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