View Mark Montgomery's presentation.

advertisement
The Demography of ‘Natural’ Disasters
What Can We Learn from DesInventar?
Mark R. Montgomery
Stony Brook University and Population Council
April 20, 2015
Montgomery
April 20, 2015
1 / 44
Outline
1
Overview
2
Disaster Recording: National versus Spatially-Specific
3
Exposure Estimates from Global Hazards Data and Models
4
Local Disaster Records: DesInventar
5
The Demographics of Risk Exposure
6
Conclusions and Next Steps
Montgomery
April 20, 2015
2 / 44
Overview
The Hyogo Framework for Action, 2005–2015
This international initiative aims to:
Fully integrate disaster planning, early warning, preparedness, and
vulnerability reduction into economic development strategies
Ensure that multiple levels and units of government are well
coordinated and linked with NGOs, community-based organizations,
and civil society
Repeatedly emphasizes integration at the level of the community,
need to understand risk perceptions and constraints on responses at
the level of families and neighborhoods
But what data will support an evidence-based approach?
United Nations (2009) Global Assessment Report on Disaster Risk
Reduction: Risk and Poverty in a Changing Climate an excellent
overview; other GARs have followed.
Montgomery
April 20, 2015
3 / 44
Overview
Looming Extreme-Event Threats
Sudden-onset extreme weather events: typhoons, heavy precipitation,
coastal and interior flooding, landslides—these disasters believed to
disproportionately harm women and children and there may be
important differences in exposure, vulnerability, and resilience by
education as well.
Gradual-onset conditions: Less is known about droughts and in arid
regions, increasing water scarcity. Implications for rural and urban
dwellers (especially the poor) and for rural-urban migration.
Montgomery
April 20, 2015
4 / 44
Overview
Adaptation to Extreme Events and Climate Change
National, regional, and municipal governments in poor countries will
need adaptation strategies that are spatially-specific and
evidence-based, focusing on individual cities and neighborhoods
within them.
Exposure to climate-related risks is being studied systematically by
bio-geophysical scientists at the level of world regions and sub-regions.
But exposure varies greatly sub-nationally and across cities and their
neighborhoods—these smaller-scale differences are not being
systematically examined by social scientists.
Montgomery
April 20, 2015
5 / 44
Overview
The Urban Dimension
National climate adaptation and risk-reduction plans often ignore city dwellers
Poor countries are approaching urban majorities
Some extreme-weather events (e.g., floods) repeatedly strike
city-dwellers, harming lives and damaging assets and livelihoods
Apart from national censuses, demographers collect little or no
city-specific data
Majority of urban residents live in small- and intermediate-size cities,
where officials have limited abilities to anticipate risks and guide local
development.
Substantial mobility and both rural-to-urban and urban-to-urban
migration: changes of neighborhood and locale generate a sequence
of place-specific exposures.
Montgomery
April 20, 2015
6 / 44
Overview
What Role for Demographers?
How Do We Link Our Socoeconomic Data to Hazard Exposure?
This is the major challenge facing demographers. We have much to
contribute, but we must organize our data to join the adaptation and
risk-reduction conversations. Population censuses can make a vital
contribution—but only if they are fully disaggregated. Longitudinal
surveys—if properly designed (or retrofitted) with spatial detail—can
address issues about short-term and longer-run consequences.
Montgomery
April 20, 2015
7 / 44
Overview
Mapping Demographic Data
1
Measures of built-up area (structures) from Martino Pesaresi’s GHSL
rasters, from 1975 to the present using LandSat and other sensors.
(Time-series!)
2
Measures of large-city population size and growth rates (UN
Population Division) and increasing coverage of smaller cities and
towns. (Time-series!)
3
High resolution population density rasters, from Andy Tatem’s
WorldPop team, see http://www.worldpop.org.uk/
4
Economic activity and city spatial extents: Night-time lights from
NOAA, both OLS (Time-series!) and the newer Viirs.
5
Sub-national census and survey (IPUMS, DHS, MICS) data on
population numbers and composition (Time-series at admin level)
6
DHS sampling cluster coordinates: fairly precise spatially (Not exactly
a time-series, but multiple points in time)
Montgomery
April 20, 2015
8 / 44
Overview
Themes and Fundamental Issues
Selectivity biases in disaster reporting. What kinds of extreme events
go under-reported? What are the spatial and social differentials in
reporting?
When is spatial specificity not enough? Droughts versus floods.
Characteristics of those exposed to risk and those actually harmed
Montgomery
April 20, 2015
9 / 44
Disaster Recording: National versus Spatially-Specific
Where, Exactly, Have Disasters Struck?
Improving the Empirical Record
Surprisingly, the spatial locations of disasters have not been recorded very
specifically.
EM-DAT : The OFDA/CRED International Disaster Database.
National-level disaster reports are available here:
http://www.emdat.be. In recent years, sub-national units where
disasters occurred are listed, but effects not disaggregated.
Much of the literature on disaster exposure and consequences has
employed whole countries as the units of analysis. Poor, less poor, and
higher-income countries; High, moderate, and low exposure to risks.
Helpful, but hardly enough.
Montgomery
April 20, 2015
10 / 44
Disaster Recording: National versus Spatially-Specific
Disasters in Guatemala: EM-DAT (2014)
1
2
3
4
5
6
7
8
9
10
Montgomery
Disaster
Earthquake
Drought
Earthquake
Flood
Storm
Storm
Drought
Flood
Drought
Storm
Date
4/2/1976
3/1/2009
7/11/2012
12/10/2011
1/10/2005
28/05/2010
6/1/2012
22/10/2008
9/1/2001
26/10/1998
Total Affected
4,993,000
2,500,000
1,321,742
528,753
475,314
397,962
266,485
180,000
113,596
105,700
April 20, 2015
11 / 44
Disaster Recording: National versus Spatially-Specific
The DesInventar Program
Since the late 1980s in Latin America and a smattering of other countries
in Africa and Asia, a coherent system of comprehensive, small-scale
disaster reporting has been in place, with protocols, software, and
public-domain data: http://www.desinventar.org.
Curiously under-exploited by researchers to date, apart from thin
summaries in GAR (2013). Not clear whether funding exists to sustain the
initiative.
Montgomery
April 20, 2015
12 / 44
Disaster Recording: National versus Spatially-Specific
The Core of the DesInventar Program
Other countries (not shown) have tested parts of the protocol: see GAR (2013)
Montgomery
April 20, 2015
13 / 44
Disaster Recording: National versus Spatially-Specific
Disasters in Guatemala: DesInventar 1988–2011
Estimates from 5000+ spatially-specific reports
Type of Harm
Deaths
Injuries
Otherwise Harmed
Missing
Evacuated
Relocated
Indirectly Harmed
Houses Destroyed
Houses Damaged
Montgomery
Flooding
882
1,204
1,042,530
991
507,723
20
1,940,733
Landslides
836
808
26,409
131
22,255
779
1,668,509
26,426
153,180
1,823
3,490
April 20, 2015
14 / 44
Disaster Recording: National versus Spatially-Specific
The DesInventar Approach
Event occurs (e.g., storm)
No human harm ⇒ No record of event (Design bias)
Human harm occurs:
But not reported in the media ⇒ No record of harm (Reporting biases)
Reported in the media ⇒ Disaster record(s) created via a
well-conceived process of:
Verification by two or more coders
Spatial pinpointing with a record created for each spatial
(administrative) unit affected
Coding at most disaggregate level possible given the verified data
Notes on street addresses often included for urban events
Multiple types of harm detailed for each unit
Montgomery
April 20, 2015
15 / 44
Disaster Recording: National versus Spatially-Specific
Shortcomings
Issues to consider in designing future reporting systems
What is the reach of the media?
What events and harms are deemed “newsworthy”?
As we’ll see, improves greatly on the who is exposed to risk analysis,
but —
Does not address the fundamental problem affecting most disaster
data: Who is harmed? Sex, age, education, poverty, . . .
Montgomery
April 20, 2015
16 / 44
Exposure Estimates from Global Hazards Data and Models
Built-Up Area from GHSL, 300m Resolution
Percent_1975
Percent_1990
100
80
60
Percent_2000
Percent_2014
40
20
0
Montgomery
April 20, 2015
17 / 44
Exposure Estimates from Global Hazards Data and Models
Urban and Rural Structures: GHSL
Percent_1975
Percent_1990
100
80
60
Percent_2000
Percent_2014
40
20
0
Montgomery
April 20, 2015
18 / 44
Exposure Estimates from Global Hazards Data and Models
Populations Densities from the WorldPop Raster
50
45
40
35
30
25
20
15
10
Montgomery
April 20, 2015
19 / 44
Exposure Estimates from Global Hazards Data and Models
VIIRS High-Resolution Lights
Even more detailed measures of structures will appear in the next 2 years . . .
8
7
6
5
4
3
2
1
Montgomery
April 20, 2015
20 / 44
Exposure Estimates from Global Hazards Data and Models
Guatemalan Cities (population 10,000+) and VIIRS Lights
8
7
6
5
4
3
2
1
Montgomery
April 20, 2015
21 / 44
Exposure Estimates from Global Hazards Data and Models
Probabilistic Models of Risk Exposure
I use the 2005 “Hotspots” measures put into the public domain by
the World Bank, CIESIN at Columbia University, and a host of
partners. For data and documentation, see
http://sedac.ciesin.columbia.edu/data/collection/ndh
Provides model-based estimates for a range of risks; global coverage
Spatial resolution often coarse, but gives a starting-point for more
detailed country-specific research
Montgomery
April 20, 2015
22 / 44
Exposure Estimates from Global Hazards Data and Models
Locating Hazards using Models: Hurricane Risks
From storm tracks; Decile among all exposed areas world-wide
9
8
7
1
Montgomery
April 20, 2015
23 / 44
Exposure Estimates from Global Hazards Data and Models
Landslide Risks
Probability models use slope, soil type and moisture
11
10
9
8
7
6
Montgomery
April 20, 2015
24 / 44
Exposure Estimates from Global Hazards Data and Models
River Flooding Risks
Dartmouth Flood Observatory model
21
20
19
17
14
13
10
7
2
Montgomery
April 20, 2015
25 / 44
Exposure Estimates from Global Hazards Data and Models
Drought
Montgomery
April 20, 2015
26 / 44
Exposure Estimates from Global Hazards Data and Models
Estimated Population at Risk: Total and Urban
Using WorldPop raster and city-specific populations
Risk
Total Population
Hurricanes
1st decile
10,367,818
7th decile
45,420
8th decile
2,476
Landslides
Category 6
3,716,903
Category 7
1,633,098
Category 8
1,772,364
Category 9
823,010
Category 10
524,186
Inland Flood Risk
“Low”
275,702
“Medium”
79,761
“High”
2496
Drought
12,145,434
Montgomery
City and
Town Population
3,597,841
10,299
1,418,933
214,657
257,605
134,055
68,566
61,123
40,274
3,819,705
April 20, 2015
27 / 44
Local Disaster Records: DesInventar
Flood Events, 1988–2011, Recorded by DesInventar
Units are municipios
160
55
52
45
41
37
33
32
31
29
26
25
24
23
22
21
20
19
18
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
Montgomery
April 20, 2015
28 / 44
Local Disaster Records: DesInventar
Recorded Landslides
202
34
28
27
25
23
20
15
14
12
11
10
9
8
7
6
5
4
3
2
1
0
Montgomery
April 20, 2015
29 / 44
Local Disaster Records: DesInventar
Homes Destroyed by Floods, 1988–2011
1680
1504
1177
953
943
888
734
707
617
586
577
562
555
551
500
486
480
470
451
443
420
415
409
400
390
360
353
316
305
300
271
254
249
235
200
159
151
150
148
144
139
136
134
130
120
118
117
107
106
102
100
98
96
92
82
80
78
75
73
72
71
69
68
64
63
58
56
55
54
52
50
48
45
40
38
37
36
34
33
30
29
28
27
24
23
22
21
20
18
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
Montgomery
April 20, 2015
30 / 44
Local Disaster Records: DesInventar
Homes Damaged by Floods
22000
9700
8576
7028
5966
5941
4568
4503
3343
3325
3168
2615
2354
2146
2113
2081
2066
1779
1761
1717
1617
1585
1439
1419
1300
1258
1135
1070
1015
1010
900
850
723
665
626
600
585
580
575
573
537
504
469
463
425
424
418
401
384
378
367
365
358
342
335
329
325
322
318
315
277
276
270
269
265
257
250
244
233
217
215
214
213
201
200
192
191
185
180
173
170
167
160
158
156
153
145
140
139
135
132
128
126
121
119
117
112
110
108
107
100
94
90
85
84
81
77
75
74
73
70
69
68
66
64
60
56
54
53
52
51
50
47
46
45
44
43
42
41
40
36
35
34
33
32
31
30
29
28
27
26
24
23
21
20
19
16
15
13
12
11
10
9
8
5
4
3
2
1
0
Montgomery
April 20, 2015
31 / 44
Local Disaster Records: DesInventar
Homes Destroyed by Landslides, 1988–2011
454
140
125
110
85
80
69
53
47
41
39
36
33
31
28
25
24
23
21
19
18
16
15
13
11
9
8
7
6
5
4
3
2
1
0
Montgomery
April 20, 2015
32 / 44
Local Disaster Records: DesInventar
Homes Damaged by Landslides
432
395
336
260
217
166
147
110
92
88
81
62
60
51
50
47
46
45
38
35
34
31
27
26
23
20
18
16
15
13
11
8
7
6
5
4
3
2
1
0
Montgomery
April 20, 2015
33 / 44
Local Disaster Records: DesInventar
Flooding and landslides within Guatemala City, 1988–2011
By the zonas of the municipio
25
19
24
15
22
13
18
8
12
7
9
6
6
5
5
4
4
3
3
2
2
1
1
0
0
Montgomery
April 20, 2015
34 / 44
Local Disaster Records: DesInventar
Garbage pickers in a Guatemala City ravine
Montgomery
April 20, 2015
35 / 44
Local Disaster Records: DesInventar
Pinpointing disasters in Guatemala City by address
Parts of Zone 7 as seen in Google Maps
Montgomery
April 20, 2015
36 / 44
The Demographics of Risk Exposure
We’ve Seen Where Risky Events Occur—but Who is
Exposed to Them?
Using 2002 Guatemalan Population Census (we have the full
micro-records) we can characterize all municipios and (eventually, we
hope) the zonas of Guatemala City by: education, age, sex, proxies
for poverty in both urban and rural settings.
A 2002 small-area poverty map, based on a living standards survey
and the census data, also summarizes municipios by mean
consumption and indexes of poverty and inequality.
Problem: Both sources represent Guatemala City as a whole. Neither
disaggregates it even by zonas—despite the well-known and enormous
within-city inequalities! Within-city differentials are commonly
overlooked!.
Montgomery
April 20, 2015
37 / 44
The Demographics of Risk Exposure
Illustration: Risk Exposure Differentials by Education
2002 Population Census provides completed schooling, by levels and
grades within level. Examine “no schooling” and “any secondary or
tertiary”. Census also records municipio of residence.
Linking these records on people to DesInventar records on
municipios, and thereby to counts of disasters, we can:
Estimate the average number of disasters by schooling level
Investigate place-of-residence and environmental risk association: For
each schooling level, are people likelier to live in lower-risk municipios,
or in higher-risk municipios.
A very simple descriptive analysis follows—it reveals the consequences
of naively representing all Guatemala City as one unit, rather than
disaggregating this highly unequal city by within-city zone.
Montgomery
April 20, 2015
38 / 44
The Demographics of Risk Exposure
s be the proportion of schooling group s living in municipio m.
Let πm
Let Tm be the risk index for municipio m (e.g., total disasters over
1988–2011).
P
s
As = M
m=1 πm · Tm is the average risk for schooling group s.
We can correlate

 s
T1
π1
 T2 
 πs 
 
 2
 ..  and  .. 
 . 
 . 

TM
s
πM
to see whether people with s schooling are likelier to live in lower-risk
municipios or higher-risk municipios. Then we’ll compare across schooling
groups.
Montgomery
April 20, 2015
39 / 44
The Demographics of Risk Exposure
All municipios For flooding (and also true for landslides):
Anone = 12.5 floods over 1988–2011
Asecondary + = 47.5 floods
none and T flood risk: 0.54
Correlation between πm
m
secondary +
Correlation between πm
and Tm flood risk: 0.77
Consequence of aggregating the zones of Guatemala City!
All municipios except Guatemala City Again for flooding:
Anone = 8.4 floods over 1988–2011
Asecondary + = 14.1 floods
none and T flood risk: 0.37
Correlation between πm
m
secondary +
Correlation between πm
and Tm flood risk: 0.38
Montgomery
April 20, 2015
40 / 44
The Demographics of Risk Exposure
What Does This Illustration Teach Us?
1
The unfortunate tendency to ignore within-city heterogeneity (in SES
and in incidence of hazards) can produce highly misleading estimates
of SES differentials in risk exposure
2
Had the zonas of Guatemala City been identified in the Census, a
different risk profile for the educated and uneducated would have
emerged
3
But—even with Guatemala City out of the picture—there is
suggestive evidence of positive associations of areal risk and individual
education.
4
Spatial specificity is not enough. Even if the educated do tend live in
riskier areas, they may be able to protect themselves against harm,
safeguard their own assets, and rebound more quickly than the
less-educated living in the same areas.
Montgomery
April 20, 2015
41 / 44
Conclusions and Next Steps
Next Steps
Analysis is underway of all core DesInventar datasets. Not every
country supplies micro-level census records, but much can be done to
map socioeconomic exposure even without them.
The low-elevation coastal zone (not very important in Guatemala)
will be a significant feature in many of these countries.
Montgomery
April 20, 2015
42 / 44
Conclusions and Next Steps
Data Issues Needing Attention
What other countries are maintaining DesInventar-like records of
disaster events? What is going on in Asia?
How successful are any of these efforts in locating gradual-onset
conditions and identifying the harm coming from them? I am skeptical
about whether DesInventar adequately measures the incidence of
droughts. How often is a sustained drought “newsworthy”?
Census data can describe the neighborhoods in which disasters took
place, but not the characteristics of affected individuals and
households. Within-neighborhood heterogeneity will remain a
challenge even in estimating exposure to risk.
Montgomery
April 20, 2015
43 / 44
Conclusions and Next Steps
Not even DesInventar systematically collects records of harm
according to sex, age, education of the affected people. Relief and
humanitarian agencies evidently unable to fill this data gap. Who can?
Potential of longitudinal studies to contribute—if they link in spatially
detailed hazard event information, and craft questions that identify
individuals and households who were harmed (not simply in the
vicinity of a hazardous event). Big challenges in dealing with mobility,
migration, and mortality.
Some extreme events—droughts—can do damage at a distance (for
instance, by raising food prices thereby harming the urban poor) as
well as in the immediate vicinity. Spatial specificity helps, but is not
enough.
Montgomery
April 20, 2015
44 / 44
Download