Residential electricity demand projections for Italy: A spatial downscaling approach

Highlights

• The 2050 residential electricity demand for Italy is downscaled at 1 km resolution.

• The projected residential electricity is increasing and becoming less concentrated.

• Residential area in 2050 in Italy is also projected and increasing.

• Electricity demand per residential area is bound to increase at the NUTS-3 level.

• Results are driven by population density, GDP per capita, and Cooling Degree Days.

Abstract

This work projects future residential electricity demand in Italy at the local (1 km grid) level based on population, land use, socio-economic and climate scenarios for the year 2050. A two-step approach is employed. In the first step, a grid-level model is estimated to explain land use as a function of socio-economic and demographic variables. In the second step, a provincial-level model explaining residential electricity intensity (gigawatt hours [GWh] per kilometre of residential land) as a function of socio-economic and climatic information is estimated. The estimates of the two models are then combined to project downscaled residential electricity consumption. The evidence suggests not only that the residential electricity demand will increase in the future but, most importantly, that its spatial distribution and dispersion will change in the next decades mostly due to changes in population density. Policy implications are discussed in relation to efficiency measures and the design of green energy supply from local production plants to facilitate matching demand with supply.

Introduction

The availability of high-resolution data on residential electricity demand might benefit energy transition. While many scenarios relevant for decision making, such as the Intergovernmental Panel on Climate Change (IPCC) Shared Socioeconomic Pathways (SSPs) (Riahi et al., 2017) for emissions, land use, and energy, are elaborated at the country level, finer depictions of the same are required for data-informed decision making at sub-national administrative levels. Sub-regional data are often either unavailable or available with considerable delay (Murakami and Yamagata, 2019; Höwer et al., 2019), and electricity demand data are seldom published at levels such as the municipal level (Murakami et al., 2015). Such practice is in contrast with the need for downscaled data at even higher resolutions. The availability of research-based evidence at a more refined geographic scale is key for facilitating detailed spatial studies. Indeed, information available on finer grids makes it possible to arrange the evidence on the basis of policy needs, which might differ among political and administrative units (Gaffin, 2004). For instance, Murakami and Yamagata (2019) discuss how important granular resolutions are for the analysis of climate change mitigation and adaptation processes, arguing that relevant features such as urban form typologies can be captured at this level only.

This work aims to contribute to this policy requirement by developing a spatially detailed dataset of residential electricity demand projections in Italy to 2050 at a very fine grid of approximately 1 km by 1 km (1 km × 1 km). The year 2050 is a key target in the 2018 long-term European Union (EU) strategy to be climate-neutral, an objective at the heart of the European Green Deal (European Commission, 2020). The analysis is focused on residential electricity demand, a relevant contributor to greenhouse gas (GHG) emissions that globally account for about 17% of carbon dioxide ($C{O}_2$) emissions (Nejat et al., 2015). Residential electricity demand is subject to potential shifts due to the increase in urbanisation processes (United Nations, 2018) that may, however, be decoupled from population growth (Guastella et al., 2019). The focus on a single country, Italy in our case, allows for the computational feasibility of our approach.¹ In addition, the heterogeneity of the climatic and socio-economic conditions of Italy makes the country a good case study to test this new methodological approach. Replication of the analysis in other countries is undoubtedly possible and warranted for future research, although with some caveats. For instance, extending the framework to a multi-country setting would require controlling for institutional and regulatory differences among countries, differences in energy prices and supply mix, and increased heterogeneity of the geographical conditions.

Our methodology relies on statistical spatial downscaling, a ‘top-down’ approach based on available aggregate information that has been of growing interest in several fields (Swan and Ugursal, 2009; van Ruijven et al., 2019). For instance, Murakami et al. (2015) use spatial downscaling to derive disaggregated electricity demand for Japan at 1 km × 1 km grid resolution in relation to sustainability and energy transition. Several other works have employed the downscaling methodology to retrieve local-level information about population, income and emissions. Gaffin et al. (2004) propose a linear downscaling of global population and GDP until 2100 under different Special Report on Emissions Scenarios (SRES) at approximately a 30 km grid resolution. van Vuuren et al. (2007) obtain global population, GDP and emissions for IPCC- SRES scenarios downscaled to the grid level of approximately 27.75 km. Nam and Reilly (2013) downscale the global population density from The National Aeronautics and Space Administration (NASA)'s Socioeconomic Data and Application Centre (SEDAC) for 1990–2015 to a 27.75 km resolution grid with a rank-size rule-based approach to estimate city size. Jones and O'Neil (2013) downscale the population in the US to a grid of approximately 11 km for selected SRES by employing a population potentials model. The same authors adopt a gravity approach to downscale the global population from all five SSPs (Jones and O'Neil, 2016). Finally, Murakami and Yamagata (2019) downscale the GDP and population scenarios developed by the SSPs to an approximately 50 km grid map by applying a four-step approach based on city population projections and urban and rural areas potential.

Few works at the global level seem to be dedicated to achieving resolutions lower than 10 km × 10 km, something that instead can be achieved in works at the national level. The main advantage in the latter case would be the availability of a higher number of relevant variables to be considered in the downscaling exercise and the computational feasibility of parametrised models applied to reduced geographical areas (Gao, 2017).

The downscaling approach proposed in this work follows a relatively simple two-step procedure grounded on the intensity method (Yamagata et al., 2015; Seya et al., 2016). This procedure uses essential information on electricity intensity, residential area and population projections. This feature ensures its replicability in most countries in Europe and other world regions and has the advantage of being computationally feasible. In this approach, the provincial (NUTS-3 in the European Nomenclature Unit for Territorial Statistics) residential electricity intensity (GWh per ha residential area) is regressed on population density, income and cooling degree days (CDDs). The grid-level total residential area is regressed on population, income and geographical factors, such as the distances of each grid from the most populated areas and the road density (Murakami and Yamagata, 2019). The estimated parameters from the grid-level model are used to project urbanisation in 2050 following the changes in a given socio-economic scenario. The projected urbanised area is then combined with population, income and CDD projections for the selected scenario, and the first model estimates are used to project electricity intensity. Finally, the grid-level demand is obtained as the product of NUTS-3 level projected intensity and grid-level projected urbanised area.

The choice to include both weather and income information in the electricity intensity equation is grounded on the existing literature about electricity and energy consumption determinants. Indeed, the effect of the interconnections between local weather, a changing climate, and residential electricity demand has been well addressed in the literature (Atalla and Hunt, 2016 and references therein).

The literature on electricity demand determinants is vast and covers multiple explanations in addition to the interconnections between socio-economic and climate changes. For instance, Ko (2013) reviews the effects of a city's physical shape on energy use considering housing size and type, density, community layout, planting, surface coverage, building design, heating efficiency, ventilation, air-conditioning systems and dwellers' behaviour. Yoshida et al. (2019) use location activity data to estimate quasi-real-time energy consumption in commercial buildings. van Ruijven et al. (2019) study the effects of the changing climate on amplifying energy demand. They develop long-term projections of electricity, oil and natural gas demands, considering climate shocks. Testing this model over an array of scenarios, they find that moderate and vigorous warming might increase the world energy demand by 11%–27% and 25%–58%, respectively. Damm et al. (2017) look at the effect of a 2 °C temperature increase on electricity demand for 26 European countries. Maintaining current socio-economic variables, they find a general reduction in consumption. Interestingly in terms of the scope of this work, they find that Italy is the only country where this does not apply, with an increase of 0.6%. Hostick et al. (2014) estimate US energy demand in 2050, finding a decrease in electricity intensity for the residential sector caused by the greening of buildings, more stringent building codes, improved appliance and equipment standards and the level of research on ultra-efficient buildings. Schweizer and Morgan (2016) also look at the US and perform a bounding analysis for electricity demand in 2050 based on changes in adaptation and mitigation policies and low or high GDP growth, finding modest or substantial improvements in energy intensities. The projected values for US electricity demand might then span 3100–17000 TW h (TWh). Huang and Hua (2019) study the balance between economic and green growth, focusing on the eco-efficiency of 191 Chinese cities from 2003 to 2013 using a spatial approach and data envelopment analysis, finding evidence of convergence for efficiency scores. Zaman et al. (2012) study the determinants of electricity consumption in Pakistan between 1975 and 2010 with an autoregressive distributed lag (ARDL) model, finding a positive and significant effect of GDP per capita, population growth and foreign direct investment.

Few works in the literature attempted to project energy and electricity consumption in Italy, and to our knowledge none proposed a spatial downscaling to the 1 km × 1 km level resolution. Bianco et al. (2009) relate national economic and demographic variables, such as population and GDP per capita, to find price and GDP consumption elasticities for the historical energy consumption in Italy, spanning the years 1970–2007. They then attempt to provide an accurate forecasting model up to 2017. In line with the Terna (2007) estimates, they forecast a yearly average of 2% growth. Employing a sample of households’ hourly electricity load for 2011, Alberini et al. (2019) study residential electricity demand to isolate the effect of temperature values. They find an irrelevant impact until 24.4 °C, after which temperature increases result in a sharp rise in consumption up to 9%–12%. For Italy, the effect of cooling is lower, given the predominance of gas heating. This feature seems to disprove the usual asymmetric V-shaped effect of temperature on electricity use, providing a linear increase instead. Besagni and Borgarello (2018) try to characterise the relationship between household composition and energy demand in Italy, segmenting the Italian population using the Italian Household Budget Survey of 2017. They find that socio-economic characteristics play a more important role compared to dwelling and appliance characteristics. Regarding electrical consumption, they find relevant differences related to geography and floor surface areas, pointing to the importance of evaluating the spatial dimension of electricity consumption models.

The remainder of the article is structured as follows. Section 2 describes the downscaling and projection approach in detail and presents the sources of employed data with summary statistics. Section 3 presents model estimates and projections and discusses the characteristics of projected electricity consumption data highlighting the changes in the spatial distributions of residential electricity demand in Italy and at the regional level. The implications for policy and decision makers, as well as future research directions, are summarised in Section 4.