150 likes | 279 Views
This study evaluates four global AVHRR-NDVI datasets—PAL, GIMMS, LTDR, and FASIR—through intercomparison and validation against Landsat imagery. The analysis spans from 1982 to 1999, identifying significant discrepancies in trends and estimates, especially in regions like the Congo and Sahel. Employing the FAO Landsat database for validation, the results highlight the GIMMS dataset as the most popular yet distinctively different, while LTDR shows potential for better calibration. This research emphasizes the crucial implications of dataset selection for remote sensing applications in global vegetation monitoring.
E N D
Global evaluation of four AVHRR–NDVI datasets: Intercomparison and assessment against Landsat imagery Hylke E. Becka,*, Tim R. McVicarb, Albert I.J.M. van Dijkb, Jaap Schellekensc, Richard A.M. de Jeua, L. Adrian Bruijnzeela a VU University Amsterdam, The Netherlands b CSIRO Land and Water, Australia c Deltares, The Netherlands * E-mail: h.e.beck@vu.nl Corresponding publication: Beck, H.E., et al., Global evaluation of four AVHRR–NDVI data sets: Intercomparison and assessment against Landsat imagery, Remote Sensing of Environment (2011), doi:10.1016/j.rse.2011.05.012
Outline • Introduction • AVHRR-NDVI dataset intercomparison • AVHRR-NDVI dataset validation • Conclusion/Summary
(1) Introduction • NOAA’s AVHRR sensors operating since 1981 • Confused as to which global AVHRR-NDVI dataset to use:PAL, GIMMS, LTDR, FASIR, GVI, PAL-II, or …? • All are based on the AVHRR Global Area Coverage archive • Significant differences between these datasets! • Which one do I use? • Validation studies limited (e.g., regional, small sample size) • Idea: validate using forest cover change data? • Better idea: use FAO global database of Landsat samples
(1) Introduction • Four global AVHRR-NDVI datasets: • PAL (8 km, 10 days) • GIMMS (8 km, 15 days) • LTDR V3 (8 km, 10 days) • FASIR (12 km, 10 days) • GIMMS the most popular • LTDR still in development
(2) AVHRR-NDVI dataset intercomparison AVHRR-NDVI dataset intercomparison: • 1982-1999 • Annual means of ‘growing season’ months • Global assessment at 0.5° resolution • Where do the datasets agree/disagree • Median, variance, trend, and correlation (here only trend is discussed)
(2) AVHRR-NDVI dataset intercomparison GIMMS dataset distinctly different patterns! Large differences in Congo and Sahel! Trends in desert areas!
(2) AVHRR-NDVI dataset intercomparison • AVHRR-NDVI dataset intercomparison: • Average for 0.5° latitude bands • LTDR V3 overestimates variance 40°S-30°S • GIMMS has lowest trends • GIMMS higher trends in tundra • Positive trends over almost whole latitudinal range for all datasets
(2) AVHRR-NDVI dataset intercomparison • AVHRR-NDVI dataset intercomparison: • Kruskal-Wallis test used for hypothesis of equal trends • Blue: similar trends • Red: different trends • Inconsistent trends in Africa and Europe • Highly consistent trends in Australia
(2) AVHRR-NDVI dataset intercomparison • AVHRR-NDVI dataset intercomparison: • The most popular dataset (GIMMS) is also the most different • Greening almost the whole latitudinal range and in most regions for all datasets • Most greening in Europe • More favorable conditions globally for vegetation growth
(3) AVHRR-NDVI dataset validation AVHRR-NDVI dataset validation: • FAO Landsat database of 20 x 20 km2 samples • 11,764 Landsat-5 samples covering all major land-cover types • Landsat suitable for validation • NDVI from bands 3 and 4 • Absolute-values comparison (see paper) • Temporal-change comparison • MODIS-NDVI for verification Every dot represents one or more Landsat samples!
(3) AVHRR-NDVI dataset validation • AVHRR-NDVI dataset validation: • Landsat sample pairs • x-axis: AVHRR- or MODIS-NDVI change • y-axis: Landsat-NDVI change • Root Mean Square Difference (RMSD) indicates performance • GIMMS second best • MODIS best
(3) AVHRR-NDVI dataset validation • AVHRR-NDVI dataset validation: • Higher RMSD in dense canopy land-cover types • NDVI saturation and non-linearity • Tropical forests: water vapor and clouds • Boreal forests: large SZA’s
(3) AVHRR-NDVI dataset validation • AVHRR-NDVI dataset validation: • LTDR V3 most accurate in terms of absolute values (see paper) • GIMMS probably (!) most accurate in terms of temporal change • MODIS more accurate than all AVHRR-NDVI datasets, confirms method • Interesting: simple average of the AVHRR-NDVI datasets is better than GIMMS, information is lost by maximum-value compositing?
(4) Conclusion/Summary • Significant differences in trends for almost half of the total land surface • Dataset choice has large implications • PAL and LTDR V3 lack calibration • GIMMS (the most popular dataset) is the most different • GIMMS probably has the best calibration • However, LTDR dataset still in development; may surpass GIMMS
Thank you! Questions? Also check our publication, ask for a hardcopy: Beck, H.E., et al., Global evaluation of four AVHRR–NDVI data sets: Intercomparison and assessment against Landsat imagery, Remote Sensing of Environment (2011), doi:10.1016/j.rse.2011.05.012