Potential impact of global warming on electricity demand in Niger

This study examines the potential impacts of climate change on electricity demand in Niger. Fourteen (14) regional climate simulations from the Coordinated Regional Climate Downscaling Experiment (CORDEX) were analyzed for the study. The study evaluates the ability of the simulations to reproduce the present-day climate variables over Niger, builds a climate-electricity demand model to link the electricity demand with climate variables, and quantifies the potential impact of climate change on daily electricity demand at various global warming levels (GWLs: 1.5 °C, 2.0 °C, 2.5 °C, and 3.0 °C) above the pre-industrial level. The climate-electricity demand model was built by combining the principal component analysis and the multiple linear regression analysis (hereafter, MLR). The residual analysis indicates that the MLR model complies with the assumptions of the regression analysis. The coefficient of determination (R2) of the MLR prediction is about 0.81, and the root mean square error (RMSE) is about 149.9 MWh day−1. The ensemble mean of the model simulations projects a future increase in electricity demand at all the GWLs, and more than 75% of the simulations agree on the projection. The study demonstrates how climate services could be used in quantifying the impacts of climate change on electricity demand, and the results would be valuable for reducing future climate risks in the energy sector.


Introduction
In most African countries, the energy sector faces several challenges, including insecure energy supply, continuous growth of electricity demand, and recurrent blackouts. For example, in Niger, electricity consumption has increased by more than 150% since 2001, with the largest increase in the residential and commercial sectors. As the cities are becoming more populated and the extreme hot days more frequent, electricity demand may continue to rise. Yet, the electricity supply systems have not been able to adequately keep up with the peak demand during the hot periods when the demand exceeds the available generating capacity, resulting in blackouts in several localities (NIGELEC 2019). For example, during the hot periods in 2016, the inability of a power company in Niger (called NIGELEC) to meet half of its electricity demand resulted in blackouts in many areas (NIGELEC 2019). This problem may be compounded in the future because the government of Niger plans to achieve 100% access to electricity in urban areas by 2030. In addition to the rise in electricity demand, the mean and extreme temperatures are also projected to increase over West Africa due to climate change (Climate Change 2014; Klutse et al. 2018;Nikulin et al. 2018). Meanwhile, previous studies have shown a good relationship between electricity demand and climate variables and demonstrated that changing climate might substantially alter energy demand (Akil et al. 2014;Cellura et al. 2018;Kaufmann et al. 2013;Jovanović et al. 2015;Valor et al. 2001;Guan et al. 2017;Shen 2017;Wang et al. 2017;Pardo et al. 2002). Hence, for better management of future electricity supply in any country, there is a need to quantify the potential impacts of climate change on electricity demand in the country.
Many studies have addressed the impact of climate change on electricity demand in several countries across the world and show that the impacts differ from one climate zone to another and from one city to the other. For instance, in cold climates, it is projected that climate change would decrease energy demand since less energy would be needed for heating the buildings during the winter. In Finland, Jylhä et al. (2015) reported a decrease of 20-35% in total energy consumption by 2100, depending on the magnitude of climate change. Wan et al. (2012) showed that the reduction in heating demand may be up to 22.3% in Harbin,23.6% in Hong Kong,26.6% in Beijing,and 55.7% in Shanghai. On the other hand, the energy demand is projected to increase in tropical countries, where more electricity will be required for cooling the buildings. Shourav et al. (2018) revealed that climate change would increase the daily electricity and peak demand in Dhaka City by up to 5.9-15.6% and 5.1-16.7% (respectively) toward the end of this century under different climate change scenarios. Ahmed et al. (2012) reported that an increase in temperature alone may lead to 1.36, 2.72, and 6.34% rise in per capita demand during the summer season and 2.09, 4.5, and 11.3% rise in per capita demand during the spring of 2030, 2050, and 2100, respectively. In Brazil, Invidiata et al. (2016) reported that climate change would induce an increase in the annual energy demand from 19 to 65% in 2020, 56 to 112% in 2050, and 112 to 185% in 2080. While it could be speculated that the electricity demand in Niger would also increase because of climate change, no study has quantified the percentage increase in the country. Given that the impacts of climate change may differ from one geographic location to another, there is a need to provide information on the potential impacts of climate change at specific locations, especially where the impacts could be felt. Such information would help the policymakers in formulating and implementing appropriate adaptation strategies.
Global climate models are a viable tool for projecting the impacts of climate change on various climate variables. However, the spatial resolution of these GCMs (typically 100-300 km) are too coarse for providing climate projections for impact assessments at local scales. Therefore, there is a need to downscale GCM outputs to local scale for impact assessments using downscaling techniques like dynamical downscaling with regional climate models. To assist developing countries that lack resources for the dynamic downscaling, the Coordinated Regional Climate Downscaling Experiment (CORDEX) has made their datasets available publicly. Several studies have already used the CORDEX data to assess the impacts of global warming on various sectors across the continent (Abiodun et al. 2017;Kumi and Abiodun 2018;Klutse et al. 2018Maúre et al. 2018;Abiodun et al. 2019). For instance, Kumi and Abiodun (2018) used the CORDEX dataset to project the impacts of climate change on rainfall onset, cessation, and length of the rainy season over West Africa, while Klutse et al. (2018) analyzed it for consecutive dry and wet days. In addition, Sawadogo et al. (2019) used the dataset to project the impacts of climate change on wind energy potential, and Sawadogo et al. (2020) also used it to quantify the impacts on photovoltaic solar generation. Nevertheless, to our knowledge, no study has utilized the dataset to investigate potential impacts on energy demand in West Africa. The present study intends to fill in the gap over Niger.
Hence, the present study aims to investigate the potential impacts of climate change on electricity demand in Niger at specific global warming levels using the multi-RCMs. In the study, we develop a multiple linear regression model (MLR) based on the historical relationship between electricity demand and climate variables, analyze the multi-simulation datasets CORDEX RCMs, and project the impacts of climate change on electricity demand based on the MLR model. Section 2 of the paper describes the data and methods used in the study; Section 3 presents and discusses the results; and Section 4 concludes the paper.

Study area
Our study area is Niger (12°N-24°N; 0°E-16°E), a West African country with a land area of 1,267,000 km 2 (Fig. 1). Niger has four distinct climate zones: the Soudano-Sahelian zone (about 1% of the total) with an annual rainfall ranging from 600 to 800 mm; the Sahelian zone (about 10% of the total land area) with an annual precipitation ranging from 350 to 600 mm; the Sahelo-Saharan zone (12% of the total land area) with an annual rainfall ranging from 150 to 350 mm; and the Saharan zone, which occupies about 77% of the total land area, with an annual precipitation less than 150 mm. The study focuses on Niamey, which is the capital and the largest city of Niger with a total land area of 256 km 2 and a population of about 1.5 million inhabitants. Niamey belongs to the Sahelian zone, with average temperatures ranging in the summer (March-June) from 30 to 35 and from 20 to 27 in the winter (December-February). It is the most important city in Niger in terms of infrastructure, institutions, and industries. This makes this city more attractive to rural dwellers. Indeed, in 2015, the electricity consumption of Niamey was about 63% of the total electricity consumed by the whole country (Fig. 2). The household sector is the main end user of energy consumption in Niger and represents 90% of the total energy consumption, followed by transport with 8% and industry, which accounts for 2% (Adamou et al. 2021).

Datasets
Two types of datasets were analyzed for the study, namely electricity demand and climate datasets.

Electricity demand datasets
The electricity demand datasets consist of observed and simulated datasets. The observed electricity dataset is the daily electricity demand (ED) data for Niamey that spans from January 2005 to December 2017. The ED dataset was obtained from the National Company of Electricity of Niger (NIGELEC), which is the only company responsible for generating electricity for the country. Figure 3 shows the time series of the observed ED data used in the study. The observed outliers, which mostly occurred during the rainy season, are due to network failure and strong winds. The extremely low values of the ED observed on August 28, 2005 ( Fig. 3; O1) and July 1, 2008 ( Fig. 3; O2) were due to the collapse of pylons on Birnin Kebbi's line, which provides about 68% of the country's electricity supply, resulting in blackouts in several localities. On the other hand, the minimum values observed in November 2010 ( Fig. 3; O3) resulted from the revisions of the thermal plant PC4 in Niamey to increase its production capacity. The ED data was used in developing a multiple linear regression (MLR) model. Detailed information on the model's development and application is presented in Section 2.3.

Climate datasets
The climate datasets used include the station observation, reanalysis, and climate simulation datasets.

Station observation data
The station data are from the automatic weather station in AGRHYMET and the Meteorological Service of Niger (DMN), extending from January 2005 to December 2017. These data include the mean temperature, the maximum temperature, the minimum temperature, the relative humidity, the wind speed, and the solar radiation. The descriptive statistics of climate variables are summarized in Table 1. The observed meteorological data was used in developing the multiple linear regression (MLR) model (see Section 2.3).

Reanalysis dataset
The reanalysis dataset is the Princeton Global Forecasting (PGF) dataset, an observational-reanalysis hybrid gridded dataset that provides near surface meteorological data for driving land surface models and other terrestrial modeling systems (Sheffield et al. 2006). PGF data consists of global 50 years , 3 hourly data at 1° × 1° resolution. It is constructed from a reanalysis of the combination of a set of global observation datasets with the National Center for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) (Sheffield et al. 2006). The biases in the precipitation and near surface meteorology reanalysis have been corrected using observation-based datasets of precipitation, air temperature, and radiation (Sheffield et al. 2006). The dataset was used to obtain the spatial distribution of climate variables over Niger.

Climate simulation dataset
The climate simulation dataset is from the Coordinated Regional Climate Downscaling Experiment (CORDEX; Nikulin et al. 2018), in which global climate model (GCM) simulations of past and future climates were downscaled with regional climate models (RCMs) over various regions of the world. The CORDEX data analyzed in the study include the daily temperature, humidity, radiation, and wind from 14 multi-model simulations produced by four RCMs (RCMs: RCA, CCLM, REMO, and ALADIN). The RCMs and GCMs downscaled are given in Table 2. The future climate projection data were analyzed at various global warming levels (GWLs; Table 2). More information on CORDEX and the definition of GWL periods can be obtained in Nikulin et al. (2018).

Obtaining the climate-induced electricity demand
The daily ED time series (Fig. 3) features a seasonal variation and an increasing trend. While the former is due to the seasonal variation in atmospheric conditions, the latter might be due to socio-economic development. Hence, in order to obtain the influence of climate variables on electricity demand, the trend (due to socio-economic development) and the seasonality (due to weekends and holidays) were removed following the procedure of Apadula et al. (2012). The general linear regression equation for the time series can be expressed as: where ED t is the aggregated daily electricity demand, t is the time variable (t = 0,1,2,3…), I W is the dummy variable taking the value 1 if the observation of the demand corresponds to holidays (weekends included) and 0 otherwise, i the regression constants, and y t is the climate-induced electricity demand (CED). Hence, we obtained the time series of CED by removing the deterministic parts (time variable due to socio-economic development and dummy variable due to holidays effect) from the time series of ED shown in Fig. 3. The results (CED) are shown in Fig. 4.

Calculation of cooling degree days
Several studies have shown that the relationship between electricity and temperature is nonlinear and thereby have used two branches in studying the relationship (Valor et al. 2001;Ahmed et al. 2012;Yi-Ling et al. 2014;Moral-Carcedo and Vicens-Otero 2005;Shin and (1) Do 2016). For convenience, the temperature can be introduced with the concept of "degree days": the cooling degree days (CDD) and the heating degree days (HDD). While HDD provides an indication of the sensible heating requirements for a particular location, CDD provides the same but for sensible cooling requirements (Giannakopoulos and Psiloglou 2006). The difference between the two branches is usually identified on electricity-temperature scatter plots using the base temperature, which is the point where the electricity shows no sensitivity to temperature. However, unlike the studies of Giannakopoulos and Psiloglou (2006) and Valor et al. (2001), in this study, the relationship between the electricity demand and temperature is quite linear and presents its minimum value around 22 °C (Fig. 5). This implies that HDD has no significant influence on the electricity demand in Niger. Hence, the CDD calculated using Eq.
(2) will be used in the present study.
where T is the mean temperature (°C) and 22 °C is the temperature at which the electricity shows no sensitivity to air temperature.

Calculation of heat index
Another factor that can influence electricity demand is the heat index (HI), which reflects an increased operation of air conditioning during hot and humid summer days (Apadula et al. 2012). The effect of relative humidity on electricity demand is supposed to be relevant only in warm and hot temperatures, because the perceived temperature can be higher in such meteorological conditions, and thus, the use of cooling appliances increases (Apadula et al. 2012). Following Steadman (1979), the HI formula can be defined as follows (Eq. 3): where HI is the heat index (°C), T is the air temperature (°C); H is the relative humidity (%), and C i are constants (C0 = − 42.39; C1 = 2.05; C2 = 10.14; C3 = − 0.22; C4 = − 0.6.84e-03; C5 = − 0.05; C6 = 1.23e-03, C7 = 8.53e-04; C8 = − 1.99e-06). The HI is applied only when the temperature is equal or exceeds 27 °C, and simultaneously, the relative humidity is higher than 40%. Such meteorological conditions occur only in Niger during the summer months (June-September) when the mean temperature is greater than 27 °C and the relative humidity relatively higher than 40%. If the above conditions are not satisfied, the HI is set equal to the maximum temperature.

Multiple linear regression model development
We developed a multiple linear regression (MLR) model to generate CED from climate variables. By definition, a MLR more for CED can be written as: where i are the regression constants and x i the climate variables.
Prior to the model development, the principal components analysis (PCA) was used to identify climate variables that are highly correlated with the climate-induced electricity demand. Indeed, the PCA is used in this study to group the key climate variables that are highly correlated with the electricity demand. Table 3 provides the loading of the PCA using two principal components (the PC1 and PC2). From this analysis, it is noticeable that the PC1 is highly correlated with the CED, T mean , T min , T max , radiation, and HI. In other words, there is a process that couples an increase of the electricity demand with these climate variables. Therefore, the PCA result suggests that the CDD, T max , T min , and HI are highly correlated with the CED.
Then, the stepwise MLR model was developed based on the PCA results. CED is used as a response variable, and the climate variables are used as independent variables. Following the approach of Braun et al. (2014), we split the datasets into two sets: the first part (80% of the data) was used to train the model, while the second part (20% of the data) was used to validate the model. The performance of the model is then assessed through its ability to estimate historical values of observed electricity demand. However, the step-wise linear regression model selected CDD, Tmax, HI, RH, Radiation, Wind based on their p-values. The regression coefficients and their corresponding p-values are given in Table 4, while the resulting regression model is in Eq. (5). Hence, Eq. (4) becomes: where 0 = −599.2 ; 1 = 77.27 ; 2 = 3.09 ; 3 = −12.07 ; 4 = 13.14 ; 4 = 13.14 5 = −0.33 ; 6 = −37.38 Moreover, we checked the basic linear models' assumptions using the residual plots (homogeneity of the variance and histogram of residuals) to see how well the model complies with the basic assumptions of linear models. These methods have already been applied in previous studies to check the assumptions of linear models (Aranda et Bianco et al. 2009). Then, the coefficient of determination R 2 , the root mean square error (RMSE), and the mean bias error are used to assess the accuracy of the model.

Simulation datasets evaluation
To evaluate the performance of the simulation datasets in reproducing the climate of Niger, we compared the simulated climate data for the period 1971-2000 (hereafter, the reference period) with the PGF data for the same period. However, the evaluation focuses on the variables needed for building the model. Furthermore, to assess the impact of climate change at various global warming levels (GWL1.5, GWL2.0, GWL2.5, and GWL3.0), we subtract the climate data in the reference period from that in GWL periods. The GWL period is defined as a 30-year period in which the climatology of global mean temperature is higher than that of the pre-industrial baseline period (1861-1890) ). As observed in Table 2, this period varies with GCM simulations. low mean square error (RMSE = 140.9 MWh/day). This indicates that the model performs well in estimating the CED based on the meteorological variables alone, without including the impact non-meteorological variables. Table 4 indicates that all the model parameters are significant.

Evaluation of the multiple linear regression model
Furthermore, we checked the assumptions of linearity to find out whether the model complies with the basic assumptions of the linear regression model. The residual error from the regression model is the difference between the observed and fitted climate-electricity demand. The residuals error should be normally distributed to comply with the basic assumptions of regression models (Bianco et al. 2009;Aranda et al. 2012). The residuals plot shows that there is no specific pattern or relationship between the residuals and the fitted values (Fig. 7a) and the distribution follows approximately the normal distribution (Fig. 7b). Consequently, we can conclude that the models comply with the basic assumptions of regression models. Therefore, the model can be used to project the impact of climate change on CED. Figure 8 shows that the RCMs reproduce well the annual cycle of daily energy demand and the climate variables (CDD, T max , heat index, radiation, and wind). In most cases, the observed annual cycles lie within the RCMs' ensemble spread except for CDD, for which the RCMs fail in reproducing the peak value observed in May and October, and also for wind, where the RCMs fail in reproducing the minimum values of wind speed observed in April. Furthermore, both observed and simulated cycles show high values of CED, CDD, HI, and T max in April-June and October, and high values of wind in June-July, reflecting the seasonal movement of the inter-tropical discontinuity (ITD) and the prevailing harmattan conditions. In both the observed and simulated curves, the minimum values of solar radiation occur in August, while the maximum values occur in March-April.

Evaluation of CORDEX simulations
Despite the good performance of the RCMs ensemble to reproducing the annual cycle of the climate variables (Fig. 8), the models struggle to reproduce the spatial distributions of some climate variables (Fig. 9). For instance, the observation features a maximum CDD (> 8 °C) over the southwestern part of the country. The RCMs ensemble mean fails to adequately reproduce this pattern (r ≈ 0.53); instead, it shows a relatively uniform distribution of the number of CDD across the country (2-6 °C). Thus, the bias in simulating the CDD is up to − 3 °C. Similarly, the same could be observed for the T max and HI where the observation features a maximum value of T max (≈ 40 °C) and HI (≈ 28 °C) over the southwestern part of the country. But, the RCMs have not been able to adequately reproduce these patterns (r = 0.06 for HI and r = 0.83 for T max ). The bias in simulating the T max is up to -5 °C over the central part of the country, while it is for HI to be + 1 °C over the northern and -1 °C over the southern part. Moreover, the RCMs' ensemble also struggles to reproduce the spatial variability of the observed radiation and wind (r ≈ 0.35). While the observation features a maximum radiation over the northern part of up to 300 W/m 2 , the RCMs' ensemble mean shows a maximum value of about 280 W/m 2 over a narrower area. The associated bias is up to − 50 W/ m 2 , suggesting that the models highly underestimate the solar radiation. Contrary, the models overestimate the wind speed with a bias up to 1.5 m/s. While these biases might result from the deficiency of RCMs, they may also come from the deficiency of the PGF data used for the validation. For instance, PGF data is a hybrid observation-reanalysis dataset created by combining global observation datasets and reanalysis datasets (NCEP-NCAR) (Sheffield et al. 2006). Hence, because of the very low density of the observational network over the country, the PGF data might not be able to capture the

Projected changes
The CORDEX ensemble models project an increase in daily electricity over the entire country at all the GWLs (Fig. 10a-d). However, the magnitude of the increase varies across Fig. 9 Spatial distribution of climate variables over Niger as depicted by PGFD and CORDEX RCMs ensemble mean in reference period . The climate variables are CDD (°C), HI (°C), T max (°C), humidity (%), radiation (Watt/m 2 ), and Wind (m/s). The dots indicate areas where the observation is within the RCMs spread; r shows the spatial correlation between the observation and simulation while asterisk (*) denotes correlation that is significant at 95% confidence level the country and grows with the increase of GWLs. For instance, at GWL1.5, the changes are rather homogenous (between 4 and 8% increase in CED) over the entire country. In addition, the changes are robust (i.e., statistically significant at a 99% confidence level). However, for GWL2.0, CED increase varies across the country, from 4 to 8% in the central part of the country to 8-12% in the remaining part of the country. Compared with changes at GWL1.5, an additional increase (up to 3%) in CED is observed over most parts of the country. Conversely, a further increase in warming level beyond 2 °C will enhance the CED over the entire country, such that, at GWL3.0, most parts of the country will become hotspots of increased in CED due to climate change. This suggests that failing to keep the global warming level below or at 2 °C (the level set by the Paris agreement) may have serious consequences for CED over the entire country. Indeed, an additional increase (up to 9.5% compared to GWL1.5) could be seen in most parts of the country, with the highest increase around Niamey. The increase in CED over the entire country is robust (i.e., statistically significant at a 99% confidence level) at all the GWLs. These findings are consistent with the notion that climate change will increase the electricity consumption of tropical countries (Santamouris et al. 2015;Scapin et al. 2015;Huang and Hwang 2016;Ang et al. 2017).
Moreover, Fig. 10 shows that the projected changes in CED are consistent with the changes in CDD, T max , HI, and humidity variables. For instance, the increase in CED may be attributed to increases in CDD, T max , HI, and humidity. This is expected since high CDD will require more CED for cooling purposes. In fact, Fig. 10e-h indicates that the spatial correlation between the changes in CED and CDD is very high (> 0.9) and significant (99% confidence level) at all the GWLs. Moreover, the increase in CDD is in agreement with the results of (Klutse et al. 2018) who found an increase in temperature over the region as a result of climate change. In the same way, the increase is also followed by increases in HI and T max . Indeed, a high T max would result in increased electricity demand peaks, hence contributing to increasing the overall CED. So the T max is an important factor that influences the CED. For instance, the spatial correlation between the changes in CED and CDD is high (> 0.9) and significant (99% confidence level) at all the GWLs. Finally, the changes in CED are also in agreement with the changes in humidity since our previous work has established that humidity and CED are negatively correlated (Bonkaney et al. 2019). So a decrease in relative humidity will lead to an increase in CED, which is observed in Fig. 10a-d. Nonetheless, for both radiation and wind, the projected changes are not consistent with changes observed in CED. For instance, one might have expected a decrease in radiation would result in a decrease in CED, because of the positive relationship between CED and radiation (Table 3). But, the reverse is the case. This might be due to the fact that the impact of the other climate variables (CDD, T max , HI, and humidity) overwhelms the impact of radiation on CED. Indeed, the spatial correlation between CED and radiation is weak (r < 0.5) and not significant. Similarly, an increase in wind speed could also have resulted in a decrease in CED because of the negative relationship between CED and wind (Table 3), but this is not the case. This can also be explained by the fact that the wind has a weak influence on electricity demand. The spatial correlation between wind and CED is weak (r < 0.3) and not statistically significant.
However, it is worth studying the impact of global warming on CED for individual months. Figure 11 shows that the impact of climate change differs from 1 month to another. Positive values indicate an increase, while negative values depict a decrease. Some variables, such as CED, CDD, and wind speed, would have a net increase for all the months at all the GWLs. The highest increase in CED and CDD are observed during the hot periods (March-June and October-November), whereas the lowest are in the cold periods (December-February and August). Hence, the impacts of global warming may be more severe in hot periods than the cold period. However, for wind, the highest is observed in November (about 0.1 m/s). The other months depict a slight increase (< 0.1 m/s) in wind speed associated with uncertainties. Conversely, for the variables, such as heat index (HI), humidity, and radiation, both positive and negative values can be observed depending on the months considered and the specific GWL. For example, the highest decrease in relative humidity is observed in May, with the magnitude of the decrease increasing with GWLs. But, a slight increase is observed in August for all the GWLs and September for the GWL1.5 and GWL2.0. Regarding the radiation, a general decrease can be observed except in May, where the changes are positive for GWL15.
The level of agreement among the models on the projection (a measure of robustness in the projected changes in the electricity demand over Niger) depends on the GWLs and the various variables (Fig. 12). In general, agreement among simulations is better for the projections of T max , CED, CDD, and HI than for humidity, radiation, and wind speed projections. For instance, almost all the simulations agree on the projections of the T max , CED, CDD, and HI for all the GWLs. This indicates that the projections of CED, CDD, and HI are robust at all the GWLs. The ensemble median of CED indicates an increase of about 5%, 7%, 12%, and 15% for GWL1.5, GWL2.0, GWL2.5, and GWL3.0 respectively. The least agreement among the simulations is observed for the radiation, where the simulations do not agree on the projections of these variables for any of the GWLs. Nevertheless, the ensemble median of radiation indicates a decrease for this variable, with the magnitude of the decrease increasing with global warming level. However, for the humidity and wind, more than 75% of the simulations agree on the projections at GWL2.5 and GWL3.0. As a result, changes in wind and humidity are only robust for warming levels greater than 2°. It may also be noted that for all the variables, the spread among simulations increases with increasing global warming (Fig. 12).

Conclusion
As part of the efforts to understand and quantify the impact of climate change on key economic sectors, this study has investigated the potential impacts of gradual global warming on electricity demand in Niger. The principal component analysis (PCA) was utilized to group the key climate parameters that influence the electricity demand. The  multiple linear regression (MLR) model has been developed to predict the electricity demand based on the PCA results. Moreover, for the projections, 14 multi-model regional climate simulations from the Coordinated Regional Climate Downscaling Experiment (CORDEX) at four specific global warming levels (GWL1.5, GWL2.0, GWL2.5, and GWL3.0) have been analyzed. The results are summarized below: • The principal component analysis (PCA) result revealed that the climates variables such as the air temperature, maximum and minimum temperature, relative humidity (RH), heat index (HI), cooling degree days (CDD), radiation, and wind are highly correlated with the climate-electricity demand. • The stepwise regression results suggest that only the variables such CDD, humidity, heat index, radiation, wind, and T max are statistically significant at 99% confidence level. The accuracy of the regression results shows a high value of coefficient of determination R 2 (0.808) and a reasonable root mean square error (140.87 MWh/day). Moreover, the residual plots indicated that the residuals from the regression model are normally distributed, suggesting that the model complies with the assumptions of regression models. • The CORDEX simulations realistically reproduce the annual cycle of the ED and climate variables used in this study and in most cases; the observed annual cycle is within the RCMs ensemble spread. However, discrepancies do exist between the individual simulations. • The CORDEX simulations project an increase in CED, T max , HI, CDD, and wind and a decrease in humidity, and radiation at all GWLs. The highest increase in CED is projected in hot periods (Mach-June) and October-November. • The simulations agree on the projections of CED, HI, and CDD at all GWLs. Conversely, there is no agreement among simulations for the radiation at any of the global warming levels. However, more than 75% of the simulations agree on the projections of wind and humidity at GWL2.5 and GWL3.0.
These findings have implications for mitigating the effects of climate change in the energy sector. The results show that climate change would increase the electricity drivers, including the cooling degree day (CDD), heat index (HI), and T max , while decreasing the relative humidity and the radiation over Niger. The overall impacts would significantly increase the electricity demand, thereby overstretching the supply. These findings agree with those of Ahmed et al. (2012); Invidiata et al. (2016); Craig et al. (2018), and Shourav et al. (2018) who also found an increase in demand as a result of climate change. The projected increase in climate-induced electricity demand over Niger (5-15%) in this study is comparable to that projected over Dhaka City (5.9-15.6%; Shourav et al. 2018) though lower than those projected over Brazil (56-112%; Invidiata et al. 2016). In Niger, the government is still struggling to keep up with the escalating demand for electricity induced by population and economic growth. The additional increase due to climate change would further exacerbate the problem. As most policies and plans on energy management in Niger do not account for the impact of climate change, the impacts might substantially affect electricity dispatching and lead to supply shortfalls. Policymakers in the energy sector can use the results of the study as a basis for developing policy and strategy to reduce the future impact of climate change on the energy sector countrywide. They can also use it to advocate for proactive actions toward increasing the resilience and adaptive capacity of the power sector.
To provide more robust information for policymakers, the results of the study can be improved in different ways. First, besides the factors related to climate, other factors such as population, GDP, policy, consumer behavior, urbanization, and so on may also determine the future electricity demand. For instance, climate influences the electricity demand through the response of people to weather (Valor et al. 2001). In other words, depending on the weather conditions, people will increase or decrease the demand. Secondly, the current study used aggregated electricity demand, including the residential, commercial, and industrial sectors since disaggregated data were not available. So, the future may look at the impacts of global warming on different sectors. Thirdly, the study also used electricity demand data for Niamey since the data for other locations were not available. Hence, subject to the availability of the relevant data, future studies could investigate how the disparity may alter the results of the present study. Conducting biased corrections on GCMs and RCMs simulations may further reduce the disagreement among the models for the projections of humidity, radiation, and wind. Such considerations will make the results more relevant for policymakers. Nevertheless, the present study has shown that climate change could increase electricity demand in Niger.