Spatiotemporal Drought Patterns from Normalized Difference Vegetation Index NDVI

1
Spatiotemporal Drought Patterns from Normalized
Difference Vegetation Index (NDVI)
Deana D. Pennington and Greg Shore University of
New Mexico
2
Normalized Difference Vegetation Index (NDVI)
Spectral Response By Wavelength
GREEN VEGETATION
DARK-TONED SOIL
Chlorophyll Absorption
Leaf Structural Reflectance
TURBID WATER
Greenness Correlated with LAI Biomass
REFLECTED INFRARED
VISIBLE
B G R NIR
NDVI (NIR R ) / (NIR R)
3
Sensors
GREEN VEGETATION
DARK-TONED SOIL
TURBID WATER
REFLECTED INFRARED
VISIBLE
AVHRR
1
2
LANDSAT THEMATIC MAPPER
5
7
1
2
3
4
Sensor Spatial Temporal Product Resolution Res
olution Availability AVHRR 1 km Daily 14 day
composite Post 1989 Maximum NDVI TM 30
m Revisit 16 days 2x per year Post
1972 Spring, Fall
4
Land Cover Change from NDVI

• Vegetation Type
• Phenology
• Site Productivity
• Succession
• Disturbance (fire, disease, grazing, harvest)
• Climate Change

Drought gt Decrease biomass gt Decrease NDVI
5
Climate, Vegetation and NDVI

• ENSO effects on Sevilleta climate, especially
  winter-spring (Molles and Dahm 1990, Swetnam and
  Betancourt 1990, Redmond and Koch 1991, Molles et
  al. 1992)
• PDO monsoon effects on Sevilleta climate, late
  summer-early fall (Gutzler et al, 2002)
• Climate and NDVI correlations (Weiss, 2002)
• 5 km radii around 6 meteorological stations
• 6 different vegetation biomes
• 1990-2000 AVHRR NDVI data
• Factor analysis NDVI correlated with 3
  variables temperature, wind speed,
  precipitation
• Spring vs summer components

6
Hypotheses
Assuming For a given location, significantly
low NDVI values at a given point in time reflect
drought effects, after controlling for phenology
and disturbance.

• Spatial patterns of drought effects differ in the
  winter/spring season (ENSO related) compared with
  late summer season (monsoon related).
• Spatial patterns of drought effects are
  influenced by non-climatic, abiotic features,
  such as geomorphic and manmade features.
• Spatial patterns of drought-affected vegetation
  influence the initiation, maintenance, expansion
  and disintegration of vegetation patches across
  the landscape

7
Study Site
Sevilleta East 63,000 ha McKenzie Flats 5650
ha
8
Vegetation
Muldavin and Milne, 1998
9
NDVI Time Series
Sevilleta East
McKenzie Flats
10
Z Statistics

• For all cells in a given period, compared by year
  (site productivity)
• For all years of a given cell, compared by period
  (cell extremes drought)

Broad scale AVHRR Local scale TM
11
AVHRR Z statistics Site Productivity
For all cells in a given period, compared by year
12
AnimationofDrought-effects

• Low NDVI compared with historical range due to
• Disturbance
• Mortality drought

No Data Zero occurrences One Two
13
AVHRR Z Statistics Drought Effects
14
Temporal Patterns of Drought Effects
Spring, Periods 9-12
Summer, Periods 17-20
Sev Station 40 Average Precipitation
Residuals from Period Average
89 90 91 92 93 94 95 96 97 98
99 00 01
Year
Time Series
15
NDVI Response at Landscape Scale
Logistic Regression Year Season Site
Productivity Elevation Precipitation from
lightning (Gosz et al. 1993, 1995) Distance
from stream Distance from road Latitude Longitude
16
Logistic Regression
Global Test of Null Hypothesis Test DF Chi-Squa
re Pr gt ChiSq Wald 18 1078.7 lt
0.0001 Effect DF Chi-Square Pr gt
ChiSq Year 12 1050.4 lt 0.0001 Season
(Summer) 2 19.1 lt 0.0001 PrecipSeaso
n 2 18.8 lt 0.0001 (-Latitude) 1
7.5 0.0062 (Distance to road) 1
4.9 0.271
17
TM Z statistics Site Productivity
5 Highest
5 Lowest
18
Animationof Cell Z-statistics
19
TM Drought Effects
10
20
NDVI Response at McKenzie Flats
21
NDVI Standard Deviation
22
Logistic Regression
Global Test of Null Hypothesis Test DF Chi-Squa
re Pr gt ChiSq Wald 1 183.4 lt
0.0001 Effect DF Chi-Square Pr gt
ChiSq -StDev 1 183.4 lt 0.0001
23
Conclusions

• Spatial patterns of drought effects differ in the
  winter/spring season (ENSO related) compared with
  late summer season (monsoon related).
  Winter/spring drought effects are less common,
  and randomly dispersed.
• Spatial patterns of drought effects are
  influenced by geomorphic (paleochannels?) and
  manmade (roads) features.
• Spatial patterns of drought-affected vegetation
  may influence patch dynamics. At the landscape
  scale, the limit of northward migration of
  creosote coincides with a northwest trending line
  of low drought-effects. At McKenzie Flats, there
  may be an interaction between drought-effects,
  subsurface flow paths and soil moisture
  retention, geomorphic features and grass type.

24
Future Work 2002 Data More statistical
analysis Geomorphic data NPP
25
Spectral Signatures at the Sevilleta
Spectral signatures courtesy of Carol Wessman and
Ann Bates
26
TM/AVHRR Differences
AVHRR 14 day Composites Maximum NDVI value for
a 14 day period. TM Revisits the site every 16
days
27
TM/AVHRR Differences
2001 Period 18
1999 Period 20
28
Climate