Energy poverty is a term widely used to define living conditions under unaffordable and inaccessible energy resources. There are a number of factors that cause energy poverty in a broader sense. On the one hand, local factors such as natural resources, geographical location, local policies, household income or education level play an important role in individual energy accessibility and affordability [1
]. On the other hand, global factors such as macroeconomic, geopolitical and climatic factors also influence the level of energy poverty [3
]. The fight against energy poverty is of increasing relevance as the population on our planet is projected to increase by slightly more than one billion people over the next 13 years, reaching 8.6 billion in 2030 and to increase further to 9.8 billion in 2050 and 11.2 billion by 2100 [7
]. The main challenges our society will face will be to provide accessibility and affordability of energy to people. Since economic growth is closely related to energy access and affordability, it is a topic of highest importance since energy access is the foundation of wealth, freedom, health and human dignity [8
In this article, we summarize the current discussion on energy poverty, the definitions and measures, as well as the resulting problems in policy implementation and their implications up to now. One key element we stress is the variety of measurement methods of energy poverty, which makes it almost impossible to conduct comparative analyses among countries. Moreover, we highlight the issue that these indicators most often just capture the space dimension and neglect the time dimension. This issue is problematic since policy measures usually do not take into account the dynamics of changing factors that lead to energy poverty and therefore risk maintaining policies that are not efficient and effective over the course of time. As the political decision making process is a tedious one, the flexibility and speed of administrations to react to changes in underlying causal factors is questionable. Therefore, it is important to consider the dynamics of these changes already at the basis of decision making.
The literature on the topic of energy poverty is widely spread among different disciplines, ranging from financial economics, energy economics, biomedical science to humanities, socioeconomic and even philosophical areas [5
]. We contribute to the current discussion by considering the potential of big data and usage of AI for the enhancement of socioeconomic data through environmental, political and climatic data for policy decision making.
This paper is structured as follows. We first start by giving a definition of energy poverty, which is followed by summarizing the development of significant legislation in various countries. We then explain the vicious circle of low productivity, health risk, environmental pollution and energy poverty and present thereafter currently used energy poverty measures and alleviation policies and stress their problems in application due to the underlying dynamics. Out of that, we show how big data could improve energy poverty alleviation policies and discuss the challenges that the data collection process would bear.
2. Energy Poverty
Energy poverty has become the main focus in many fields, from social and economic studies to engineering studies and political campaigns. The number of publications dedicated to energy poverty shows the public interest in the issue in both academic and public communities. Figure 1
shows the number of research publications in “Elsevier Science Direct” and “Wiley Online Library” with the keywords “Energy Poverty” or “Fuel Poverty”. As can be seen, the number of publications increased exponentially in recent years.
The study of energy poverty has raised serious challenges in these fields. Even the definition of the subject has been an issue. For instance, in 2013, the UK government (David Cameron’s cabinet) changed the definition of fuel poor households from “households which spend more than 10% of their income on fuel to maintain an adequate level of warmth” to “households which spend above average fuel costs that would leave them with a residual income below the official poverty line” [13
]. Social activists, political parties, commons committee and media, accused the government of redefining fuel poverty to bring down the number of households classified as fuel poor [13
The definition of energy poverty will lead to designing energy poverty measures and, consequently, solutions and policies [15
]. For instance, defining energy poverty as “households which spend more than 10% of their income on fuel to maintain an adequate level of warmth” would consider “low income low cost”, “high income high cost” and “low income high cost” as energy poor (when the energy cost is considerably higher than the income) and takes into account the financial changes during the time (the definition takes in to account not only energy consumption and price but also a household’s income relative to energy cost). Hence, all policies should consider decreasing energy cost (through decreasing energy price or increasing energy efficiency) for all households in energy poverty (even households with incomes higher than the poverty line). Consequently, energy poverty should not be considered as a part of income poverty. Changing the fuel poverty definition to “households which spend above average fuel costs that would leave them with a residual income below the official poverty line”, energy poverty would only consider “low income high cost” as fuel poor. Moreover, this definition has two conditions for a household to fall into the fuel poor category. The first condition is to “spend above average fuel costs”, which means energy inefficiency and the second is “residual income below the official poverty line”, which means low income. With this definition, fuel poverty would become a subclass of income poverty. Consequently, the policies for fuel poverty reduction would focus only on financially poor households, reducing the poverty threshold and increasing energy efficiency.
There is no universal definition of energy or fuel poverty and in different countries and regions, there are different social, economic and environmental situations. However, in developed countries, energy (fuel) poverty is usually defined based on energy (or fuel) affordability, while in less developed countries, it is defined based on clean or modern energy accessibility [16
]. Different countries and institutions have given their own definitions based on the social, economic and environmental situations of their regions. The UK was the first country in the EU that defined and addressed energy poverty (as fuel poverty) officially Schumacher et al. [17
]. In the UK, the first definition of fuel poverty was given in “Warm Homes and Energy Conservation Act (WHECA) 2000”: “a person is to be regarded as living “in fuel poverty” if he or she is a member of a household living on a lower income in a home which cannot be kept warm at reasonable cost” [18
]. The same definition is used later in the 2010 and 2013 Energy Acts. The definition is imprecise and impractical. The practical meanings of “reasonable cost”, “warm” and “lower income” are unclear in this definition. Moreover, the definition is only based on the warmth of the house, which is the issue in cold seasons and does not take into account issues such as cooking proper meals and access to proper lighting and air conditioning during hot seasons. Recently, scholars and authorities have raised concerns about the fuel poverty definition see References [19
]. The European Commission (EC) has defined energy poverty as a set of conditions where “individuals or households are not able to adequately heat or provide other required energy services in their homes at affordable cost ”Thomson and Bouzarovski [22
]. This definition takes into account other energy needs (cooling, lightning and powering the appliances needed for a decent standard of living and health). With this definition, the policies for reducing energy poverty should guarantee a standard level of life with reasonable price. So, energy poverty policies should address low incomes, high energy prices and accessible appliance technologies at the same time. The African Energy Commission (AFREC) uses a definition similar to the UK’s 10% definition, that is, “a household that spends more than 10% of its revenues to meet its energy needs is in energy poverty”. However, the energy poverty indicators are not entirely based on revenue and energy costs since some areas are not well developed and do not have access to clean or modern energy. AFREC uses two additional indicators to measure energy poverty: rate of electrification and access to biomass for cooking [23
]. With this definition, energy poverty policies should take into account developing sustainable energy resources too.
2.1. Vicious Circle of Energy Poverty
Energy poverty is one of the central factors in socioeconomic and environmental issues. In some cases, it is the key link in socioeconomic crisis chain. Abundant research in the last two decades has revealed an interactional relation between energy poverty and a number of socioeconomic and environmental issues. For instance, there is a linkage between economic development and energy poverty [2
]. In the developed countries, social support programs would prevent energy poverty and its consequences [24
], although it would not solve the problem [26
]. On the other hand, in less developed regions, such programs are not well funded and, due to socioeconomic situations, housing poverty is usually greater. Other socioeconomic issues such as migration, education level and employment could make the energy poverty situation worse [27
In a region with energy poverty, there usually is low productivity (due to low access to energy or high energy cost), low surplus and, consequently, low income. Low productivity would result in low cash flow, which means there is not adequate financial resources to develop energy supplies, energy conversion or energy efficiency. This is called the vicious circle of energy poverty [29
]. At the household level, the vicious circle exists between energy poverty and housing poverty [30
Energy poverty affects an individual’s health and well-being. In developed countries such as the U.S., the UK and Australia, households with high levels of energy poverty usually are forced to reduce their food expenditures while cutting other basic needs as well as heating/cooling energy demand [24
]. Reducing expenditures for basic needs potentially would cause the physical/mental illness [31
]. In less developed countries (regions in which households have less access to modern energy), households with high levels of energy poverty usually use biomass and inefficient fuels (such as “firewood”, “dung cake”, “crop residue”, “kerosene” and “coal/charcoal” [32
]), which would cause indoor pollution and put households’ well-being at risk. In both situations (developed and less developed regions), as a consequence of energy poverty, a household’s income decreases (due to a decreased ability to offer workforce), while its health costs increases.
At larger scale, as a result of inefficient fuels, regional air pollution will increase. For instance, energy poverty (and using firewood as the main source of energy) is the main cause of air pollution in Southern Chile [33
]. Air pollution caused by energy poverty is not just an issue in less developed countries (with low access to modern energy). Between 2010 and 2012, as the result of economic crisis and increased tax rate on heating fuel, a large number of Greek households lost their electricity connections due to unpaid electricity bills [34
]. The households had become energy poor and started using firewood as an alternative heating fuel, resulting in a new environmental crisis. In 2013, a high level of winter air pollution was reported, in large Greek cities. Smog worsened through nights, which shows firewood’s role during dark and cold hours [35
]. Chemical analysis of the air pollutants showed high levels of “lead”, “arsenic” and “cadmium” particles, meaning people were burning painted wood (most likely old furniture) [36
]. Due to indoor and environmental pollution, energy poverty reduces productivity and increases health costs in a region. This is the vicious circle of energy poverty, environmental pollution and health risk (in which energy poverty would increase health risks and increased health risks would worsen the energy poverty situation). Energy poverty takes part in another vicious circle, as it has a synergic relation with local and global climate change [39
2.2. Energy Poverty Measures and Alleviation Policies
The next step in alleviating energy poverty is to develop a measure that shows all aspects of energy poverty. There are a variety of methods and indices for measuring energy poverty. Each study has developed a method based on socioeconomic and environmental situations of the region at hand. Table 1
and Table 2
show the variety of energy poverty measures and indicators proposed in recent years.
As seen in Table 1
and Table 2
, different energy poverty indices are proposed for different environmental situations. Since environmental situations are dynamic over time, the energy poverty index (and consequently the energy poverty alleviation policies) would change not only based on region but over time.
Different policies have been proposed to alleviate energy poverty subsequent to social awareness. One common policy (mostly executed in developed countries) is a series of social supports for energy poor households. Although this approach could prevent the issues caused by energy poverty, it would not solve the energy poverty itself. In addition, there is an ongoing debate about the type of social aid in both academic society (see References Karásek and Pojar [60
], Oppenheim [61
], Scarpellini et al. [62
]) and social/political campaigns (for instance, see References Thomson and Bouzarovski [22
], Enet English [37
], Daily Mail Press Association [63
], Long-Bailey [64
]). Three other popular strategies to alleviate energy poverty are to “increase energy efficiency of dwelling” [65
], “increase energy affordability and demand management” [61
] and “increase access to clean/modern energy” [69
]. The first two strategies are more common in developed regions (regions with high access to clean/modern energy), while the third one is the policy in less developed regions. A comprehensive study of energy poverty alleviation policies is provided by Reference Santamouris [72
]. Although there are a variety of policies, the choice of policies is usually local, based on the indicators’ situation in each region. As seen in Table 1
and Table 2
, many of the indicators and pressure factors mentioned above are global (such as climate, migration and economic development). The policies however, should be made at the local level [62
4. Use of Satellite Imaging Data to Predict Energy Poverty
While there are many ways to measure energy poverty, there is still lack of data on energy poverty metrics and supporting indicators in the developing world and in rural areas in the developed world. This lack of information is because the data are typically gathered through surveys, which are time-consuming, expensive and sometimes impossible to carry out in particular regions of countries and are thus less common. Given the difficulties of scaling up survey-based methods, the opportunities that satellite imaging data offer could be a key to assessing the global distribution of poverty.
The satellite-based Earth Observation market is in an emerging state today and is expected to boom over the next decades owing to recent technical breakthroughs regarding data collection and storage. Indeed, there is an increasing offer of high-quality Earth observation data, both in terms of variety of sources and of quality of resolution [75
]. The usefulness of satellite images can be examined along several dimensions of satellite design characteristics and acquisition conditions, such as spatial resolution, temporal resolution, spectral range, spectral resolution, cloud coverage, time of capture and off-nadir angle [76
The large quantity of data that can be captured using satellite technology means that a larger and more accurate picture can be quickly created. By identifying the areas that need better energy efficiency measures with satellite images, scientists will have a more accurate view of how to reduce unnecessary energy consumption and they will be able to precisely assess poverty in remote and rural areas. This technique could help revolutionize the ways in which groups find impoverished areas and eventually provide relief to people living in those specific parts of the world.
In September 2018, the European Space Agency (ESA) in partnership with the gas and electricity supplier E.ON. and Astrosat announced a trial project that uses satellite imaging data from across the UK to identify the communities that are in most need of support [77
]. By using orbiting satellites, the project will use near real-time and historical data including optical sources and thermal-infrared for heat mapping, air quality and pollution tracking. The data will be combined with Astrosat’s ThermCERT software to help address issues such as housing conditions, insulation, air quality and even traffic management. This project has the potential to be rolled out across other countries once the UK trial has successfully concluded [78
Remote sensing data such as high-resolution satellite imagery are therefore becoming increasingly available and inexpensive and can be used as aids in deriving essential conclusions regarding initiatives to tack energy poverty, as has been reported in recent literature. Measuring intra-urban poverty using urban texture and structure metrics and land cover was examined as a means of determining spatial unit analysis for energy poverty in the slum areas of Medellin, Columbia [79
]. The authors used very high spatial resolution (VHR) Quickbird satellite images and examined 139 analytical regions, which were broken down from 243 city neighborhoods. They found that these features explain up to 59% of the variability in a measure of poverty known as the Slum Index.
Understanding relationships between poverty and the environment is crucial for sustainable development and ecological conservation. Monitoring annual progress toward the United Nations’ sustainable development goals (SDGs) using household surveys is prohibitively expensive. Watmough et al. [80
] explored whether remotely sensed (RS) satellite data could be used to monitor rural poverty in low-income and middle-income countries. The authors analyzed RS land use and land cover data for a cluster of rural villages in Kenya at multiple spatial levels, from individual homesteads to the wider village periphery. Using this approach, the authors identified the poorest households with 62% accuracy, compared with 52% using a single-level approach in which RS data are aggregated over a fixed area. According to the authors, appropriate refinement of such a method, combined with the increasing availability of RS data and volunteered geographic information, could provide a more cost-effective means of monitoring sustainable development than household surveys.
Several studies have explored the correlations between night-time lights (NTL) satellite imagery, economic activity and welfare in a given area [81
]. For instance, dark areas of the NTL (Figure 2
) show the areas with low access to energy. These are either less populated areas or areas with high energy poverty. NTL appear less useful for distinguishing differences in economic activity between poor, densely populated areas and wealthy, sparsely populated areas [91
]. The latest studies looked at daylight images that can help distinguish different levels of economic well-being in developing countries [91
Even though there is access to satellite imagery for the entire globe, many of these raw data are not in a usable format for machine learning frameworks, making it difficult to extract actionable insights at scale [92
]. The earliest approach that combined machine learning with satellite imagery and crowdsourced assistance to identify poor villages in Kenya was presented by two NGO Give Directly volunteer researchers in 2014 [76
]. The authors estimated the ratio of thatched and metal roofs in each village and targeted the villages with large percentages of thatched roofs as poor villages. While the algorithm was not accurate enough, it showed the potential of using satellite imagery and machine learning for leveraging in development work. Next, Xie et al. [92
] trained a convolutional neural network (CNN) to predict nighttime light intensities from daytime satellite imagery and then used the learned filters to predict poverty areas by classifying each location as either “in-poverty” or “poverty-free”. In response to the binary classification problem there was the study of Jean et al. [91
], who provided more fine-grained poverty data in five countries of Africa by using continuous wealth measures. In particular, the authors established a transfer learning approach to estimate poverty for the countries of Malawi, Nigeria, Rwanda, Tanzania and Uganda by using a set of 4096 unstructured features extracted from the penultimate layer of a CNN that used Google Static Maps (API) to obtain daytime imagery and predict the nighttime light luminosity. Using ridge regression, these features were then used to predict the average per capita consumption, taken from living standard measurement surveys. The resulting model had good predictions and explained 40 to 50% of the variation in village per capita consumption. The analysis of Head et al. [93
] replicated and extended results from Jean et al. [91
]. The authors confirmed that transfer learning predicts asset index with R
of approximately 0.7 in Rwanda and Nigeria. However, when they performed the same technique in Haiti and Nepal, the predictive model lost some of its accuracy (R
of Haiti’s predictions for asset index falls to 0.51 and Nepal’s is 0.64), suggesting that NTL step in the transfer learning process may be not suitable for poor, low-density areas.
Engstrom et al. [94
] investigated the ability of object and texture features derived from high spatial resolution imagery (HSRI) to estimate and predict poverty rates in a sample of 1291 villages in Sri Lanka. The results indicated that the correlation between indicators from high-resolution daytime imagery and economic well-being was remarkably strong. Simple linear models explained nearly 60 percent of the variation in poverty and average log per capita consumption. In comparison, models built using NTL captured only 15 percent of the variation in poverty or income in the same area. Babenko et al. [97
] trained CNNs to estimate poverty directly from high and medium resolution satellite images provided by both Planet and Digital Globe (without using transfer learning or NTL), covering all 2 million km
of Mexico. The authors found that in urban areas, the predictive power of Digital Globe was slightly better than for Planet imagery, which explained 61% and 54% of poverty variation, respectively. They concluded that CNNs can be trained end-to-end on satellite imagery to estimate poverty, although there is much work to be done to understand how the training process influences out-of-sample validation.
Zhao et al. [98
] used a random forest regression (RFR) model by combining features extracted from multiple data sources, including National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (NPP-VIIRS) Day/Night Band (DNB) nighttime light (NTL) data, Google satellite imagery and land cover map, road map and division headquarter location data to estimate poverty based on household wealth index (WI) at a 10-km spatial resolution. The authors trained the RFR model using data in Bangladesh and applied the model to both Bangladesh and Nepal to evaluate the model’s accuracy. With a relatively high R
between estimated and actual WI, the authors concluded that the data that were estimated to measure poverty were relatively accurate and suggested that those countries with no accurate data or that lacked survey data implement the RFR model.
The very high cost of VHR imagery per image or collection of images motivated Perez et al. [99
] and Gram-Hansen et al. [95
] to carry out their research by using freely available multi-spectral imagery, which is typically of a much lower resolution. In particular, Perez et al. [99
] trained CNN models on free and publicly available multispectral daytime satellite images of the African continent from the Landsat 7 satellite, which has collected imagery with global coverage for almost two decades. Although the Landsat images are of lower resolution than the Google Static Maps images used by Jean et al. [91
], the authors achieved equivalent or better results than the previous benchmarks. Furthermore, Gram-Hansen et al. [95
] introduced a brand new machine learning dataset, purposely developed for informal settlement detection and they showed that it is possible to detect informal settlements using freely available low-resolution (LR) data, in contrast to previous studies that use very-high resolution (VHR) satellite and aerial imagery.
The consideration of three additional layers to socioeconomic data, namely, environmental, political and climatic data, for policy decision making allows policy makers to capture not only the spatial dimension of the development of energy poverty factors but also the dynamics over time. The collection and consolidation of these datasets would create a new data basis for policy makers and enable them to deduce direct implications of certain policy decisions and conduct scenario and sensitivity analyses. To be able to make use of this powerful big data set, the following obstacles would need to be overcome:
Definition of which data to gather: The first requirement would be to agree on a definition and delineation of the required data, determination of the data sources and the dedication of the data, for example, what they should be used for.
Data collection: The need of an agenda to ensure a proper collection of datasets. This includes the enforcement of data provision and in some cases the prior provision of measurable equipment to countries that lack sufficient infrastructure and equipment.
Standardization: Since the collection of data from various sources will inevitably include inhomogeneities, it is important to agree on measurable and comparable units.
Interaction of data: Once the data have been gathered and standardized, it is important to learn about the relationships among the datasets.
Definition of target function: After the relationships have been studied, direct conclusions can be drawn and certain scenarios simulated. These scenarios depend on a target function, which would need to be defined, as the desired outcome (minimization of energy poverty) can be reached via a set of solutions.
The challenges for local policy decisions will be smaller than for global policy decisions since the implementation of energy poverty measures will be easier on a local basis, while evaluation of certain energy poverty alleviation policies might be harder to measure on a global scale.
Through the development of the Internet of Things (IoT), data gathering and the implementation and tracking of policy measures will be enabled and allow for frequent evaluation and adaption of these policies. AI will play an important role when it comes to the optimization of energy distribution and utilization since the amount of data in decentralized networks includes so much information that those data can be better, faster and more efficiently processed by AI than by humans. Nevertheless, human oversight will remain crucial.
Considering the evolution of AI, the question arises regarding which fundamental rules would need to be set up in order to ensure a peaceful and welfare-improving set of solutions to reduce energy poverty while not at the expense of energy wealth. Nevertheless, several prerequisites must be met before the implementation of the IoT can evolve at such scale, such as the provision and building of critical infrastructure, education and increased awareness of the importance of energy poverty.