全球长时序植被聚集指数的年际周期变化研究

Year-to-year cyclical variations on global long time series of clumping index product

  • 摘要: 植被聚集指数(clumping index,CI)是表征植被冠层叶片空间聚集程度的结构参数。基于作者所在研究团队研发生产的全球长时序(2001—2019年)、逐月的植被聚集指数(CI)遥感产品,使用傅里叶分解的方法逐像元地对CI的年际周期变化规律进行探究.步骤如下:1)预处理去除数据缺失较多的像元并填补部分缺失较少的像元,以产生相对完整的年际时间序列,预处理筛选出有效研究区占全球植被区面积的75.58%;2)假设CI年际变化可以分解为由多组余弦波信号与随机噪声构成的序列,使用离散傅里叶变换提取时间序列中振幅最大的余弦波(主波),发展了表征CI年际周期(1 a 1周期)变化参数指标体系;3)基于模拟数据和部分高质量数据检验了该方法的抗噪性,并将提取结果与MODIS物候产品(MCD12Q2)的1 a物候周期数(NumCycles,NC)及峰值时间(Peak_1)进行对比验证.结果表明:主波周期12(月)像元占比显著大于其他周期像元,占研究区的76.22%,表明CI最显著、最普遍的年周期性变化特征是周期长度12个月的年际变化;在研究区中主波周期等于12(月)区域与物候产品的1 a物候周期数(NC = 1)高度重合,精度达到96%;对主波周期12月的像元,主波低峰值月与物候产品peak值较接近,全球平均差异为1.37月,且CI低峰值月(植被最聚集月份)普遍提前于peak值,说明年周期植被叶片的最聚集状态(CI的季节变化低峰值)要普遍早于该周期植被叶片的绿度峰值(物候产品的peak最大值).该研究为理解植被聚集效应的年际周期变化提供了证据.

     

    Abstract: Vegetation clumping index (CI) is a structural parameter characterizing spatial clustering degree of vegetation canopy leaves, and playing an important role in extracting other vegetation parameters and in modeling land surface. Currently, remote sensing CI products are limited to a few multiangular satellite-borne sensors, compared to other remote sensing products (e.g., leaf area index, LAI). Therefore, study on long time series of CI products is needed: many models do not consider the cyclical variations of CI product in many applications. The global long-time series (2001-2019) and monthly CI products developed and produced by the author’s research team are therefore used here, to explore interannual periodic variations of 19-year CI product on a global scale, with Fourier decomposition technique. The CI product is first preprocessed by screening out and smoothening outliers, and filling in gaps, to generate a relatively complete year-to-year time series of CI product, with effective pixel proportion accounting for 75.58% of vegetation area on the global scale. Assuming that interannual variations of time-series CI product can be decomposed into a sequence composed of multiple sets of cosine wave signals with possible additive noise of random distribution, discrete Fourier transform is used to extract the cosine wave with the largest amplitude (main wave) in the time series. This is further used as the main index to characterize interannual periodic variation frequency. The period and amplitude of main waves are analyzed and further cross-compared with annual phenological NumCycles (NC) and peak time (Peak_1) of MODIS phenological products (MCD12Q2). Anti-noise performance of this method is tested by simulation data and high-quality CI pixels. The pixel proportions with = 12 are found significantly larger than in other periods, accounting for 76.22% of the study area. This indicates that interannual variations with 12-month period length is most significant among various periodic variations of CI time frequencies. Validation results show that the period with = 12 in the study area highly coincided with annual phenological period of MODIS phenological products with = 1, with validation accuracy reaching to 96%. For the pixel with = 12, low-peak month of the main wave is close to value of MODIS phenological product and average difference between them is about 1.37 months. In general, low-peak month of CI product (i.e., month with most clumping vegetation foliage) is earlier than peak value of MODIS phenological product, probably related to greenness of vegetation foliage. This implies that maximum clumping status is probably earlier to maximum greenness status for vegetation foliage. The present study provides evidence for improved understanding of interannual periodic variations for vegetation foliage clumping effect.

     

/

返回文章
返回