Grouping-Based Time-Series Model for Monitoring of Fall Peak Coloration Dates Using Satellite Remote Sensing Data

Qu Zhou,Wenkai Li,Jian Li,Liqiao Tian,Xianghan Sun

doi:10.3390/rs12020274

Abstract

Accurate monitoring of plant phenology is vital to effective understanding and prediction of the response of vegetation ecosystems to climate change. Satellite remote sensing is extensively employed to monitor vegetation phenology. However, fall phenology, such as peak foliage coloration, is less well understood compared with spring phenological events, and is mainly determined using the vegetation index (VI) time-series. Each VI only emphasizes a single vegetation property. Thus, selecting suitable VIs and taking advantage of multiple spectral signatures to detect phenological events is challenging. In this study, a novel grouping-based time-series approach for satellite remote sensing was proposed, and a wide range of spectral wavelengths was considered to monitor the complex fall foliage coloration process with simultaneous changes in multiple vegetation properties. The spatial and temporal scale effects of satellite data were reduced to form a reliable remote sensing time-series, which was then divided into groups, namely pre-transition, transition and post-transition groups, to represent vegetation dynamics. The transition period of leaf coloration was correspondingly determined to divisions with the smallest intra-group and largest inter-group distances. Preliminary results using a time-series of Moderate Resolution Imaging Spectroradiometer (MODIS) data from 2002 to 2013 at the Harvard Forest (spatial scale: ~3500 m; temporal scale: ~8 days) demonstrated that the method can accurately determine the coloration period (correlation coefficient: 0.88; mean absolute difference: 3.38 days), and that the peak coloration periods displayed a shifting trend to earlier dates. The grouping-based approach shows considerable potential in phenological monitoring using satellite time-series.

Highlights

Vegetation phenological events, which play a crucial role in determining spatial and temporal patterns of net primary production, are highly sensitive to climate change [1]
Moderate Resolution Imaging Spectroradiometer (MODIS) time-series data were presented as nodes and edges to construct the network graph, and peak coloration dates were determined by the network metrics
We considered the fall foliage transition dates as variables and divided the time-series data into three groups according to the assumed transition dates, namely pre-transition, transition, and post-transition groups

Summary

Introduction

Vegetation phenological events, which play a crucial role in determining spatial and temporal patterns of net primary production, are highly sensitive to climate change [1]. Tracking the change in multiple spectral signatures by combining different vegetation properties may outperform the tracking of a single VI time-series trajectory and can accurately estimate the transition dates of complex leaf coloration [16]. To better monitor the complex foliage coloration processes, Diao [16] developed a pheno-network model to detect peak coloration transition dates using collections of spectral signatures In this model, MODIS time-series data were presented as nodes and edges to construct the network graph, and peak coloration dates were determined by the network metrics. To avoid threshold selection and take advantage of multiple spectral signatures, we proposed an innovative grouping-based method for peak foliage coloration estimations using MODIS data and field measurements from 2002 to 2013 at Harvard Forest. The performance of the grouping-based model was evaluated in this study

Data Sources and Preprocessing

Determination of Optimal Spatial and Temporal Scales

Determination of Peak Coloration Periods

Discussion

Conclusions