Vegetation variability, malnutrition, and armed conflict
in the Horn of Africa
Pedram Rowhani (McGill University)Olivier Degomme (CRED –
UCLouvain)
Debarati Guha‐Sapir (CRED –
UCLouvain)Eric Lambin (UCLouvain & Stanford)
KlimaCampus, Hamburg University, November 20, 2009
Climate change‐famine‐war
Study area
4 million km²
supporting 56 million people
3 of the world’s poorest countries hit by
conflicts and famines (> 6 million IDP)
• Sudan (states‐wilayat):
Southern Sudan since early 1980’s;
Darfur, war opposing nomad Arab
militia to different local non‐Arab
groups
• Ethiopia (zones):
The Ogaden province; Border dispute
with Eritrea; Involved in insurgency in
Somalia
• Somalia (regions‐gobollada):
No permanent national government
Data•
Complex Emergency Database (CE‐DAT): Global Acute Malnutrition
(GAM) represents the proportion of children below 2 standard deviation
from the average ratio weight‐over‐height
•
MODIS: Vegetation variability (SCV), production (iEVI), land
degradation (decreases in iEVI)
•Armed Conflicts database, version 3‐2005: Conflicts with at least 25
battle deaths/year, LAT/LONG and Radius
• Gridded Gross Domestic Product (GDP)
•Road density
Global Acute Malnutrition (GAM)The nutritional status of a population is one of the basic indicators
to assess the severity of a humanitarian crisis.
The weight and height of children between 6 and 59 months are
measured and the results are used as a proxy indicator for the
general health of the entire population.
Global Acute Malnutrition (GAM) = weight‐for‐height index less than
‐2 standard deviations from the mean weight of a reference
population of children of the same height and/or having oedema.Thresholds:
<5% = acceptable5% to 9.9% = poor10% to 14.9% = serious>15% = critical
First day of data collected by MODIS on the TERRA platform
Remote sensing
MODerate resolution Imaging Spectroradiometer (MODIS)
• MODIS data– Geo‐location ~ 50 m – 250 –
1000 m resolution– Enhanced radiometric quality– Real‐time atmospheric correction– Designed for vegetation
monitoring
• MOD43B4– Feb 2000 – DEC 2006– 1 km resolution– 16‐day composites– BRDF corrected reflectance
products (Schaaf et al. 2002)
Global MODIS Enhanced Vegetation Index
L C CGEVI
blue2red1nir
rednir
ρρρ
ρρ
Vegetation variability (1)
• Change Vector– Difference of annual
profile vectors
Change Vector
EVI Profiles
12 IICV
I1 I2
CV
Vegetation variability (2)Sum of the absolute values of the Change Vector (SCV)
Linderman et al., 2005
n
i
refi
yeari II
1SCV(year)
• Changes in amount of EVI
• Changes in phenology and/or timing of activity
Vegetation variability (3)
Mean SCV 2000‐2006 period at 1km resolution (SCV/iEVI(ref))
Vegetation availability / Degradation
integrated EVI
(iEVI)
Six year profile
1000
1500
2000
2500
3000
3500
4000
4500
1 9 17 25 33 41 49 57 65 73 81 89 97 105 113 121 129 137
Time
EVI
23
1jjEVIiEVI
Land degradation (iEVI)
MethodLogistic regressions:1.Malnutrition ~ f(Conflict)2.Malnutrition~f(SCV,iEVI,degradation,GCP,Roads)3.Conflict~f(SCV,iEVI,degradation,GCP,Roads)
Administrative &
Village level
Results (1) malnutrition ~ conflict
Logistic
regression
model results
measuring
the relationship
between
GAM
and conflict
(pseudo‐R2 = 0.19, model chi‐square = 0.0008,
area under
ROC = 0.74, AIC = 46.506)
In the Horn of Africa, GAM values over 15 are
11.7 times more likely
to be
found
in conflict
areas.
Results
(2) Malnutrition
•
Regions
with
unpredictable
vegetation
productivity
have an increased
likelihood
of acute malnutrition.
• Malnutrition
is
15.4
times more likely
to be
found
in a poor
area.
• The likelihood
of observing
malnutrition decreases
with
road density.
Multivariate
logistic
regression
model results
analyzing
the spatial
distribution of conflict
(pseudo‐R²
= 0.23/0.04, area under
ROC = 0.81/0.64)
Variable Parameter estimate Standard error P-value Odds ratio
admin. unit village admin. unit Village admin. Unit village admin. unit village
Intercept 22.61 7.36 9.85 2.06 0.0217 0.0003 - -
log10 GCP -3.57 -0.87 1.36 0.258 0.0089 0.0007 0.0283 0.419
Roads -53.55 25.28 0.0341 5.5·10-24
SCV 17.38 8.66 0.0446 3.55·1007
Results (3) Armed conflict
Multivariate logistic regression model results analyzing the spatial
distribution of conflict
(pseudo‐R2 = 0.08/0.13, area under ROC = 0.71/0.74)
• Better economic situation reduces likelihood of armed conflict.
• Armed conflicts are more likely in regions with
more vegetation.
•
Interannual variability in
vegetation
and land degradation do not
explain the presence of conflict
in the Horn of Africa.
Variable Parameter estimate Standard error P-value Odds ratio
Admin. unit village admin. unit Village admin. unit village admin. unit village
Intercept 7.235 2.108 4.08 2.63 0.0762 0.01 1387.06 8.23
log10 GCP -1.089 0.548 0.0471 0.3366
iEVI 3.6·10-05 5.73·10-05 1.50·10-05 1.16·10-05 0.0163 8.04·10-07 1.00004 1.00007
Conclusions • Malnutrition and armed conflict are closely related
•
Direct association between interannual variability in vegetation
productivity and malnutrition
• Vegetation variability indirectly associated with conflict
•
Short‐term land degradation not related to malnutrition and
armed conflict
•
Better economic situation found in peaceful areas with low
levels of malnutrition (scale independent)
• High vegetation production associated with conflict
• Different processes at different scales
• Data quality/availability!!!
Thank you!
Pedram.rowhani‐[email protected]