Tropical Geography ›› 2020, Vol. 40 ›› Issue (6): 1051-1062.doi: 10.13284/j.cnki.rddl.003286

Previous Articles    

Normalized Difference Vegetation Index Reconstruction of Cloudy and Rainy Areas Based on Weighted Smooth Spline: A Case Study of Guangdong Province

Zhu Ruan1,2(), Yaoqiu Kuang1,3()   

  1. 1.Guangzhou Institute of Geochemistry, Chinese Academy of Sciences, Guangzhou 510640, China
    2.University of Chinese Academy of Sciences, Beijing 100049, China
    3.School of environment, Jinan University, Guangzhou 511443, China
  • Received:2019-11-28 Revised:2020-04-15 Online:2020-11-30 Published:2020-12-10
  • Contact: Yaoqiu Kuang E-mail:zhuruan89@163.com;kuangyaoqiu@jnu.edu.cn

Abstract:

Compared with other regions, there are more low-quality pixels in the Normalized Difference Vegetation Index (NDVI) of cloudy and rainy areas in terms of temporal and spatial distribution. This would increase the difficulty of reconstructing a high-quality NDVI in cloudy and rainy areas. Guangdong Province—located in the subtropical monsoon region of southern China—is a typical cloudy and rainy province with a high intensity of human activities. Therefore, it is particularly significant to study vegetation change and the response of vegetation to human activities in Guangdong and reconstruct a high-quality NDVI within it. To verify that a weighted smooth spline (sspw) can reconstruct a high-quality NDVI in cloudy and rainy areas and provide reliable data for follow-up studies, this paper takes Guangdong Province as an example and uses MOD13Q1 NDVI from 2000 to 2005 to reconstruct its NDVI. Further, to verify the reliability of the NDVI reconstructed by sspw, the results are compared with those of smooth spline (ssp), Savitzky-Golay (SG), and weighted SG filter (SGw). The following conclusions were drawn. First, compared with SGw, sspw, and SG, sspw-reconstructed NDVI had the largest number of strong correlations and low error pixels. In the correlation coefficient results, 41% of the NDVI pixels reconstructed by sspw had a correlation coefficient of more than 0.7 with the original; in the root mean square error, 64% of the NDVI pixels reconstructed by sspw had an error of less than 0.1 with the original. In terms of Akaike's Information Criterion (AIC) and Bayesian Information Criterion (BIC), the NDVI reconstructed by sspw had 41% of all pixels' AIC < 197 (lower AIC) and 52% of all pixels' BIC < 235 (lower BIC), respectively. This indicates that sspw can reconstruct a high-quality NDVI in cloudy and rainy areas. Second, while the NDVI curve characteristics of typical sample points can be used to determine the reasonable range of the spar coefficient, spar = 0.25 is more suitable for NDVI reconstruction in Guangdong Province. Third, compared with the unweighted ssp and SG, the NDVI time series curve reconstructed by sspw and SGw with quality weight is closer to the original value. Finally, compared with the NDVI reconstructed by sspw, in cloudy and rainy areas, the NDVI reconstructed by SG and SGw will retain some noise and abnormal values in the low- and high-value parts. Therefore, sspw is more suitable for NDVI reconstruction in cloudy and rainy areas.

Key words: cloudy and rainy areas, weighted smooth spline comparison, NDVI reconstruction, filter, Guangdong Province

CLC Number: 

  • TP751