Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Reply to Comment on ‘High-income does not protect against hurricane losses’

Reply to Comment on ‘High-income does not protect against hurricane losses’ Any further distribution Recently a multitude of empirically derived damage models have been applied to project future of this work must maintain attribution to tropical cyclone (TC) losses for the United States. In their study (Geiger et al 2016 Environ. Res. Lett. the author(s) and the title of the work, journal 11 084012) compared two approaches that differ in the scaling of losses with socio-economic drivers: citation and DOI. the commonly-used approach resulting in a sub-linear scaling of historical TC losses with a nation’s affected gross domestic product (GDP), and the disentangled approach that shows a sub-linear increase with affected population and a super-linear scaling of relative losses with per capita income. Statistics cannot determine which approach is preferable but since process understanding demands that there is a dependence of the loss on both GDP per capita and population, an approach that accounts for both separately is preferable to one which assumes a specific relation between the two dependencies. In the accompanying comment, Rybski et al argued that there is no rigorous evidence to reach the conclusion that high-income does not protect against hurricane losses. Here we affirm that our conclusion is drawn correctly and reply to further remarks raised in the comment, highlighting the adequateness of our approach but also the potential for future extension of our research. Recently, we applied various empirically derived dam- on one of the conflicting findings only. As correctly age models to project future TC losses for the United observed by RPK2017 both approaches only slightly States [1]. In particular, we distinguished between two differ with respect to their explanatory power. However, model types that differ with respect to their scaling of only one approach analyses the scaling of income with socio-economic drivers, i.e. using total affected GDP losses explicitly, while in the other approach income is as a single predictor or separating the effect of affected only implicitly accounted for via total GDP. In this sense population and average per capita GDP. Although it is correct to say that ‘high income does not protect statistics cannot determine which approach is prefer- against hurricane losses’, asthissuper-linear scaling able, we argue that in terms of process understanding with rising income remains hidden if GDP is used as a there exists a dependence of the loss on both GDP single predictor, due to the very pronounced sub-linear per capita and population, such that an approach that scaling of losses with population. This finding has also accounts for both separately is preferable to one which been confirmed recently [2]. assumes a specific relation between the two dependen- Generally, the availability of socio-economic data cies. Interestingly, we found that separating population limits our analysis to a certain temporal period. In par- and per capita GDP (i.e. income) results in a super- ticular the TC-affected average income does not vary as linear increase of TC losses with income, leading to the strongly as e.g. TC-affected population between 1963 conclusion that high-income does not protect against and 2012, as correctly observed by RPK2017, mak- hurricane losses. ing it a worthwhile endeavor to explore potentially Thereafter Rybski et al (hereafter RPK2017) argued different functional scalings of socio-economic drivers that our conclusion needs to be revisited as it is based with hurricane losses in future research. This point © 2017 The Author(s). Published by IOP Publishing Ltd Environ. Res. Lett. 12 (2017) 098002 TGeiger et al directly links to the question raised by RPK2017 on how using hurricane wind-field extension. Our analysis does well can hurricane damage be predicted at all. The var- not distinguish between urban areas and rural regions ious drivers of TC damage are complex and have been and, due to large hurricane sizes, encompasses various controversially discussed in the literature, including forms of settlements. subtle elements as the angle of landfall, the associated In conclusion, our original paper has made an effort precipitation, and the storm’s duration or gustiness [3]. to ease comparability between various TC damage Attempts to build models with higher certainty would models and has shifted the emphasis to the socio- consequently require many more predictors, most of economic drivers, in particular the scaling of losses with which would be unavailable for future projections. On income. However and in agreement with RPK2017, the more aggregate level, as in our approach, devia- future research is needed to understand the origin tions remain quite large but smaller than in previous of this scaling, to reduce model uncertainties and to attempts [4, 5], where explained variances are smaller improve the accuracy of future projections. and uncertainty is usually not discussed. Moreover, we also did not aim to provide the best possible dam- Acknowledgments age model but rather re-applied models that have been proposed in various contexts previously [2, 4–8], in a The work was supported within the framework of the Leibniz Competition (SAW-2013-PIK-5 and SAW- consistent setup. The fact that all analyzed models only slightly differ in the explained variance also illustrates 2016-PIK-1). that previous discussions about best model types, in ORCID iDS particular with respect to wind-speed scaling, are mis- leading, and have overlooked the importance of the Tobias Geiger https://orcid.org/0000-0002-8059- socio-economic component. In this respect, the cur- rently missing propagation of model-uncertainties to projections, as argued by RPK2017, would not alter the main findings of our paper, i.e. the large divergence References between both basic model types. [1] Geiger T, Frieler K and Levermann A 2016 High-income does RPK2017 also raised a concern about the appar- not protect against hurricane losses Environ. Res. Lett. 11 ent discontinuity in GDP-by-state time series at 1997, introduced by the transition from SIC to NAICS [2] Bakkensen L A and Mendelsohn R O 2016 Risk and adaptation: evidence from global hurricane damages and fatalities J. Assoc. industry definitions. While merging both series we Environ. Resour. Econ. 3 555–87 encountered no significant changes in levels (for the [3] Czajkowski J and Done J 2014 As the wind blows? overlapping year 1997) and trends in growth rates for Understanding hurricane damages at the local level through a total real GDP-by-state. We therefore see no reason case study analysis Weather Clim. Soc. 6 202–17 [4] Schmidt S, Kemfert C and Ho¨ppe P 2009 The impact of that model performance may be obscured by the GDP socio-economics and climate change on tropical cyclone losses data. As a side remark, the discontinuity, however, in the USA Reg. Environ. Change 10 13–26 exists when looking at specific industries due to the [5] Nordhaus W D 2010 The economics of hurricanes and reshuffling of the various contributing sectors. implications of global warming Clim. Change Econ. 1 1–20 [6] Mendelsohn R, Emanuel K, Chonabayashi S and Bakkensen L Finally, RPK2017 pointed out that a recently 2012 The impact of climate change on global tropical cyclone reported power-law correlation between urban GDP damage Nat. Clim. Change 2 205–9 and population [9] could be used to unite both basic [7] Zhai A R and Jiang J H 2014 Dependence of US hurricane model types. While providing an interesting insight economic loss on maximum wind speed and storm size Environ. Res. Lett. 9 064019 we doubt that there exists a one-to-one correspon- [8] Murnane R J and Elsner J B 2012 Maximum wind speeds and dence between our analysis and [9], questioning the US hurricane losses Geophys. Res. Lett. 39 707 applicability of equation [6] from RPK2017. Where [9] Bettencourt L M, Lobo J, Helbing D, Kuh ¨ nert C and West G B [9] relates a city’s population to their urban economic 2007 Growth, innovation, scaling, and the pace of life in cities Proc. Natl Acad. Sci. USA 104 7301–6 activity, we determine exposed population and GDP http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Environmental Research Letters IOP Publishing

Reply to Comment on ‘High-income does not protect against hurricane losses’

Loading next page...
 
/lp/iop-publishing/reply-to-comment-on-high-income-does-not-protect-against-hurricane-nInvTw2WCa

References (42)

Copyright
Copyright © 2017 The Author(s). Published by IOP Publishing Ltd
eISSN
1748-9326
DOI
10.1088/1748-9326/aa88d6
Publisher site
See Article on Publisher Site

Abstract

Any further distribution Recently a multitude of empirically derived damage models have been applied to project future of this work must maintain attribution to tropical cyclone (TC) losses for the United States. In their study (Geiger et al 2016 Environ. Res. Lett. the author(s) and the title of the work, journal 11 084012) compared two approaches that differ in the scaling of losses with socio-economic drivers: citation and DOI. the commonly-used approach resulting in a sub-linear scaling of historical TC losses with a nation’s affected gross domestic product (GDP), and the disentangled approach that shows a sub-linear increase with affected population and a super-linear scaling of relative losses with per capita income. Statistics cannot determine which approach is preferable but since process understanding demands that there is a dependence of the loss on both GDP per capita and population, an approach that accounts for both separately is preferable to one which assumes a specific relation between the two dependencies. In the accompanying comment, Rybski et al argued that there is no rigorous evidence to reach the conclusion that high-income does not protect against hurricane losses. Here we affirm that our conclusion is drawn correctly and reply to further remarks raised in the comment, highlighting the adequateness of our approach but also the potential for future extension of our research. Recently, we applied various empirically derived dam- on one of the conflicting findings only. As correctly age models to project future TC losses for the United observed by RPK2017 both approaches only slightly States [1]. In particular, we distinguished between two differ with respect to their explanatory power. However, model types that differ with respect to their scaling of only one approach analyses the scaling of income with socio-economic drivers, i.e. using total affected GDP losses explicitly, while in the other approach income is as a single predictor or separating the effect of affected only implicitly accounted for via total GDP. In this sense population and average per capita GDP. Although it is correct to say that ‘high income does not protect statistics cannot determine which approach is prefer- against hurricane losses’, asthissuper-linear scaling able, we argue that in terms of process understanding with rising income remains hidden if GDP is used as a there exists a dependence of the loss on both GDP single predictor, due to the very pronounced sub-linear per capita and population, such that an approach that scaling of losses with population. This finding has also accounts for both separately is preferable to one which been confirmed recently [2]. assumes a specific relation between the two dependen- Generally, the availability of socio-economic data cies. Interestingly, we found that separating population limits our analysis to a certain temporal period. In par- and per capita GDP (i.e. income) results in a super- ticular the TC-affected average income does not vary as linear increase of TC losses with income, leading to the strongly as e.g. TC-affected population between 1963 conclusion that high-income does not protect against and 2012, as correctly observed by RPK2017, mak- hurricane losses. ing it a worthwhile endeavor to explore potentially Thereafter Rybski et al (hereafter RPK2017) argued different functional scalings of socio-economic drivers that our conclusion needs to be revisited as it is based with hurricane losses in future research. This point © 2017 The Author(s). Published by IOP Publishing Ltd Environ. Res. Lett. 12 (2017) 098002 TGeiger et al directly links to the question raised by RPK2017 on how using hurricane wind-field extension. Our analysis does well can hurricane damage be predicted at all. The var- not distinguish between urban areas and rural regions ious drivers of TC damage are complex and have been and, due to large hurricane sizes, encompasses various controversially discussed in the literature, including forms of settlements. subtle elements as the angle of landfall, the associated In conclusion, our original paper has made an effort precipitation, and the storm’s duration or gustiness [3]. to ease comparability between various TC damage Attempts to build models with higher certainty would models and has shifted the emphasis to the socio- consequently require many more predictors, most of economic drivers, in particular the scaling of losses with which would be unavailable for future projections. On income. However and in agreement with RPK2017, the more aggregate level, as in our approach, devia- future research is needed to understand the origin tions remain quite large but smaller than in previous of this scaling, to reduce model uncertainties and to attempts [4, 5], where explained variances are smaller improve the accuracy of future projections. and uncertainty is usually not discussed. Moreover, we also did not aim to provide the best possible dam- Acknowledgments age model but rather re-applied models that have been proposed in various contexts previously [2, 4–8], in a The work was supported within the framework of the Leibniz Competition (SAW-2013-PIK-5 and SAW- consistent setup. The fact that all analyzed models only slightly differ in the explained variance also illustrates 2016-PIK-1). that previous discussions about best model types, in ORCID iDS particular with respect to wind-speed scaling, are mis- leading, and have overlooked the importance of the Tobias Geiger https://orcid.org/0000-0002-8059- socio-economic component. In this respect, the cur- rently missing propagation of model-uncertainties to projections, as argued by RPK2017, would not alter the main findings of our paper, i.e. the large divergence References between both basic model types. [1] Geiger T, Frieler K and Levermann A 2016 High-income does RPK2017 also raised a concern about the appar- not protect against hurricane losses Environ. Res. Lett. 11 ent discontinuity in GDP-by-state time series at 1997, introduced by the transition from SIC to NAICS [2] Bakkensen L A and Mendelsohn R O 2016 Risk and adaptation: evidence from global hurricane damages and fatalities J. Assoc. industry definitions. While merging both series we Environ. Resour. Econ. 3 555–87 encountered no significant changes in levels (for the [3] Czajkowski J and Done J 2014 As the wind blows? overlapping year 1997) and trends in growth rates for Understanding hurricane damages at the local level through a total real GDP-by-state. We therefore see no reason case study analysis Weather Clim. Soc. 6 202–17 [4] Schmidt S, Kemfert C and Ho¨ppe P 2009 The impact of that model performance may be obscured by the GDP socio-economics and climate change on tropical cyclone losses data. As a side remark, the discontinuity, however, in the USA Reg. Environ. Change 10 13–26 exists when looking at specific industries due to the [5] Nordhaus W D 2010 The economics of hurricanes and reshuffling of the various contributing sectors. implications of global warming Clim. Change Econ. 1 1–20 [6] Mendelsohn R, Emanuel K, Chonabayashi S and Bakkensen L Finally, RPK2017 pointed out that a recently 2012 The impact of climate change on global tropical cyclone reported power-law correlation between urban GDP damage Nat. Clim. Change 2 205–9 and population [9] could be used to unite both basic [7] Zhai A R and Jiang J H 2014 Dependence of US hurricane model types. While providing an interesting insight economic loss on maximum wind speed and storm size Environ. Res. Lett. 9 064019 we doubt that there exists a one-to-one correspon- [8] Murnane R J and Elsner J B 2012 Maximum wind speeds and dence between our analysis and [9], questioning the US hurricane losses Geophys. Res. Lett. 39 707 applicability of equation [6] from RPK2017. Where [9] Bettencourt L M, Lobo J, Helbing D, Kuh ¨ nert C and West G B [9] relates a city’s population to their urban economic 2007 Growth, innovation, scaling, and the pace of life in cities Proc. Natl Acad. Sci. USA 104 7301–6 activity, we determine exposed population and GDP

Journal

Environmental Research LettersIOP Publishing

Published: Sep 1, 2017

There are no references for this article.