Detecting spatiotemporal dynamics of global electric power consumption using DMSP-OLS nighttime stable light data

Highlights

• The global NSL data were intercalibrated using the MIR method.

• Global EPC at 1 km resolution were modeled from 1992 to 2013.

• The spatiotemporal dynamics of global EPC were analyzed at multiple scales.

Abstract

The rapid development of global industrialization and urbanization has resulted in a great deal of electric power consumption (EPC), which is closely related to economic growth, carbon emissions, and the long-term stability of global climate. This study attempts to detect spatiotemporal dynamics of global EPC using the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP-OLS) nighttime stable light (NSL) data. The global NSL data from 1992 to 2013 were intercalibrated via a modified invariant region (MIR) method. The global EPC at 1 km resolution was then modeled using the intercalibrated NSL data to assess spatiotemporal dynamics of EPC from a global scale down to continental and national scales. The results showed that the MIR method not only reduced the saturated lighted pixels, but also improved the continuity and comparability of the NSL data. An accuracy assessment was undertaken and confined that the intercalibrated NSL data were relatively suitable and accurate for estimating EPC in the world. Spatiotemporal variations of EPC were mainly identified in Europe, North America, and Asia. Special attention should be paid to China where the high grade and high-growth type of EPC covered 0.409% and 1.041% of the total country area during the study period, respectively. The results of this study greatly enhance the understanding of spatiotemporal dynamics of global EPC at the multiple scales. They will provide a scientific evidence base for tracking spatiotemporal dynamics of global EPC.

Introduction

Electric power consumption (EPC) is indispensable to modern society, playing an important role in supporting socioeconomic development and human life [1], [2], [3], [4]. As such, EPC is often one of the main sources of carbon emissions which are the important sources to drive and accelerate global warming [5], [6], [7], [8]. After World War II, the world has entered a rapid development stage of industrialization and urbanization. Global EPC showed a massive increase of about four fold from 4512 billion kW h in 1971 to 21,725 billion kW h in 2012 [9]. This rapid growth is not only closely related to the world energy market and global sustainable development, but also affects the long-term stability of global climate. Hence, accurately and reliably detecting spatiotemporal dynamics of global EPC is crucial for understanding both the impacts and the mechanisms of EPC and its interactions with socioeconomic activities and the environment.

Previous studies have detected spatiotemporal dynamics of EPC in several ways. For example, Ranjan and Jain [10] modeled spatiotemporal dynamics of EPC in Delhi using linear multiple regression models. Tso and Yau [11] used regression analysis, decision tree and neural networks for the evaluation of EPC. Huang et al. [12] used a Grey-Markov forecasting model to estimate the electric power supply and demand in China from 1985 to 2001. Wang et al. [13] analyzed the changes in industrial EPC in China from 1998 to 2007, using a logarithmic mean Divisia index I decomposition method. Bianco et al. [14] presented temporal EPC in Italy from 1970 to 2007. Bildirici et al. [15] evaluated the causality relationship between EPC and economic growth for US, China, Canada and Brazil. However, most of these investigations used statistical EPC data based on administrative units. In spite of their authoritativeness, the statistical EPC data were both time lagging and short of spatial details, which severely limited their usefulness [5]. More importantly, compared with statistical EPC data, detection of spatiotemporal dynamics of EPC at finer scales is a more realistic application. The results are easier to be integrated with other spatial data layers, so as to carry out interdisciplinary studies [5]. Consequently, more efficient ways of detecting spatiotemporal dynamics of EPC at finer scales are urgently demanded.

Satellite remote sensing imagery, such as the nighttime light data obtained by the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP-OLS), is an effective proxy for socioeconomic indicators [16], [17], [18], [19], [20], [21], [22], [23], [24], [25]. Since nighttime lights can directly reflect economic activity intensity which is closely related to EPC [26], [27], [28], the DMSP-OLS data can be integrated with statistical EPC data to capture spatiotemporal dynamics of EPC in detail over larger areas. Previous studies have demonstrated that DMSP-OLS data have a great potential to model EPC. For example, Elvidge et al. [27] firstly found that DMSP-OLS data were highly correlated to statistical EPC data in 21 countries in the mid-1990s. Amaral et al. [29] demonstrated a close correlation between DMSP-OLS data and statistical EPC data in Brazilian Amazonia in 1999. Similarity, Lo [30] gave a logarithmic relationship between DMSP-OLS data and statistical EPC data for 35 Chinese capital cities. Townsend et al. [31] reported a high correlation between DMSP-OLS data and statistical EPC data in Australia from 1997 to 2002. In addition, Chand et al. [32] analyzed spatial characterization of EPC patterns over India using temporal DMSP-OLS data from 1993 to 2002. He et al. [1] built logarithmic regression models between DMSP-OLS data and statistical EPC data to model spatiotemporal dynamics of EPC at county scale in China from 1995 to 2008. Cao et al. [5] proposed a top-down method to estimate pixel-based EPC in China from 1994 to 2009 using DMSP-OLS data, gross domestic product (GDP), and population as the variables. He et al. [33] detected spatiotemporal dynamics of EPC at the sub-county level in China from 2000 to 2008 using DMSP-OLS data. Using DMSP-OLS data, Xie and Weng [34] modeled spatiotemporal dynamics of EPC in urban cores and suburban regions at Chinese cities from 2000 to 2012. Although previous studies have documented the effectiveness of DMSP-OLS data for estimating EPC with varying degree of success, most of these studies focused on national, regional, or city level evaluation. Due to lack of complete statistical EPC data for all countries in the world, what is missing to date is the detailed exploration of DMSP-OLS data for detecting spatiotemporal dynamics of global EPC. It is, therefore, worthwhile to introduce an effective approach for linking incomplete statistical EPC data with DMSP-OLS data so that the spatiotemporal dynamics of global EPC can be modeled accurately and reliably.

There are two other constraints which limit the reliability and accuracy of modeling spatiotemporal dynamics of global EPC using DMSP-OLS data. The first one is pixel saturation of DMSP-OLS data in the urban center of large cities, and the other is the lack of continuity and comparability of the data [35]. Pixel saturation in DMSP-OLS data results from the OLS sensor’s low radiometric resolution of six bits. The OLS sensor was designed to detect low-radiance light sources with radiance ranging from 10⁻¹⁰ to 10⁻⁸ W/cm²/sr/μm under normal operation [36]. Light sources with radiance ranging >10⁻⁸ W/cm²/sr/μm, which often exist in the urban center, were all given the digital number (DN) value of 63 in DMSP-OLS data [33]. Consequently, the different radiance of various lights in the center of urban could not be distinguished. In addition, due to lack of in-flight calibration, discrepancies appeared between DN values derived from different satellites for the same year, and abnormal fluctuations appeared in DN values for different years derived from the same satellite [37], [38].

To improve the availability of DMSP-OLS data, many studies have attempted to resolve these two constraints. Among these studies, the invariant region method were widely used to intercalibrate the DMSP-OLS data [39]. The essence of this method is to use a small administrative unit as an invariant region to intercalibrate DMSP-OLS data in the broader regions. For example, Elvidge et al. [40] assumed Sicily, Italy as an invariant region to intercalibrate DMSP-OLS data from 1994 to 2008. Similarly, using Jixi as an invariant region, Liu et al. [37] applied second-order regression models to intercalibrate DMSP-OLS data in China from 1992 to 2008. Although these studies solved the discontinuity problem in DMSP-OLS data, they failed to reduce pixel saturation. In addition, Letu et al. [41] presented an invariant region method within some administrative units to remove the saturated pixels of DMSP-OLS data of Japan. Although this study had advantages of correcting the saturated pixels, it could hardly to optimize continuity and comparability of DN values within DMSP-OLS data. Most recently, Wu et al. [42] has defined Mauritius, Puerto Rico, and Okinawa as the invariant regions to some extent reduce pixel saturation and enhance continuity of DMSP-OLS data in the world. However, all the saturation pixels were assigned to a fix value which means that there is no spatial difference across the corrected saturation pixels for each year. In regard to the invariant region method, research questions remain regarding how to not only effectively reduce saturation pixels but also accurately solve the discontinuity problem in DMSP-OLS data.

To address the above deficiencies, this study attempts to detect spatiotemporal dynamics of global EPC using DMSP-OLS data from 1992 to 2013. The objectives are (1) using a modified invariant region (MIR) method for intercalibrating DMSP-OLS data; (2) linking incomplete statistical EPC data with the intercalibrated DMSP-OLS data for constructing global EPC at 1 km resolution; and (3) evaluating spatiotemporal dynamics of EPC from a global scale down to continental and national scales.

This study is organized as follows. Section 2 briefly describes data sources. Section 3 introduces methodology, presenting the methods for intercalibrating DMSP-OLS data, estimating EPC, and evaluating spatiotemporal dynamics of EPC. Section 4 presents the results and discussion, and conclusions are drawn in Section 5.