1. Toward a Stable Real-Time Green Vegetation Fraction Le Jiang, Dan Tarpley, Felix Kogan, Wei Guo and Kenneth Mitchell JCSDA Science Workshop May 31 – June 1, 2006

2. The Problem: AVHRR NDVI and green vegetation fraction are not stable over time NOAA-7NOAA-9NOAA-11NOAA-14NOAA-16

3. Approach: Develop an algorithm for operational adjustment to real-time global NDVI to ensure the time series consistency Assume the “climatology” of the Earth’s distribution of NDVI is stable over time for each week Select a climatology of “standard” years without problems with: –Instrument calibration –Equator crossing time –Natural perturbations of the atmosphere (volcanic eruptions) Evaluate mathematical or statistical procedures that adjust real time NDVI distributions to climatological “standard” –Method has to retain regional information about vegetation condition –Method should be simple and require minimum assumptions –Method should not rely on outside data sources Select procedure for operational use

4. NDVI Time Series “Benchmark NDVI Climatology” selected from years with the best known data quality: 1989, 1990, 1995 (after wk 14), 1996, 1997, 1998 from NOAA-11 and -14

5. Approaches EvaluatedDescription a), b) Range Re-Scaling (RRS) c. Normalization (NML) d. Linear Regression (LR) e. Adjusted Cumulative Distribution Function (ACDF) Adjusting the CDF of real-time weekly global NDVI to match that of the benchmark climatology for week i f. NML+RRSNML followed by RRS g. Adjusting Satellite-By- Satellite (ASBS)+NML ASBS followed by NML h. GIMMS Data (for comparison) Independent datasets from Global Inventory Monitoring & Modeling Studies (GIMMS) group: global 8-km, semi- monthly, time series decomposed / reconstructed, RT corrected Solve ND 0 (equivalent ND within the context of benchmark climo. a) Use pixel max ND as ND max b) Use Avg of top 1% max ND as ND max

6. Schematic Illustration of the ACDF Approach (option e) Benchmark CDF

7. Before Correction

8. After CDF Correction

9. Before and After

10. Results & Evaluation abcdefgh Results after simple adjustment Yellow – maximum NDVI Green – average of top 1% NDVI Blue – average NDVI Red – standard deviation of NDVI a. RRSb. RRS_Top 1% c. NML d. LR e. ACDFf. NML+RRS g. ASBS+NML h. GIMMS Data

11. E.g. Performance of different fixes on Class 7&12 (Short Ground Cover and Cropland): Yellow – maximum NDVI Green – average of top 1% NDVI Blue – average NDVI Red – standard deviation of NDVI a. RRS b. RRS_Top 1%c. NML d. LRe. ACDF f. NML+RRSg. ASBS+NML h. GIMMS Data Un-adjusted

12. Satellite ECTs for the period 1982 to 2003 Effects after different adjustment for week 27 from 1982 to 2003. a. RRS b. RRS_Top 1% c. NML d. LRe. ACDF f. NML+RRS g. ASBS+NML h. GIMMS Data Un-adjusted

13. Comparison of drought detection signatures over CONUS (Weeks 16, 20, 24, 29, 33 and 37 in 2005) Left column: ND* from un-adjusted NDVI; Middle column: ND* from ACDF fixed NDVI; Right column: Vegetation Condition Index (VCI) based on manual adjustment of unfixed NDVI Indicator (or quantity examined):

14. Summary ACDF correction looks best Successfully compensates for sensor change and orbit drift Local vegetation anomalies retained in corrected data More validation needed Applicable to VIIRS NDVI on day 1 Possible long-term NDVI trends removed

15. Stop

16. Before Correction

19. 13-class global land surface type map 1) Broadleaf-evergreen trees (tropical forest); 2) Broadleaf-deciduous trees; 3) Broadleaf and needle leaf trees; 4) Needle leaf evergreen trees; 5) Needle leaf deciduous trees (larch); 6) Broadleaf trees with ground cover (savanna); 7) Short groundcover (in perennial); 8) Broadleaf shrubs with perennial ground cover; 9) Broadleaf shrubs with bare soil; 10) Tundra (dwarf trees and shrubs with ground cover); 11) Bare soil; 12) Cropland (cultivated); 13) Glacial.

20. Evolution of percentages for different NDVI intervals from 1982 to 2003 ( Black – 0.0% ~ 20%, Blue – 20% ~ 40%, Green – 40% ~ 60%, Yellow – 60% ~ 80%, Red – 80% ~ 100% of maximum NDVI) a. RRS b. RRS_Top 1% c. NML d. LRe. ACDF f. NML+RRS g. ASBS+NML h. GIMMS Data Un-adjusted

21. ND* of the unfixed, adjusted, and GIMMS datasets for week 26 over US Great Plains from 1982 to 2003 a. RRS b. RRS_Top 1% c. NML d. LR e. ACDF f. NML+RRSg. ASBS+NML h. GIMMS Data Un-adjusted