The Assessment Of The Correlation Between Poverty And Natural Disasters Based On Multi-source Data

Poverty is a serious social problem for the whole world.In 2015,700 million people around the world still lived below the extreme poverty line.Poverty reduction has been an important mission for all countries.The United Nations(UN)has proposed Sustainable Development Goals(SGDs)which include the elimination of all forms of poverty in the world.China has also proposed an important strategy of“Precise Poverty Alleviation”to help reduce poverty.In the context of global climate change,natural disasters pose a huge threat to the elimination of poverty.Compared with other groups of people,poor people have higher social vulnerability.They are more likely to suffer from natural disasters and more difficult to recover.In the past 50 years,the number of natural disasters in the world has shown an increasing trend.The situation of poor people at risk of natural disasters and the impact of natural disasters on the poor need to receive more attention.Assessing the correlation between poverty and natural disasters is of great significance for understanding the living conditions of the poor,targateing poverty alleviation and poverty reduction programs,and promoting social equity.The analysis of the correlation between natural disasters and poverty requires poverty datasets and natural disaster datasets.Traditional ways of poverty measurements largely rely on statistical and survey data.However,most statistical data used administrative regions as the units and cannot reflect the distribution of poverty within the administrative region while survey data have problems such as low spatial coverage and low slow update frequency.Furthermore,countries that are extremely poor or in war can even lack of these statistical and survey data for years.Current research used remote sensing data,GIS data,and other statistical data to estimate poverty instead.Given that the causes of poverty and the characteristics of poor households are complex,poverty variation is difficult to explain by a single data type.On the other hand,the poverty estimation using multi-source were limited by the availability and computational cost of the source data.In addition,current poverty statistics in different countries are inconsistent,leading most studies to focus on poverty estimation in individual countries or regions,and lack of large-scale,cross-country poverty estimation methods.Therefore,it is necessary to propose a method that can quickly integrate multi-source remote sensing and GIS data to estimate poverty and improve the accuracy of poverty estimation.Current research on the exposure of poor people to natural disasters as well as the correlation between poverty and different types of natural disasters is limited due to the lack of spatial explicit data on poverty.In addition,although nighttime day data has been proven to be useful for assessing the impact of some natural disasters,it is still necessary to evaluate the usefulness and limitations of this source of data when assessing the impact of different types of natural disasters.At the same time,few studies have explored the correlation between poverty and the recoverty rate after disasters.To fill these research gaps,this study proposed a machine learning model to estimate poverty distribution by combining features extracted from multiple data sources and explored the estimation ability of the model in a single country,multiple countries and a city.This study then evaluated the correlation between poverty and natural disasters.At last,this study summarized the usefulness and limitations of the NPP(National Polar-Orbiting Partnership)–VIIRS(Visible Infrared Imaging Radiometer Suite)nighttime light(NTL)data when assessing the impact of natural disasters.The main research contents and results of this paper are as follows:(1)This study proposed a machine learning model to estimate poverty.Firstly,various features were extracted from multiple data sources,including NTL data,land cover data,high-spatial resolution Google satellite imagery,accessibility map,and Digital Elevation Model(DEM)data.These features were classified into 4 types which were socioeconomic features,land cover features,topographic features,and accessibility features.Secondly,a Random Forest Regression(RFR)model was developed and trained using these features as independent variables to estimate the Wealth Index(WI)as the dependent variable for a single country(Bangladesh),multiple countries(South Asia and Southeast Asia)and a city(Vishakhapatnam,India).The R~2 between the actual and estimated WI was 0.71 in Bangladesh,0.61 in South Asia and Southeast Asia,and 0.63 in Visakhapatnam,respectively.Compared with traditional research,this study improved the accuracy and generalization ability of the poverty estimation.(2)From the perspective of pre-disaster rick assessment,this study assessed the hazardous distribution of four natural disasters including landslides,earthquakes,floods,and tropical cyclones in low-income and middle-low-income countries in South and Southeast Asia.This study then assessed the exposure of poor area to natural disasters and the correlation between poverty and the hazardous level of natural disasters.Specifically,the exposure to natural disasters in poor areas was first analyzed.14.64%of the poor areas were found to be at high risk of at least one natural disaster.Then,this study analyzed the distribution of poverty in high-hazard areas of natural disasters.It is found that the proportion of poor areas in high-hazard areas were generally large,but the population was relatively sparse.At last,the relationship between the hazardous level of natural disasters and the poverty level was analyzed.The results show that for earthquakes and landslides,the higher the hazardous level of disasters,the poorer the place.However,for floods and tropical cyclones,the higher the hazardous level of disasters,the richer the place in most countries.Only in India,the higher the hazardous level of disasters,the poor the place.These results were related to the nature of disasters and the countries’ability to prevent and respond to disasters.(3)From the perspective of post-disaster assessment,this study explored the application of daily NPP-VIIRS NTL data in assessing natural disasters and then analyzed the impact of the tropical cyclone Hudhud on poor areas in the city of Visakhapatnam.Specifically,this study used a Percent of Normal Light(PNL)index to measure the differences between the pre-and post-disaster NTL images.Then the usefulness and limitation of PNL in assessing the impact of natural disasters were analyzed in six typical disasters,including earthquakes,tropical cyclones,and floods.The analyses showed that PNL were useful for detecting damages caused by the Nepal Earthquake with an overall accuracy of 75.5%Kappa of 0.31.However,it was not possible to use PNL to detect the damage after the Italy Earthquake.In the case of Hurricane Maria,the PNL value was significantly correlated with the power outage rate(R~2=0.94).In the case of tropical cyclone Hudhud,the power outage rate estimated by PNL was in good agreement with the reported power outage rate.In the flood events,power outages were detected by NTL after the Yulin Flood,but were not detected after the Louisiana Flood.The analyses of power outages and restorations in Visakhapatnam after cyclone Hudhud showed that the power outage condition in poorer areas and richer areas were similar,but the power restoration in poorer areas was lower than that in richer areas.