Christopher Doll JSPS-UNU Postdoctoral Fellow United Nations - PowerPoint PPT Presentation

Population detection profiles of DMSP- OLS night-time imagery by regions of the world Christopher Doll JSPS-UNU Postdoctoral Fellow United Nations University – Institute of Advanced Studies & University of Tokyo Asia-Pacific Advanced Network Hanoi Monday 9 th August 2010

Overview � Background: Using night-time lights to estimate access to electricity access � (Known) unknowns with using night-time light imagery � Population detection by region � Ambient vs residential population � Population density and DN � Summary and final thoughts

Night-time lights and electricity access � Energy access is often regarded as ‘the missing Millennium Development Goal’ � Electricity in particular is seen as crucial to development, not least because electrical lighting, lengthens the productive time available in the day � Around 1.6 billion people do not have access to electricity � Can night-time lights help in monitoring access to electricity

Overview of the study � The properties of the DMSP-OLS sensor prompted the evaluation of rural electrification rates – Overglow effects in urban areas deemed it unsuitable for assessing the urban component of electricity access � Access to electricity was assessed by evaluating population present in unlit areas of the world and comparing it to the total rural population � Regional and national electrification rates were calculated

Population density in areas without electricity access (no light) � Areas with the lowest levels of electricity access are also some of the least densely populated From Doll & Pachauri, 2010

� Sub-Saharan Africa has 93% of rural areas without access to electricity � The population density in (these) unlit areas is 26 people/km 2 � Compared to India where the unlit population density is over 200people/km 2 � The geographical dimension of access to electricity will require differentiated solutions in order to be economically attractive

Obstacles to using night-time lights for estimating access to electricity � Where population density can be detected, usage is not dense enough to be detected � Population density is not high enough to be detected – What is the fundamental population density that DMSP- OLS can observe?

Population detection rates vary by region � This graph shows how the proportion of unlit pixels varies with population density � We see that in the developed world population is consistently detected from around 25persons/sqkm � But for developing countries, lower levels of access to electricity mean this only higher population densities are detectable � At 250 persons/km 2 : – Africa 80% undetected – Asia 50% undetected – Latin America 25% undetected Source: Doll & Pachauri, 2010 Energy Policy

The study.. � Using this method of evaluating % unlit cells by population density class, this paper investigates further the variation in detection rates of DMSP-OLS with regard to two spatially explicit population datasets – CIESIN’s GRUMP (census based population rendering) – ORNL’s Landscan (modelled ambient population dataset) � Detection profiles of population density with respect to DN are then evaluated by assessing the weighted average of population density per DN value

Regional groupings used

OECD regions are 90% detected between 50- 100persons/km 2 Fraction of unlit pixels Population Density (persons.km ‐ 2 ) Range 0 ‐ 500

Developing regions show more variation Fraction of unlit pixels Population Density (persons.km ‐ 2 ) Range 0 ‐ 2,000

Former Soviet Union, Centrally Planned Asia and the Middle East Fraction of unlit pixels Population Density (persons.km ‐ 2 ) Range 0 ‐ 2,000

Over the long range detection settles around 10-15% globally Fraction of unlit pixels Population Density (persons.km ‐ 2 ) Range 0 ‐ 15,000

Summary � Detection rates vary widely over levels of development � Interestingly, ambient population is consistently less detectable than residential in all regions � Globally around 90% of population is detected at 10,000persons/km 2 � What about the population density within lit areas?

GRUMP DN-Population density profiles DN Value Population Density (persons.km ‐ 2 ) Range 0 ‐ 8,000

Landscan DN-Population density profiles DN Value Population Density (persons.km ‐ 2 ) Range 0 ‐ 8,000

Summary table of typical population densities by region average DN 0 10 25 40 50 60 63 desnity NAM GRUMP 4 29 82 159 251 522 1620 53 WEU GRUMP 17 107 327 557 762 1449 3355 114 PAO GRUMP 4 184 467 702 961 1639 4879 126 MEA GRUMP 16 280 575 910 1027 1297 2771 54 LAM GRUMP 11 254 502 760 1022 1658 4115 41 CPA GRUMP 77 796 1422 1762 2200 3639 6800 155 FSU GRUMP 9 129 467 869 1444 1849 3222 31 EEU GRUMP 39 137 476 942 1160 2192 3876 109 AFR GRUMP 28 693 979 1405 1954 2547 3260 40 SAS GRUMP 153 665 1530 2113 2249 3954 9547 277 PAS GRUMP 49 527 914 1288 1794 3441 7439 126

Summary table of typical population densities by region Average desnity DN 0 10 25 40 50 60 63 (regional) NAM GRUMP 4 29 82 159 251 522 1620 53 WEU GRUMP 17 107 327 557 762 1449 3355 114 PAO GRUMP 4 184 467 702 961 1639 4879 126 MEA GRUMP 16 280 575 910 1027 1297 2771 54 LAM GRUMP 11 254 502 760 1022 1658 4115 41 CPA GRUMP 77 796 1422 1762 2200 3639 6800 155 FSU GRUMP 9 129 467 869 1444 1849 3222 31 EEU GRUMP 39 137 476 942 1160 2192 3876 109 AFR GRUMP 28 693 979 1405 1954 2547 3260 40 SAS GRUMP 153 665 1530 2113 2249 3954 9547 277 PAS GRUMP 49 527 914 1288 1794 3441 7439 126

Summary � South Asia has the most linear profile of DN to population density � Two regions of similar population density but different development levels can correspond to a factor of two population density difference for a given DN value � North America stands out even compared to other developed nations and the ratio is even higher between other regions

Conclusions � This study gives a broad regional overview of the relationship between population night-time lights with respect to two parameters: – likelihood of detection – DN-population density profiles � Such descriptions should help form assumptions when using DMSP-OLS data and what you can reasonably expect to get from analysis � Longitudinal component is missing but whilst a general pattern is evident, it is not clear that all regions will converge to one profile