) and a combination of textural analysis and change detection techniques on high resolution (0.6-2.4 m) satellite imagery. For the medium resolution imagery, four Landsat 7 multi-spectral images covering the period from March 2002 to April 2003 were obtained over the midlands of KwaZulu-Natal, South Africa, and a supervised classification, based on the Maximum Likelihood classifier, as well as two unsupervised classification routines were applied to each of these images. The supervised classification routine used 12 classes identified from ground-truthing data, while the unsupervised classification was done using 10 and 4 classes. NDVI was also calculated and used to estimate vegetation status. Three change detection techniques were applied to the unsupervised classification images, in order to determine where clear-felling, planting and weed control operations had occurred. An Assisted "Classified" Image change detection technique was applied to the Ten-Class Unsupervised Classification images, while an Assisted "Quantified Classified" change detection technique was applied to the Four-Class Unsupervised Classification images. An Image differencing technique was applied to the NDVI images. For the high resolution imagery, a series of QuickBird images of a plantation forestry site were used and a combination of textural analysis and change detection techniques was tested to quantify weed development in replanted forest stands less than 24 months old. This was achieved by doing an unsupervised classification on the multi-spectral bands, and an edge-enhancement on the panchromatic band. Both the resultant datasets were then vectorised, unioned and a matrix derived to determine areas of high weed. It was found that clear-felling operations could be identified with accuracy in excess of 95%. However, using medium resolution imagery, newly planted areas and the weed status of forest stands were not definitively identified as the spatial resolution was too coarse to separate weed growth from tree stands. Planted stands younger than one year tended to be classified in the same class as bare ground or ground covered with dead branches and leaves, even if weeds were present. Stands older than one year tended to be classified together in the same class as weedy stands, even where weeds were not present. The NDVI results indicated that further research into this aspect could provide more useful information regarding the identification of weed status in forest stands. Using the multi-spectral bands of the high resolution imagery it was possible to identify areas of strong vegetation, while crop rows were identifiable on the panchromatic band. By combining these two attributes, areas of high weed growth could be identified. By applying a post-classification change detection technique on the high weed growth classes, it was possible to identify and quantify areas of weed increase or decrease between consecutive images. A theoretical canopy model was also derived to test whether it could identify thresholds from which weed infestations could be determined. The conclusions of this study indicated that medium resolution imagery was successful in accurately identifying clear-felled stands, but the high resolution imagery was required to identify replanted stands, and the weed status of those stands. However, in addition to identifying the status of these stands, it was also possible to quantify the level of weed infestation. Only wattle (Acacia mearnsii) stands were tested in this manner but it was recommended that in addition to applying these procedures to wattle stands, they also are tested in Eucalyptus and Pinus stands. The combination of textural analysis on the panchromatic band and classification of multi-spectral bands was found to be a suitable process to achieve the aims of this study, and as such were recommended as standard procedures that could be applied in an operational plantation forest monitoring environment. / Thesis (Ph.D.)-University of KwaZulu-Natal, Pietermaritzburg, 2006.

Ma Fung-yan. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1999. / Includes bibliographical references (leaves 137-141). / Abstracts in English and Chinese. / Abstract --- p.i / Acknowledgements --- p.iv / Table of Contents --- p.v / List of Tables --- p.ix / List of Figures --- p.x / Chapter CHAPTER 1 --- INTRODUCTION / Chapter 1.1 --- Introduction and background --- p.1 / Chapter 1.2 --- Objectives --- p.4 / Chapter 1.3 --- Significance --- p.5 / Chapter 1.4 --- Organization of the thesis --- p.5 / Chapter CHAPTER 2 --- LITERATURE REVIEW / Chapter 2.1 --- Introduction --- p.7 / Chapter 2.2 --- Hyperspectral remote sensing --- p.7 / Chapter 2.2.1 --- Current imaging spectrometers available --- p.8 / Chapter 2.2.2 --- Applications of hyperspectral remote sensing --- p.9 / Chapter 2.2.2.1 --- Biochemistry of vegetation --- p.10 / Chapter 2.2.2.2 --- Spatial and temporal patterns of vegetation --- p.12 / Chapter 2.3 --- Tree species recognition --- p.12 / Chapter 2.3.1 --- Factors affecting spectral reflectance of vegetation --- p.14 / Chapter 2.3.1.1 --- Optical properties of leaf --- p.14 / Chapter 2.3.1.2 --- Canopy structure --- p.15 / Chapter 2.3.1.3 --- Canopy cover --- p.16 / Chapter 2.3.1.4 --- Illumination and viewing geometry --- p.16 / Chapter 2.3.1.5 --- Spatial and temporal dynamics of plants --- p.17 / Chapter 2.3.2 --- Classification algorithms for hyperspectral analysis --- p.17 / Chapter 2.3.2.1 --- Use of derivative spectra for tree species recognition --- p.17 / Chapter 2.3.2.2 --- Linear discriminant analysis --- p.18 / Chapter 2.3.2.3 --- Artificial neural network --- p.19 / Chapter 2.3.3 --- Tree species recognition using hyperspectral data --- p.21 / Chapter 2.4 --- Data compression and feature extraction --- p.22 / Chapter 2.4.1 --- Analytical techniques of data compression --- p.23 / Chapter 2.4.2 --- Analytical techniques of feature extraction --- p.25 / Chapter 2.4.2.1 --- Feature selection by correlation with biochemical and biophysical data --- p.25 / Chapter 2.4.2.2 --- Spatial autocorrelation-based feature selection --- p.27 / Chapter 2.4.2.3 --- Spectral autocorrelation-based feature selection --- p.29 / Chapter 2.4.2.3.1 --- Optimization with distance metrics --- p.29 / Chapter 2.4.2.3.2 --- Stepwise linear discriminant analysis --- p.30 / Chapter 2.5 --- Summary --- p.31 / Chapter CHAPTER 3 --- METHODOLOGY / Chapter 3.1 --- Introduction --- p.33 / Chapter 3.2 --- Study site --- p.33 / Chapter 3.3 --- Instrumentation --- p.34 / Chapter 3.4 --- Data collection --- p.35 / Chapter 3.4.1 --- Laboratory measurement --- p.36 / Chapter 3.4.2 --- In situ measurement --- p.39 / Chapter 3.5 --- Methods of data analysis --- p.40 / Chapter 3.5.1 --- Preprocessing of data --- p.40 / Chapter 3.5.2 --- Compilation of hyperspectral database --- p.42 / Chapter 3.5.3 --- Tree species recognition --- p.42 / Chapter 3.5.3.1 --- Linear discriminant analysis --- p.44 / Chapter 3.5.3.2 --- Artificial neural network --- p.44 / Chapter 3.5.3.3 --- Accuracy assessment --- p.45 / Chapter 3.5.3.4 --- Comparison of different data processing strategies and classifiers --- p.45 / Chapter 3.5.3.5 --- Comparison of data among different seasons --- p.46 / Chapter 3.5.3.6 --- Comparison of laboratory and in situ data --- p.46 / Chapter 3.5.4 --- Data compression --- p.47 / Chapter 3.5.5 --- Band selection --- p.47 / Chapter 3.6 --- Summary --- p.48 / Chapter CHAPTER 4 --- RESULTS AND DISCUSSIONS OF TREE SPECIES RECOGNITION / Chapter 4.1 --- Introduction --- p.50 / Chapter 4.2 --- Characteristics of hyperspectral data --- p.50 / Chapter 4.3 --- Tree species recognition --- p.79 / Chapter 4.3.1 --- Comparison of different classifiers --- p.82 / Chapter 4.3.1.1 --- Efficiency of the classifiers --- p.83 / Chapter 4.3.1.2 --- Discussions --- p.83 / Chapter 4.3.2 --- Comparison of different data processing strategies --- p.84 / Chapter 4.3.3 --- Comparison of data among different seasons --- p.86 / Chapter 4.3.4 --- Comparison of laboratory and in situ data --- p.88 / Chapter 4.4 --- Summary --- p.92 / Chapter CHAPTER 5 --- RESULTS AND DISCUSSIONS OF DATA COMPRESSION AND BAND SELECTION / Chapter 5.1 --- Introduction --- p.93 / Chapter 5.2 --- Data compression --- p.93 / Chapter 5.2.1 --- PCA using in situ spectral data --- p.93 / Chapter 5.2.1.1 --- Characteristics of PC loadings --- p.95 / Chapter 5.2.1.2 --- Scatter plots of PC scores --- p.96 / Chapter 5.2.2 --- PCA using laboratory spectral data --- p.99 / Chapter 5.2.2.1 --- Characteristics of PC loadings --- p.102 / Chapter 5.2.2.2 --- Scatter plots of PC scores --- p.103 / Chapter 5.2.2.3 --- Results of tree species recognition using PC scores --- p.107 / Chapter 5.2.3 --- Implications --- p.107 / Chapter 5.3 --- Band selection --- p.108 / Chapter 5.3.1 --- Preliminary band selection using stepwise discriminant analysis --- p.108 / Chapter 5.3.1.1 --- Selection of spectral bands --- p.109 / Chapter 5.3.1.2 --- Classification results of the selected bands --- p.109 / Chapter 5.3.1.3 --- Seasonal comparison using stepwise linear discriminant analysis --- p.114 / Chapter 5.3.1.4 --- Implications --- p.116 / Chapter 5.3.2 --- Band selection using hierarchical clustering technique --- p.116 / Chapter 5.3.2.1 --- Hierarchical clustering procedure --- p.116 / Chapter 5.3.2.2 --- Selection of spectral band sets --- p.119 / Chapter 5.3.2.3 --- Classification results of the selected band sets --- p.124 / Chapter 5.4 --- Summary --- p.127 / Chapter CHAPTER 6 --- SUMMARY AND CONCLUSION / Chapter 6.1 --- Introduction --- p.129 / Chapter 6.2 --- Summary --- p.129 / Chapter 6.2.1 --- Tree species recognition --- p.129 / Chapter 6.2.2 --- Data compression --- p.130 / Chapter 6.2.3 --- Band selection --- p.131 / Chapter 6.3 --- Limitations of this study --- p.132 / Chapter 6.4 --- Recommendations for further studies --- p.133 / Chapter 6.5 --- Conclusion --- p.136 / BIBLIOGRAPHY --- p.137 / APPENDICES / Appendix 1 Reflectance of the 25 tree species in four seasons with three levels of leaf density --- p.142-166 / "Appendix 2 Confusion matrices of tree species recognition using original spectra, first derivatives spectra and second derivatives spectra with 138 bands classified by linear discriminant analysis for each season" --- p.167-178 / "Appendix 3 Confusion matrices of tree species recognition using original spectra, first derivatives spectra and second derivatives spectra with 138 bands classified by neural networks for each season" --- p.179-190 / Appendix 4 Confusion matrices of tree species recognition using 21 tree species with original spectra classified by linear discriminant analysis for seasonal comparison --- p.191-193 / Appendix 5 Confusion matrices of tree species recognition using the first eight PC scores classified by linear discriminant analysis for each season --- p.194-197 / "Appendix 6 Confusion matrices of tree species recognition using original spectra, first derivatives spectra and second derivatives spectra classified by stepwise linear discriminant analysis (Case 2) for each season" --- p.198-209 / "Appendix 7 Confusion matrices of tree species recognition using original spectra, first derivatives spectra and second derivatives spectra classified by stepwise linear discriminant analysis (Case 3) for each season" --- p.210-220 / "Appendix 8 Confusion matrices of tree species recognition using 21 tree species with original spectra, first derivatives spectra and second derivatives spectra classified by stepwise linear discriminant analysis for seasonal comparison" --- p.221-229 / Appendix 9 Confusion matrices of tree species recognition using the spectral bands selected by hierarchical clustering procedures and classified by linear discriminant analysis for each season --- p.230-257