Estimating Plant Traits of Grasslands from UAV-Acquired Hyperspectral Images: A Comparison of Statistical Approaches
Estimating Plant Traits of Grasslands from UAV-Acquired Hyperspectral Images: A Comparison of Statistical Approaches
复制标题
DOI：
10.3390/ijgi4042792
复制
发表时间：
2015-12-01
影响因子：
3.4
通讯作者：
Suomalainen, Juha
中科院分区：
文献类型：
--
作者：
Capolupo, Alessandra;Kooistra, Lammert;Suomalainen, Juha
草原生态系统覆盖了整个地球表面约40%的面积。因此，有必要在田间尺度上保证良好的草地管理，以改善草地的保护和实现最佳生长。本研究在偏最小二乘回归(PLSR)和狭义植被指数之间确定了从无人机获取的高光谱图像中估计草地结构和生化特征的最合适的统计策略。此外，还分析了化肥对草地植物性状的影响。高光谱数据是从德国克莱夫附近Haus Riswick农场的一块试验田收集的，用于5月和10月的两次不同飞行活动。收集的图像块被几何和辐射校正以获得表面反射率。利用提取的地块光谱特征，通过计算PLSR和以下狭义植被指数：MERIS陆地叶绿素指数(MTCI)、修正后的叶绿素吸收与吴修正的优化土壤调节植被指数(MCARI/OSAVI)、红边叶绿素指数(Cired-Edge)和归一化红边差(NDRE)来提取草地特征。PLSR对草地结构性状的估测结果较好，但对所选化学性状(粗灰分、粗纤维、粗蛋白质、钠、钾、代谢能)的估测结果不太理想。所建立的关系不受施肥类型和施肥量的影响，而受草原健康状况的影响。在本文分析的方法中，PLSR是探索草地结构和生化特征的最佳策略。使用基于无人机的高光谱遥感可以对草原试验田进行非常详细的评估。
Grassland ecosystems cover around 40% of the entire Earth's surface. Therefore, it is necessary to guarantee good grassland management at field scale in order to improve its conservation and to achieve optimal growth. This study identified the most appropriate statistical strategy, between partial least squares regression (PLSR) and narrow vegetation indices, for estimating the structural and biochemical grassland traits from UAV-acquired hyperspectral images. Moreover, the influence of fertilizers on plant traits for grasslands was analyzed. Hyperspectral data were collected from an experimental field at the farm Haus Riswick, near Kleve in Germany, for two different flight campaigns in May and October. The collected image blocks were geometrically and radiometrically corrected for surface reflectance. Spectral signatures extracted for the plots were adopted to derive grassland traits by computing PLSR and the following narrow vegetation indices: the MERIS Terrestrial Chlorophyll Index (MTCI), the ratio of the Modified Chlorophyll Absorption in Reflectance and Optimized Soil-Adjusted Vegetation Index (MCARI/OSAVI) modified by Wu, the Red-edge Chlorophyll Index (CIred-edge), and the Normalized Difference Red Edge (NDRE). PLSR showed promising results for estimating grassland structural traits and gave less satisfying outcomes for the selected chemical traits (crude ash, crude fiber, crude protein, Na, K, metabolic energy). Established relations are not influenced by the type and the amount of fertilization, while they are affected by the grassland health status. PLSR is found to be the best strategy, among the approaches analyzed in this paper, for exploring structural and biochemical features of grasslands. Using UAV-based hyperspectral sensing allows for the highly detailed assessment of grassland experimental plots.