Department of Geography, University of South Florida, Tampa, FL, 33620, USA.
Environ Monit Assess. 2011 Jan;172(1-4):199-214. doi: 10.1007/s10661-010-1327-5. Epub 2010 Feb 6.
A stepwise masking system with high-resolution IKONOS imagery was developed to identify and map urban forest tree species/groups in the City of Tampa, Florida, USA. The eight species/groups consist of sand live oak (Quercus geminata), laurel oak (Quercus laurifolia), live oak (Quercus virginiana), magnolia (Magnolia grandiflora), pine (species group), palm (species group), camphor (Cinnamomum camphora), and red maple (Acer rubrum). The system was implemented with soil-adjusted vegetation index (SAVI) threshold, textural information after running a low-pass filter, and brightness threshold of NIR band to separate tree canopies from non-vegetated areas from other vegetation types (e.g., grass/lawn) and to separate the tree canopies into sunlit and shadow areas. A maximum likelihood classifier was used to identify and map forest type and species. After IKONOS imagery was preprocessed, a total of nine spectral features were generated, including four spectral bands, three hue-intensity-saturation indices, one SAVI, and one texture image. The identified and mapped results were examined with independent ground survey data. The experimental results indicate that when classifying all the eight tree species/ groups with the high-resolution IKONOS image data, the identifying accuracy was very low and could not satisfy a practical application level, and when merging the eight species/groups into four major species/groups, the average accuracy is still low (average accuracy = 73%, overall accuracy = 86%, and κ = 0.76 with sunlit test samples). Such a low accuracy of identifying and mapping the urban tree species/groups is attributable to low spatial resolution IKONOS image data relative to tree crown size, to complex and variable background spectrum impact on crown spectra, and to shadow/shaded impact. The preliminary results imply that to improve the tree species identification accuracy and achieve a practical application level in urban area, multi-temporal (multi-seasonal) or hyperspectral data image data should be considered for use in the future.
利用高分辨率 IKONOS 图像开发了逐步掩蔽系统，以识别和绘制美国佛罗里达州坦帕市的城市森林树种/组。这 8 个树种/组包括砂生栎（Quercus geminata）、月桂栎（Quercus laurifolia）、弗吉尼亚栎（Quercus virginiana）、玉兰（Magnolia grandiflora）、松树（树种组）、棕榈（树种组）、樟（Cinnamomum camphora）和红枫（Acer rubrum）。该系统采用土壤调整植被指数（SAVI）阈值、运行低通滤波器后的纹理信息以及近红外波段的亮度阈值，将树冠与非植被区域从其他植被类型（如草地/草坪）中分离出来，并将树冠分为阳光照射区和阴影区。最大似然分类器用于识别和绘制森林类型和树种。在对 IKONOS 图像进行预处理后，共生成了 9 个光谱特征，包括 4 个光谱波段、3 个色调-强度-饱和度指数、1 个 SAVI 和 1 个纹理图像。识别和绘制的结果与独立的地面调查数据进行了检查。实验结果表明，在使用高分辨率 IKONOS 图像数据对所有 8 个树种/组进行分类时，识别精度非常低，无法满足实际应用水平，而将 8 个树种/组合并为 4 个主要树种/组时，平均精度仍然较低（平均精度=73%，总体精度=86%，阳光测试样本的 κ=0.76）。造成城市树种/组识别和绘制精度如此低的原因是树冠大小相对于 IKONOS 图像数据的空间分辨率较低、树冠光谱受复杂多变的背景光谱影响以及阴影/阴影的影响。初步结果表明，为了提高树种识别精度并在城市地区达到实际应用水平，未来应考虑使用多时相（多季节）或高光谱数据图像数据。
