Exploring factors affecting the relationship between light consumption and GDP based on DMSP/OLS nighttime satellite imagery

Highlights

• Time fixed panel regression is used to eliminate error caused by satellites.

• How GDP affects night lights depends on some other factors significantly.

• GDP is broken down to detect agricultural and non-agricultural contribution.

• An inverted “U” curve is found between night lights per capita and GDP per capita.

Abstract

We consider night light as a type of consumer goods and propose a model for factors affecting the relationship between night lights and GDP. It is then decomposed into agricultural and non-agricultural productions. Further, the model is modified to determine how the factors affect residents' propensity to consume lights. Models are tested with time-fixed regression on a set of 15-year panel data of 169 countries globally and regionally. We find that light consumption propensity is affected by GDP per capita, latitude, spatial distribution of human activities and gross saving rate, and that light consumption per capita has an inverted-U relationship with GDP per capita.

Introduction

Gross Domestic Product (GDP) is a crucial indicator in many societal studies and an important reference for political decision making. However, it is inadequately measured all over the world (Feige & Urban, 2008). In the least-developed countries, there are no usable national GDP data; in undemocratic regimes, statistical data are unreliable for probable unreal declaration by local governments; even in developed economies, measurement errors inevitably exist, with shadow economies ignored. Furthermore, GDP figures become more uncertain when converted into international dollars to be comparable with that in other countries. In response, many methods have been adopted in the past decades to estimate more accurate GDP, including the cash-demand approach (Tanzi, 1983), the consumer expenditure approach (Crohan & Smith, 1986) and the recently developed MIMIC approach (Schneider et al., 2010).

The method of remote sensing was not introduced to estimate GDP until Elvidge et al. (1997) found a strong positive relationship between GDP and night lights observed from outer space, which was detected by the US Air Force Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS). The data are collected every day between 9 p.m. and 10 p.m. local time (Elvidge et al., 1999). Since 1992, annual cloud-free composites with background noise and fires removed have been made by NOAA-NGDC and can be downloaded by the public for free; consequently, night lights data has been attracting growing research interests in various domains (Cinzano and Elvidge, 2004, Small et al., 2011, Sutton, 2003).

As many earlier papers have stated, the data set acquired by satellites has several advantages in estimating GDP, one of which is its objectivity clear of statistical errors (Ghosh et al., 2009), and another its spatially specific resolution without segmentation by administrative boundaries (Zhao et al., 2011). Its high-frequency accessibility also makes it a better measure compared to the statistical method carried out yearly. Ebener et al. (2005), Doll et al. (2006), Sutton et al. (2007), Elvidge et al., 2009a, Elvidge et al., 2009b and Ghosh et al. (2009) estimated GDP values or indicators of relative poverty at both national and sub-national scales based on night lights data. Earlier in 2002, Sutton and Costanza (2002) took nighttime imagery as a proxy of GDP to evaluate global ecosystem service.

However, relationships between GDP and night lights are not yet clear, and this uncertainty may lead to erroneous results when doing estimation. To solve the problem, Elvidge et al., 2009a, Elvidge et al., 2009b introduced the population factor acquired by LandScan (a series of spatially disaggregated global population count data sets by Oak Ridge National Laboratory) into the estimation; Ghosh et al. (2010) grouped all countries into 36 categories based on ratios of night lights and GDP and did regression within the groups to reduce the estimation error; Henderson et al. (2011) amended the estimation by combining statistical data together with night lights. These methods were effective but did not focus on and reveal the nature of why relationships between GDP and night lights vary from country to country. It is obvious that how much night lights are consumed by residents in a certain country is not merely determined by its GDP; rather, if we take night lights as normal goods (in economics, normal goods are any goods for which demand increases when income increases and falls when income decreases but price remains constant), the relationships are deeply affected by consumer preferences, just as other goods discussed in economics are. Albeit consumer preferences, specifically residents' light consumption propensity, seem too complicated to analyze in a microscopic view, at the national level, different natural and social factors such as latitude, per capita income and domestic saving rate may statistically link to different consumption tendencies. Therefore, what we discuss in this paper is how these factors work with or regulate the oversimplified relationships between GDP and night lights, based on which GDP estimation using night lights would be more reliable and reasonable.

Because DMSP/OLS has no on-board calibration, annual composites in different years or taken by different satellites could not be compared directly with each other (Table 1). Elvidge et al., 2009a, Elvidge et al., 2009b developed the second-order regression model to intercalibrate individual composites via an empirical procedure, in which Sicily was chosen as the reference area, with F121999 used as the reference composite. Data from other satellite years were adjusted to match the F121999 data range, assuming that night lights in the reference area have been largely stable over time. Although the calibration is valid, having successfully obtained a convergence of values in years where two satellite products are available, there are still two debatable issues:

• The assumption that lighting for the reference area, Sicily, has been largely stable over time is not convincing as the author has discussed in his paper.

• Any unnecessary calibration to the original images would bring new errors. Hence, if having an alternative to control systematic errors brought by satellites in some circumstances, we should turn to that. As to topics of GDP or other macro variable regression, time-fixed panel data models commonly applied in econometrics would be a better choice, or at least a choice.

Despite the fact that GDP and light consumption are strongly correlated, contributions of agricultural and non-agricultural productions to light consumption have not been studied quantitatively. In some studies, night lights were viewed only as urban things, as the authors believe that activities in rural areas without light produced cannot be captured by the satellites (Ghosh et al., 2010, Han et al., 2012). In other studies, night lights were implicitly considered as a reflection of all sectors of an economy, and the data set can be used to estimate total GDP (Doll et al., 2000, Doll et al., 2006). Which is closer to the real world? In our view, not only do industrial and servicing productions have immediate relationships with night lights, but also agricultural production is linked to night lights indirectly. What night lights represent is consumption rather than production. (Although this argument might be a little coarse for DMSP/OLS data sets, as gas flares, lights reflected by snow, and other light sources are all mixed with the true light consumption in the data sets, we believe it is generally reasonable.) That means both agricultural and non-agricultural productions reflect themselves in night lights indirectly to some extent. What proportions they separately contribute is one of the topics in this paper.

The study presented here contains three major parts: part A, factors affecting the relationship between light consumption and GDP; part B, contributions of agriculture and non-agriculture to global light consumption; part C, factors affecting light consumption per capita. Correspondingly, we first raise a model for light consumption and GDP with three other explanatory variables in consideration. Secondly, GDP is broken down into agricultural production and non-agricultural production to detect what proportions they separately contribute to light consumption. Thirdly, we develop a quadratic model for light consumption per capita to discover how factors, especially GDP per capita, affect residents' propensity to consume lights. After that, a set of 15-year panel data from 1995 to 2009 of 169 countries all over the world is used both globally and regionally in time-fixed regressions to test the models above. At the end, we offer some detailed discussion and some useful conclusions.