TY - JOUR
T1 - Evaluation of earth observation based global long term vegetation trends — comparing GIMMS and MODIS global NDVI time series
AU - Fensholt, Rasmus
AU - Proud, Simon Richard
PY - 2012/4/16
Y1 - 2012/4/16
N2 - A new and updated version of the AVHRR (Advanced Very High Resolution Radiometer) based GIMMS (Global Inventory Modelling and Mapping Studies) NDVI (Normalized Difference Vegetation Index) dataset is now available covering 1981 to 2010 (GIMMS3g). Earlier versions of this global coverage 15-day composite dataset have been used for numerous local to global scale vegetation time series studies during recent years. However, several aspects of the AVHRR sensor design and data processing potentially introduce substantial noise into the NDVI dataset if not corrected for. The more recent NDVI dataset from Terra MODIS (Moderate Resolution Imaging Spectroradiometer) is considered an improvement over AVHRR data and with the release of GIMMS3g an overlapping period of 11. years now provides a possibility to perform a robust evaluation of the accuracy of GIMMS3g data and derived trends. In this study the accuracy is evaluated by comparison with the global Terra MODIS NDVI (MOD13C2 Collection 5) data using linear regression trend analysis. The trends of GIMMS NDVI were found to be in overall acceptable agreement with MODIS NDVI data. A significant trend in NDVI (α = 0.05) was found for 11.8% of the MODIS NDVI pixels on a global scale (5.4% characterised by positive trends and 6.3 with negative trends) whereas GIMMS NDVI analysis produced a total of 10.5% significant pixels (4.9% positive, 5.6% negative). However, larger differences were found for the Southern Hemisphere land masses (South America and Australia) and the high northern latitude Arctic regions. From a linear regression analysis the correlation coefficient between the two datasets was found to be highly significant for areas with a distinct phenological cycle. Discrepancies between the GIMMS and MODIS datasets were found in equatorial areas (broadleaved, evergreen forest), Arctic areas (sparse herbaceous or sparse shrub cover) and arid areas (herbaceous cover, closed-open). Linear regression of QA filtered Terra and Aqua MODIS NDVI (2003-2010) revealed similar inconsistencies for Arctic and equatorial areas suggesting that robust long-term NDVI trend estimates in these areas are difficult to obtain from both GIMMS and MODIS data. Additionally, GIMMS based NDVI trend analysis in arid areas of limited photosynthetic activity should be interpreted with caution. The regression coefficient (slope value) (p < 0.01) was found to be close to 1 for most land cover types on a global scale (global land cover class average slope = 1.00) suggesting overall compatibility between MODIS and GIMMS NDVI, but with land cover class specific variations (within class and between classes).
AB - A new and updated version of the AVHRR (Advanced Very High Resolution Radiometer) based GIMMS (Global Inventory Modelling and Mapping Studies) NDVI (Normalized Difference Vegetation Index) dataset is now available covering 1981 to 2010 (GIMMS3g). Earlier versions of this global coverage 15-day composite dataset have been used for numerous local to global scale vegetation time series studies during recent years. However, several aspects of the AVHRR sensor design and data processing potentially introduce substantial noise into the NDVI dataset if not corrected for. The more recent NDVI dataset from Terra MODIS (Moderate Resolution Imaging Spectroradiometer) is considered an improvement over AVHRR data and with the release of GIMMS3g an overlapping period of 11. years now provides a possibility to perform a robust evaluation of the accuracy of GIMMS3g data and derived trends. In this study the accuracy is evaluated by comparison with the global Terra MODIS NDVI (MOD13C2 Collection 5) data using linear regression trend analysis. The trends of GIMMS NDVI were found to be in overall acceptable agreement with MODIS NDVI data. A significant trend in NDVI (α = 0.05) was found for 11.8% of the MODIS NDVI pixels on a global scale (5.4% characterised by positive trends and 6.3 with negative trends) whereas GIMMS NDVI analysis produced a total of 10.5% significant pixels (4.9% positive, 5.6% negative). However, larger differences were found for the Southern Hemisphere land masses (South America and Australia) and the high northern latitude Arctic regions. From a linear regression analysis the correlation coefficient between the two datasets was found to be highly significant for areas with a distinct phenological cycle. Discrepancies between the GIMMS and MODIS datasets were found in equatorial areas (broadleaved, evergreen forest), Arctic areas (sparse herbaceous or sparse shrub cover) and arid areas (herbaceous cover, closed-open). Linear regression of QA filtered Terra and Aqua MODIS NDVI (2003-2010) revealed similar inconsistencies for Arctic and equatorial areas suggesting that robust long-term NDVI trend estimates in these areas are difficult to obtain from both GIMMS and MODIS data. Additionally, GIMMS based NDVI trend analysis in arid areas of limited photosynthetic activity should be interpreted with caution. The regression coefficient (slope value) (p < 0.01) was found to be close to 1 for most land cover types on a global scale (global land cover class average slope = 1.00) suggesting overall compatibility between MODIS and GIMMS NDVI, but with land cover class specific variations (within class and between classes).
U2 - 10.1016/j.rse.2011.12.015
DO - 10.1016/j.rse.2011.12.015
M3 - Journal article
SN - 0034-4257
VL - 119
SP - 131
EP - 147
JO - Remote Sensing of Environment
JF - Remote Sensing of Environment
ER -
