Comparison of Harmonic Analysis of Time Series (HANTS) and Multi-Singular Spectrum Analysis (M-SSA) in Reconstruction of Long-Gap Missing Data in NDVI Time Series

Hamid Reza Ghafarian Malamiri,Terence Darlington Mushore,Emma Izquierdo Verdiguier,Haraldur Olafsson,Hadi Zare,Iman Rousta,Hao Zhang

doi:10.3390/rs12172747

Abstract

Monitoring vegetation changes over time is very important in dry areas such as Iran, given its pronounced drought-prone agricultural system. Vegetation indices derived from remotely sensed satellite imageries are successfully used to monitor vegetation changes at various scales. Atmospheric dust as well as airborne particles, particularly gases and clouds, significantly affect the reflection of energy from the surface, especially in visible, short and infrared wavelengths. This results in imageries with missing data (gaps) and outliers while vegetation change analysis requires integrated and complete time series data. This study investigated the performance of HANTS (Harmonic ANalysis of Time Series) algorithm and (M)-SSA ((Multi-channel) Singular Spectrum Analysis) algorithm in reconstruction of wide-gap of missing data. The time series of Normalized Difference Vegetation Index (NDVI) retrieved from Landsat TM in combination with 250m MODIS NDVI time image products are used to simulate and find periodic components of the NDVI time series from 1986 to 2000 and from 2000 to 2015, respectively. This paper presents the evaluation of the performance of gap filling capability of HANTS and M-SSA by filling artificially created gaps in data using Landsat and MODIS data. The results showed that the RMSEs (Root Mean Square Errors) between the original and reconstructed data in HANTS and M-SSA algorithms were 0.027 and 0.023 NDVI value, respectively. Further, RMSEs among 15 NDVI images extracted from the time series artificially and reconstructed by HANTS and M-SSA algorithms were 0.030 and 0.025 NDVI value, respectively. RMSEs of the original and reconstructed data in HANTS and M-SSA algorithms were 0.10 and 0.04 for time series 6, respectively. The findings of this study present a favorable option for solving the missing data challenge in NDVI time series.

Highlights

Monitoring vegetation coverages over time is highly important and is applicable to many research fields, including management of natural, water and agricultural resources [1,2,3,4]
Harmonic ANalysis of Time Series (HANTS) and M-Singular Spectrum Analysis (SSA) algorithms were comparatively used to fill in missing data of Normalized Difference Vegetation Index (NDVI) time series derived from the Landsat 5 TM and MODIS data
The results showed that the Multi-Singular Spectrum Analysis (M-SSA) accuracy in filling the gaps of missing data was higher than that of the HANTS algorithm

Summary

Introduction

Monitoring vegetation coverages over time is highly important and is applicable to many research fields, including management of natural, water and agricultural resources [1,2,3,4]. Rapid and wide-area monitoring of changes in natural resources including vegetation is possible due to the development of remote sensing technologies and access to satellite images of the past decades. The integrity of remote sensing data can be dramatically altered due to influences of atmospheric dust, aerosols, clouds as well as measurement sensor failure and algorithm malfunctioning, among others [5]. In this regard, clouds are one of the most important factors involved in causing missing data (gaps), outliers and noise in satellite images [6]. Clouds are often indicated in satellite images by characteristic higher reflection and lower surface temperature than other terrestrial phenomena in visible and thermal electromagnetic spectral ranges, respectively [9,10]

Methods

Results

Conclusion