Evaluating temporal consistency of long-term global NDVI datasets for trend analysis
(2015) In Remote Sensing of Environment 163. p.326-340- Abstract
As a way to understand vegetation changes, trend analysis on NDVI (normalized difference vegetation index) time series data have been widely performed at regional to global scales. However, most long-term NDVI datasets are based upon multiple sensor systems and unsuccessful corrections related to sensor shifts potentially introduce substantial uncertainties and artifacts in the analysis of trends. The temporal consistency of NDVI datasets should therefore be evaluated before performing trend analysis to obtain reliable results. In this study we analyze the temporal consistency of multi-sensor NDVI time series by analyzing the co-occurrence between breaks in the NDVI time series and sensor shifts from GIMMS3g (Global Inventory Modeling... (More)
As a way to understand vegetation changes, trend analysis on NDVI (normalized difference vegetation index) time series data have been widely performed at regional to global scales. However, most long-term NDVI datasets are based upon multiple sensor systems and unsuccessful corrections related to sensor shifts potentially introduce substantial uncertainties and artifacts in the analysis of trends. The temporal consistency of NDVI datasets should therefore be evaluated before performing trend analysis to obtain reliable results. In this study we analyze the temporal consistency of multi-sensor NDVI time series by analyzing the co-occurrence between breaks in the NDVI time series and sensor shifts from GIMMS3g (Global Inventory Modeling and Mapping Studies 3rd generation), VIP3 (Vegetation Index and Phenology version 3), LTDR4 (Long Term Data Record version 4) and SPOT-VGT (Système Pour l'Observation de la Terre VEGETATION). Single sensor time series from MODIS (MODerate Resolution Imaging Spectroradiometer) Terra and Aqua are used as reference datasets. The global land surface is divided into six regions according to the world humidity zones and averaged NDVI time series in each region are analyzed separately using a multiple structural change detection approach. We find that artifacts exist in the VIP3 and LTDR4 NDVI datasets with an abrupt shift detected in all humidity zones coinciding with the shift from NOAA-9 to NOAA-11 in 1988 and that orbital drift effects are evident in arid regions, potentially introducing uncertainties in NDVI trend analysis. Platform/sensor change from VGT-1 to VGT-2 is found to cause a significant positive break in the SPOT-VGT NDVI time series. Potential artifacts exist in humid, dry-subhumid, semi-arid and hyper-arid regions of GIMMS3g NDVI, whereas no signs of artifacts are found in the arid region. Although temporal consistency throughout all examined datasets increases after 2000 due to the usage of advanced platforms and sensors, variations in NDVI values from 2010 to 2011 still result in different trends at global and regional scales.
(Less)
- author
- Tian, Feng LU ; Fensholt, Rasmus ; Verbesselt, Jan ; Grogan, Kenneth ; Horion, Stephanie and Wang, Yunjia
- publishing date
- 2015-06-05
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Break points dectection, GIMMS3g, LTDR, MODIS, Sensor shift related artifacts, SPOT-VGT, VIP
- in
- Remote Sensing of Environment
- volume
- 163
- pages
- 15 pages
- publisher
- Elsevier
- external identifiers
-
- scopus:84937764455
- ISSN
- 0034-4257
- DOI
- 10.1016/j.rse.2015.03.031
- language
- English
- LU publication?
- no
- id
- d5c179b2-156f-4fc4-b320-bb33fd6bb4ae
- date added to LUP
- 2018-06-08 14:18:16
- date last changed
- 2022-04-25 07:46:32
@article{d5c179b2-156f-4fc4-b320-bb33fd6bb4ae,
abstract = {{<p>As a way to understand vegetation changes, trend analysis on NDVI (normalized difference vegetation index) time series data have been widely performed at regional to global scales. However, most long-term NDVI datasets are based upon multiple sensor systems and unsuccessful corrections related to sensor shifts potentially introduce substantial uncertainties and artifacts in the analysis of trends. The temporal consistency of NDVI datasets should therefore be evaluated before performing trend analysis to obtain reliable results. In this study we analyze the temporal consistency of multi-sensor NDVI time series by analyzing the co-occurrence between breaks in the NDVI time series and sensor shifts from GIMMS3g (Global Inventory Modeling and Mapping Studies 3rd generation), VIP3 (Vegetation Index and Phenology version 3), LTDR4 (Long Term Data Record version 4) and SPOT-VGT (Système Pour l'Observation de la Terre VEGETATION). Single sensor time series from MODIS (MODerate Resolution Imaging Spectroradiometer) Terra and Aqua are used as reference datasets. The global land surface is divided into six regions according to the world humidity zones and averaged NDVI time series in each region are analyzed separately using a multiple structural change detection approach. We find that artifacts exist in the VIP3 and LTDR4 NDVI datasets with an abrupt shift detected in all humidity zones coinciding with the shift from NOAA-9 to NOAA-11 in 1988 and that orbital drift effects are evident in arid regions, potentially introducing uncertainties in NDVI trend analysis. Platform/sensor change from VGT-1 to VGT-2 is found to cause a significant positive break in the SPOT-VGT NDVI time series. Potential artifacts exist in humid, dry-subhumid, semi-arid and hyper-arid regions of GIMMS3g NDVI, whereas no signs of artifacts are found in the arid region. Although temporal consistency throughout all examined datasets increases after 2000 due to the usage of advanced platforms and sensors, variations in NDVI values from 2010 to 2011 still result in different trends at global and regional scales.</p>}},
author = {{Tian, Feng and Fensholt, Rasmus and Verbesselt, Jan and Grogan, Kenneth and Horion, Stephanie and Wang, Yunjia}},
issn = {{0034-4257}},
keywords = {{Break points dectection; GIMMS3g; LTDR; MODIS; Sensor shift related artifacts; SPOT-VGT; VIP}},
language = {{eng}},
month = {{06}},
pages = {{326--340}},
publisher = {{Elsevier}},
series = {{Remote Sensing of Environment}},
title = {{Evaluating temporal consistency of long-term global NDVI datasets for trend analysis}},
url = {{http://dx.doi.org/10.1016/j.rse.2015.03.031}},
doi = {{10.1016/j.rse.2015.03.031}},
volume = {{163}},
year = {{2015}},
}