1. Ph.D. Dissertation Defense Warming-enhanced Plant Growth in the North Since 1980s: A Greener Greenhouse? Liming Zhou Department of Geography, Boston University Dissertation Committee Ranga B. Myneni Robert K. Kaufmann Yuri Knyazikhin Nathan Phillips Compton J. Tucker 1 of 43

2. Summary of Presentation 1.Motivation 2.Data 3.Quality of Satellite Data 4.Changes in Northern Vegetation Activity 5.Spatial Pattern of Changes in Vegetation 6.Drivers for Changes 7.Contributions of Research 8.Future Directions 2 of 43

3. Has Vegetation Responded to Climate Change? Pronounced warming in northern high latitudes Earlier disappearance of snow in spring Increased precipitation in northern high latitudes Increased concentration of atmospheric CO 2 Changes in Climate Increased productivity through: - enhanced photosynthesis - enhanced nutrient availability Changes in Vegetation 3 of 43

4. i. greatest warming in winter and spring ii. continental difference: overall warming in Eurasia and smaller warming or cooling trends in North America coolingwarming  Monthly land surface climate data (1981-1999) 1. NOAA precipitation: 2.5  x2.5  2. GISS temperature: 2  x2  Data 4 of 43

5.  Satellite NDVI data at 8 km resolution 1. GIMMS 15-day composite NDVI (07/81-12/99) 2. Pathfinder AVHRR Land 10-day composite NDVI (07/81- 09/94)  Solar zenith angle (SZA) from GIMMS and Pathfinder data  Monthly stratospheric aerosol optical depth (AOD) reported as zonal means  A land cover map at 8 km resolution 5 of 43

6.  Factors that may contaminate long-term satellite measures: 1. calibration uncertainties (satellite drift and changeover) 2. atmospheric and bidirectional effects (aerosol, vapor, etc) 3. soil background effects  Methods that help to reduce some non-vegetation effects 1. Maximum NDVI compositing 2. Spatial and temporal aggregations 3. Empirical methods Changes in SZAChanges in AOD El ChichonMt. Pinatubo 6 of 43

7. Are AVHRR Satellite Measures of NDVI Contaminated by Satellite Drift and Changeover? Kaufmann, R. K., Zhou, L., Knyazikhin, Y., Shabanov, N.V, Myneni, R.B., and Tucker, C.J., Effect of orbital drift and sensor changes on the time series of AVHRR vegetation index data. IEEE Trans. Geosci. Remote Sens. 38: 2584-2597, 2000. 7 of 43

8. Theoretical Analysis The effect of changes in SZA on NDVI can be examined from radiative transfer equation in vegetation media. NDVI = f (SZA, …) Sensitivity experiments Result: NDVI is minimally sensitive to SZA changes and this sensitivity decreases as leaf area increases. 8 of 43

9. Empirical Analysis  A statistically meaningful relation between NDVI and SZA? 1. Ordinary least squares (OLS) 2. Cointegration analysis (VECM) i. spurious regression results? i. a cointegrating relation? ii. the statistical ordering of this relation? 9 of 43

10. Land cover type A statistically meaningful relation? OLSCointegration analysis causal order Evergreen needleleaf forests no Evergreen broadleaf forests yesno Deciduous needleleaf forests no Deciduous broadleaf forests no Mixed forests no Woodlands no Wooded grasslands/shrubs yes SZA  NDVI Closed bushlands/shrublands yes SZA  NDVI Open shrublands yes SZA  NDVI Grasses yes SZA  NDVI Relationship between NDVI and SZA 10 of 43

11. Conclusions  Theoretical and empirical analyses indicate that NDVI is minimally sensitive to SZA changes and this sensitivity decreases as leaf area increases.  Using OLS can generate spurious regressions because of the nonstationary properties of time series.  The AVHRR NDVI do not cointegrate with satellite drift and changeover for dense vegetation types. 11 of 43

12. Has Northern Hemisphere Vegetation Changed? Zhou, L., Tucker, C.J., Kaufmann, R.K., Slayback, D., Shabanov, N.V, and Myneni, R.B., Variations in northern vegetation activity inferred from satellite data of vegetation index during 1981 to 1999. J. Geophys. Res. 106, 20069-20083, 2001. 12 of 43

13. Study Pixels  Vegetated pixels (defined by NDVI) between April to October 1. Minimize the SZA effect 2. Reduce the soil background contribution (snow, barren and sparsely vegetated areas) 3. Use data from the same pixels in the entire analysis. Map of vegetated pixels 13 of 43

14. Changes in Vegetation Activity  Changes in vegetation photosynthetic activity can be characterized by 1. changes in growing season 2. changes in NDVI magnitude Increases in NDVI magnitudeIncreases in growing season JanDecJul Aug Increase NDVI JanDecJul Aug earlier spring delayed fall NDVI 14 of 43

15. Increases in Growing Season (Increased by 18 Days) (Increased by 12 Days) 11.9 days/18 yrs (p<0.05) 17.5 days/18 yrs (p<0.05) 15 of 43

16. Increases in NDVI Magnitude (8 Percent Increase) (12 Percent Increase) 8.4/18 yrs (p<0.05) 12.4/18 yrs (p<0.05) 16 of 43

17. Spatial Pattern of Changes in NDVI Magnitude  Three methods 1. NDVI difference 2. NDVI trend 3. Persistence index of NDVI 17 of 43

18. NDVI difference between 1995-99 and 1982-86 averages 18 of 43

19. NDVI trend at the 5% significance level 19 of 43

20. Define the persistence index (PI) of NDVI increase: PI = PI(1) + PI(2) + PI(3) + PI(4) + PI(5) +PI(6), 0  PI  6 PI (i) = 1 if Trend i > 80% Trend i-1 0 otherwise Year PI (1) = 1 if Trend 1 > 80% Trend 0 0 otherwise 82 84 87 89 91 93 95 97 99 P(1), PI(2), PI(3), P(4), PI(5), PI(6) Trend 0 (82-87) Trend 1 (82-89) Trend 2 (82-91) Trend 3 (82-93) Trend 4 (82-95) Trend 5 (82-97) Trend 6 (82-99) 20 of 43

21. Persistence index of NDVI 21 of 43

22. Consistency between NDVI and Temperature R=0.79 (p<0.01) R=0.72 (p<0.01) 22 of 43

23. Statistical Results yx A statistically meaningful relation? y =  0 +  1 x +  y =  0 +  1 x+  2 time +   y =  0 +  1  x +  EA NDVIEA Tyes NA NDVINA Tyes EA NDVINA Tno NA NDVIEA Tno Note: T – Temperature; EA – Eurasia; NA - North America 23 of 43

24. Conclusions  Eurasia is photosynthetically more vigorous than North America during the past two decades: 1. Eurasia has a higher percentage of vegetated pixels (61% vs. 30%) showing a larger increase in the NDVI magnitude (12% vs. 8%) and a longer active growing season (18 vs. 12 days) than North America. 2. The temporal changes and continental differences in NDVI are consistent with land surface temperature, an important determinant of biological activity. 24 of 43

25. Is the Spatial Pattern of Greening in North America and Eurasia Different? Bogaert, J., Zhou, L., Tucker, C.J., Myneni, R.B, and Ceulemans, R., Evidence for a persistent and extensive greening trend in Eurasia inferred from satellite vegetation index data. J. Geophys. Res. 2001 (in review). 25 of 43

26. Patch Level Analysis  Patch concept: 253 255 235 253 255 235 9 pixels4 patches  Advantage 1. Enhance confidence in the observed NDVI trends  trends that show spatial proximity and coherence are more reliable than those observed only in spatially fragmented regions. 26 of 43

27. Spatial Pattern Assessment  Spatial pattern assessment for fragmentation: 1. Patch statistics: area, perimeter, number 2. Patch coherence 3. Patch fragmentation index 4. Largest patch index 5. Pixel contiguity 6. Pixel clustering 7. Conditional probability of pixel adjacency 27 of 43

28. Results  Patches with long- and short-term persistence index (PI) Long-term greeningShort-term greening PI CoherenceFragmentation Index EurasiaNorth AmericaEurasiaNorth America 30.0270.0390.0460.031 50.1940.0780.0130.020 28 of 43

29.  Patch area in North America (NA) and Eurasia (EA) 1. more large patches in EA than NA for PI=5 2. more small patches in EA than NA for PI=3 29 of 43

30.  Pixel contiguity in North America (NA) and Eurasia (EA) 1. higher degree of connectedness in EA than NA for PI=5 2. lower degree of connectedness in EA than NA for PI=3 High contiguityLow contiguity 30 of 43

31. Conclusions  Eurasia shows a more temporally persistent and spatially extensive greening than North America: 1. Eurasia shows a larger extent of long-term greening, characterized by higher contiguity and clustering, larger patches, higher probability of pixel adjacency, lower overall fragmentation level; 2. North America shows a fragmented spatial pattern of long-term NDVI increase, and is less dominated by long- term greening. 31 of 43

32. What Drives the Observed Changes in NDVI? Zhou, L., Kaufmann, R.K., Y. Tian, Myneni, R.B, and Tucker, C.J., Relation Between Interannual Variations in Satellite Measures of Vegetation Greenness and Climate Between 1982 and 1999. Global Biogeochem. Cyc. 2001 (in review). 32 of 43

33. Structure of Panel Data  Data georeferenced to 2  x 2  boxes: - Eurasia: 980 boxes - North America: 450 boxes  Data aggregated into seasons: - winter - spring - summer - fall  NDVI classified into vegetation types: - evergreen needleleaf forests - deciduous needleleaf forests - deciduous broadleaf forests - mixed forests - woodlands 33 of 43

34. Modeling NDVI NDVI it =  i +  i X it +  it i = 1,…, Boxes t = 1,…, Years Dependent variable: - NDVI (seasons, vegetation types) Independent variables (X): - temperature (T) - precipitation (P) - solar zenith angle (SZA) - stratospheric aerosol optical depth (AOD) Advantages: - increase degrees of freedom - represent effects of spatial heterogeneity and unobserved variables 34 of 43

35. NDVI smmer =  11 T winter +  12 T 2 winter +  21 T spring +  22 T 2 spring +  31 T summer +  32 T 2 summer +  41 P winter +  42 P 2 winter +  51 P spring +  52 P 2 spring +  61 P summer +  62 P 2 summer +  7 SZA summer +  8 AOD summer +  summer +  An example of one regression to model summer NDVI Climate variables (T and P) are represented using a quadratic specification (a physiological optimum). Effects of climate variables for earlier seasons are included. SZA and AOD are used to separate the non-vegetation effects from the climate effects. 35 of 43

36. Statistical Method NDVI it =  i +  i X it +  it 1. Test homogeneity of slope and intercept coefficients across boxes 2. Test homogeneity of slopes across boxes reject 3. Random coefficient model NDVI it =  i  +  i X it +  it reject Pooled ordinary least square NDVI it =  +  X it +  it Fail to reject 2a. Fixed/random effects estimator NDVI it =  i  +  X it +  it Fail to reject 36 of 43

37. Statistical Results  The 27 regressions include 144 climate variables, 27 SZA variables, and 27 AOD variables.  132 of 144 climate variables and 51 of 54 SZA and AOD variables show statistically meaningful relations with NDVI.  These relations vary by seasons and vegetation types.  R 2 ~15-33% 1. associated with the use of anomalies 2. missing variables (forest regrowth, CO 2 fertilization, etc)  Error terms are stationary -- variables cointegrate. 37 of 43

38. Changes in NDVI Associated with Climate 38 of 43

39. Changes in NDVI Associated with SZA and AOD 39 of 43

40. Reconciling the SZA Effect with No Cointegration Result Described Earlier  Two stochastic trends in NDVI 1. Sparse vegetation: 2. Dense vegetation: NDVI SZAClimate strongweak (no cointegration) (cointegration) NDVI SZAClimate weakstrong 40 of 43

41. Conclusions  There is a statistically meaningful relation between changes in NDVI and climate at the regional scale, and this relation varies by season and vegetation type. 1. Changes in temperature account for the largest fraction of the changes in NDVI between the early 1980s and the late 1990s, while precipitation has a smaller effect. 2. Artifacts associated with variations in stratospheric aerosols and solar zenith angle explain a smaller portion of satellite measured NDVI. 41 of 43

42. Contributions of Research  Evidence for a photosynthetically vigorous Eurasia relative to North America during the past two decades, possibly driven in part by temperature and precipitation.  Suggestion of a climate induced increase in northern hemispheric phytomass, which may account for some of the missing carbon.  Important economic implications: warmer planet, greener north. 42 of 43

43. Future Directions  Use finer resolution climate and higher quality satellite data.  Detect the direction of causality between NDVI and climate.  Include other variables such as forest regrowth, anthropogenic pollutants (O 3, SO 2, NO x, etc). 43 of 43

44. The End

45. Any question?