Journal of National Taipei Teachers College, Vol.17, No.1 (Mar. 2004) 173~182 NATIONAL TAIPEI TEACHERS COLLEGE
173
A Principal Component Analysis of Vegetation Characteristics in North America
Shaw-wen Sheen*
ABSTRACT
Vegetation cover is an important variable in many Earth system processes. The aim of this research is to analyze spatial variations in vegetation characteristics of North America using principal component analysis. The study area is from 120°W to 50°W and 10°N to 70°N. The results show that the principal components 1, 2, and 3 explain 73%, 25%, and 2% of the variance in the original 4 vegetation maps (evergreen, deciduous, broadleaf, and needleleaf maps). Component 1 is highly correlated with needleleaf and evergreen vegetation. Component 2 is highly correlated with broadleaf and deciduous vegetation. This research used cluster analysis to generate 11 North American vegetation clusters.
Key words: Vegetation cover, Principal Component Analysis, Cluster Analysis, North America.
*
Shaw-wen Sheen: Assistant Professor, Department of Social Studies Education, National Tainan Teachers College
174
Journal of National Taipei Teachers College, Vol.17, No.1 (Mar. 2004) 173~182 NATIONAL TAIPEI TEACHERS COLLEGE
A Principal Component Analysis of Vegetation Characteristics in North America
Shaw-wen Sheen*
INTRODUCTION
Vegetation cover is an important variable in many Earth system processes (Hansen, DeFries, Townshend, & Sholberg, 2000). Sabins (1996) indicated that vegetation, both native and cultivated, covers much of the earth and strongly influences the environment. Until recently adequate data were lacking for mapping the composition, concentration, and dynamics of the world’s vegetation. Sabins (1996) stated that remote sensing from satellites provide our first opportunity to inventory the surface resources of the earth in a systematic repetitive manner. Campbell (2002) indicated that remote sensing provides the only practical means of mapping and monitoring changes in major ecological regions that, although not directly used for production of food, have great long-term significance for mankind. Campbell (2002) stated that the tropical forests that cover significant areas of the earth’s surface have never been mapped or studied except in local regions. Yet these regions are of critical importance to mankind due to their role in maintaining the earth’s climate and as a source of genetic diversity. Humans are rapidly destroying large areas of tropical forests; it is only by means of remote sensing that we are ever likely to understand the nature and locations of these changes (Campbell, 2002). During the 1980s, pioneering research was conducted to map and monitor vegetation on continental scales using data acquired by the U.S. National Oceanographic and Atmospheric Administration’s (NOAA) meteorological satellite, the Advanced Very High Resolution Radiometer (AVHRR) (DeFries & Belward, 2000). DeFries and Belward (2000) indicated that global land cover products have
*
Shaw-wen Sheen: Assistant Professor, Department of Social Studies Education, National Tainan Teachers College
A Principal Component Analysis of Vegetation Characteristics in North America
175
been derived from AVHRR for fire monitoring, thematic land cover maps (Hansen et al., 2000), and continuous fields of vegetation characteristics (DeFries, Hansen, & Townshend, 2000). Hansen et al. (2000) produced a 1 km spatial resolution land cover classification map using data for 1992-1993 from AVHRR. DeFries et al. (2000) derived global continuous fields of percentage woody vegetation, herbaceous vegetation and bare ground using AVHRR data for the years 1982 to 1994 at 8 km spatial resolution. DeFries, Hansen, Townshend, Janetos, and Loveland (2000) used AVHRR data at 1 km spatial resolution and generated a global map depicting percentage tree cover and associated proportions of trees with different leaf longevity (evergreen and deciduous) and leaf type (broadleaf and needleleaf). The aim of this research is to use principal component analysis to analyze spatial variations in 1 km resolution leaf longevity and leaf type maps of North America. This research classified composite map derived from 3 principal components.
DATA AND METHODS
This research used Principal Component Analysis of IDRISI32 Release 2 to analyze evergreen, deciduous, broadleaf and needleleaf maps of North America. The study area is from 120°W to 50°W and 10°N to 70°N. Figure 1 shows the percent tree cover of the study area. Cluster Analysis of IDRISI32 Release 2 was used to classify composite map.
176 Journal of National Taipei Teachers College, Vol.17, No.1
Figure 1 Percent Tree Cover of North America.
Vegetation Characteristics Data
The 1 km resolution continuous fields of vegetation characteristics data were downloaded from the Global Land Cover Facility at the University of Maryland. The data were percent evergreen, deciduous, broadleaf, and needleleaf maps of North America. These maps were produced using 1992 to 1993 AVHRR data.
Principal Component Analysis and Cluster Analysis
Principal Components Analysis of IDRISI32 Release 2 on a set of images produces a new set of images, components that are uncorrelated with each other and explain progressively less of the variance found in the original set of images (Clark Labs, 2001). Cluster Analysis of IDRISI32 Release 2 provides an unsupervised classification of an image based on the information in a composite image (Clark Labs, 2001). The aim of unsupervised classification is to uncover the major classes that exist in the image without prior knowledge of what they might be. Unsupervised classification
A Principal Component Analysis of Vegetation Characteristics in North America
177
techniques search for clusters of pixels with similar reflectance characteristics in a multi-band image. They are concerned with uncovering the major classes, and thus tend to ignore those that have very low frequencies of occurrence. Cluster Analysis uses a histogram peak technique (Clark Labs, 2001). This is equivalent to looking for the peaks in a one-dimensional histogram, where a peak is defined as a value with a greater frequency than its neighbors on either side. Once the peaks have been identified, all possible values are assigned to the nearest peak and the divisions between classes fall at the midpoints between peaks. Here a three-dimensional histogram is used because the composite is derived from three images. A peak is thus a class where the frequency is higher than all of its neighbors. Once the peaks have been located, each pixel in the image can then be assigned to its closest peak, with each class being labeled as a cluster (Clark Labs, 2001).
RESULTS
Principal Components
Tables 1 and 2 lists the results of Principal Component Analysis. The components 1, 2, and 3 explain 73.2%, 25.1%, and 1.6% of the variance in the original 4 vegetation characteristics maps (evergreen, deciduous, broadleaf and needleleaf maps). Component 1 is highly correlated with needleleaf and evergreen tree types (Table 2). Component 2 is highly correlated with broadleaf and deciduous tree types. Table 1. The Principal Component Eigenvalues and Eigenvectors of Vegetation Characteristics of North America. Component 1 Component 2 Component 3 Component 4 % Variance 73.21 25.14 1.64 0.00 Eigenvalue 762.08 261.73 17.10 0.00 Broadleaf 0.110289 0.743128 -0.430761 0.500041 Needleleaf 0.691524 -0.149579 0.499421 0.499999 Evergreen 0.707629 -0.056237 -0.496165 -0.499919 Deciduous 0.094298 0.649787 0.564662 -0.500041
178 Journal of National Taipei Teachers College, Vol.17, No.1
Table 2. The Component Loadings of Vegetation Characteristics of North America. Component 1 Component 2 Component 3 Component 4 Broadleaf 0.243069 0.959816 -0.142214 0.000060 Needleleaf 0.986773 -0.125085 0.106754 0.000039 Evergreen 0.993091 -0.046252 -0.104308 -0.000038 Deciduous 0.234894 0.948558 0.210700 -0.000068 Figures 2 and 3 show principal components 1 and 2. The range of the component image 1 is from 0 to 112 (Figure 2). Northern regions of the map have higher values, indicating the higher concentrations of needleleaf and evergreen vegetation. The range of the component image 2 is from -16 to 111 (Figure 3). Southern and eastern regions of the map have higher values, indicating the higher concentrations of broadleaf and deciduous vegetation. These principal components show the major spatial variability over the original vegetation characteristics of North America.
Figure 2. Maps of Component 1 of Vegetation Characteristics of North America.
A Principal Component Analysis of Vegetation Characteristics in North America
179
Figure 3. Maps of Component 2 of Vegetation Characteristics of North America.
Clusters of Vegetation Characteristics
The research used Cluster Analysis of IDRISI32 Release 2 to classify vegetation characteristics composite map. North American vegetation characteristics composite image was produced from combining principal components 1, 2, and 3 images. Broad clustering of Cluster Analysis generated 11 clusters with dropping least significant clusters (<1% of all area) (Figure 4). Then, we could identify the vegetation cover class of each cluster. Cluster 1 indicates non-vegetated or tree cover less than 5% area. This research did not recognize all the clusters because more data (e.g., field survey, vegetation maps, remote sensing high-resolution images, and aerial photographs) are needed to identify all the vegetation clusters. The research concludes that the result could be useful in the future to improve vegetation cover classification of North America.
180 Journal of National Taipei Teachers College, Vol.17, No.1
Figure 4. Cluster Map of Vegetation Characteristics of North America.
CONCLUSIONS
This study analyzed spatial variations in vegetation characteristics maps of North America. The components 1, 2, and 3 explain 73%, 25%, and 2% of the variance in the original 4 vegetation maps. Component 1 is highly correlated with needleleaf and evergreen vegetation. Component 2 is highly correlated with broadleaf and deciduous vegetation. Eleven North American vegetation clusters were generated in this research. The result is useful to improve North American vegetation cover classification in the future work.
ACKNOWLEDGEMENTS
The author thanks the Global Land Cover Facility at the University of Maryland, U.S.A. for producing the data in their present form and distributing them.
A Principal Component Analysis of Vegetation Characteristics in North America
181
REFERENCES
Campbell, J. B. (2002). Introduction to remote sensing (3rd ed). New York: The Guilford Press. Clark Labs (2000). IDRISI32 release 2. Clark University, USA. Clark Labs (2001). IDRISI32 release 2 guide to GIS and image processing Volume 2. Clark University, USA. DeFries, R.S., & Belwad, A.S. (2000). Global and regional land cover characterization from satellite data: an introduction to the Special Issue. International Journal of Remote Sensing, 21, 1083-1092. DeFries, R., Hansen, M., Townshend, J. R. G., & Sholberg, R. (1998). Global land cover classifications at 8 km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiers. International Journal of Remote Sensing, 19, 31413168. DeFries, R. S, Hansen, M.C., & Townshend, J. R. G. (2000). Global continuous fields of vegetation characteristics: a linear mixture model applied to multi-year 8 km AVHRR data. International Journal of Remote Sensing, 21, 1389-1414. DeFries, R. S., Hansen, M. C., Townshend, J. R. G., Janetos, A. C., & Loveland, T. R. (2000). A new global 1km data set of percent tree cover derived from remote sensing. Global Change Biology, 6, 247-254. Global Land Cover Facility, University of Maryland (n.d.). Retrieved December 5, 2003, from http://glcf.umiacs.umd.edu/ Hansen, M. C., DeFries, R. S., Townshend, J. R. G., & Sholberg, R. (2000). Global land cover classification at 1 km spatial resolution using a classification tree approach. International Journal of Remote Sensing, 21, 1331-1364. Sabins, F.F. (1996). Remote sensing: principles and interpretation (3rd ed). New York: W.H. Freeman and Company.
182
國立臺北師範學院學報，第十七卷第一期（九十三年三月）173～182 國立臺北師範學院
北美地區植物特性 之主成份因子分析
沈少文*
摘
要
植物分佈是地球環境許多作用中重要因子之一，本文目標是使用主成份因 子分析方法分析北美地區植物特性之空間變化，研究地區範圍從西經 120 度到 50 度及北緯 10 度到 70 度，研究結果顯示主成份因子 1，2 及 3 分別解釋 73%， 25%及 2%之北美地區 4 類植物特性變異(常綠樹，落綠樹，闊葉樹及針葉樹)， 主成份因子 1 與針葉樹及常綠樹呈高度正相關，主成份因子 2 與闊葉樹及落綠樹 呈高度正相關，本文使用群落分析方法研究北美地區植物特性，研究結果顯示 有 11 個主要植物群落。
關鍵字：植物分佈、主成份因子分析方法、群落分析方法、北美地區
*
沈少文：國立台南師範學院社會科教育學系助理教授