Country-level energy demand for cooling has increased over the past two decades

The study examines how warming temperatures and population growth are driving cooling energy demand in buildings worldwide. It focuses on country-level changes in cooling degree days (CDD) between 2000 and 2020, using population-weighted metrics and a humidity-adjusted formulation (CDDhum) to better represent human thermal stress and potential cooling demand. The research addresses two questions: (1) how national CDD and CDDhum have changed over the past two decades, and (2) whether intense cooling demand days are becoming more temporally clustered, which has implications for peak loads, storage needs, and grid reliability. The authors motivate the work by noting rapid growth in cooling demand, especially across populous tropical and subtropical countries with increasing AC penetration, and emphasize that ignoring humidity or population distribution can misrepresent both magnitude and spatial patterns of cooling needs.

Prior work shows rising global energy use with buildings accounting for a large share and significant CO2 emissions. In IEA member countries, heating dominates residential energy use, but cooling is the fastest-growing end-use in buildings with rapidly increasing AC ownership, especially in developing and tropical regions. Degree days (HDD/CDD) are widely used to approximate heating/cooling needs and their trends under climate change; studies indicate decreasing HDD and increasing CDD, with population weighting altering regional averages (lower HDD, higher CDD in Europe) and better aligning with energy loads. Sensitivity of electricity demand to temperature varies by OECD status and urban context. Clustering methods and regression analyses have been applied to US cities and Brazilian demand management, while climate risks to energy supply include drought-induced cooling water shortages for thermoelectric plants. Literature on temporal clustering of degree-day-derived extremes is limited, though clustering has been studied for other hazards (e.g., hurricanes, European temperature extremes, extra-tropical cyclones). This study fills gaps by assessing both trends and temporal clustering of CDD and humidity-corrected CDD at national scale, employing population weighting and focusing on emerging, heat-stressed countries.

Data: The IEA–CMCC Weather for Energy Tracker (ERA5-based) provides energy-relevant climate indicators at 0.25° resolution from 2000–2022 for 234 countries. Population weighting uses NASA SEDAC GPWv4 gridded population (0.25°) consistent with national censuses (interpolated between 2000–2020). Country-level derived indicators (e.g., CDD) are provided as population-weighted aggregates. This study uses CDD at threshold X=21 °C (CDD21) and humidity-corrected CDD at 21 °C (CDDhum21), nationally averaged with population weights. Warm season definition: Based on maximizing six-month mean warmth across countries, the warm season is defined as May–October (MJJASO) for the Northern Hemisphere and November–April (NDJFMA) for the Southern Hemisphere. Country selection for detailed clustering analysis: Countries were ranked by the 75th percentile (75p) of daily CDDhum21 during 2000–2020 warm seasons; the top decile (23 countries) was selected. The ranking is invariant if using the 90th percentile or different base thresholds (21, 23, 26 °C). Trend analysis of annual potential cooling demand: For each country, yearly warm-season cumulative CDDhum21 and CDD21 (°C-year) were computed and linear regressions over 2000–2020 performed to quantify trends (global mapping in Fig. 1; country examples in Figs. 2–3). Definition of intense days and binary series: Intense days are those with daily CDDhum21 (or CDD21) above the country-specific 90th percentile threshold computed over 2000–2020 warm seasons. A binary time series marks intense days (1) and non-intense days (0). Clustering metrics: (1) Coefficient of Variation (CV) of inter-event times (days between intense days, counting consecutive non-intense days; zero for consecutive intense days). CV>1 indicates overdispersion and temporal clustering. CV computed over 2000–2020 and separately for decades 2000–2009 (I) and 2011–2020 (II). (2) Ripley’s K function in time: average number of intense days within t=10 days around each intense day; compared across decades using bootstrap significance and against Monte Carlo homogeneous Poisson processes with equivalent rates. (3) N: number of intense days per decade; N_C: number of clusters (groups of ≥2 consecutive intense days) per decade; M_C: maximum cluster duration (days) per decade. Decadal comparison and significance: Differences between decades I and II were assessed using bootstrap (simple resampling with replacement) to obtain 95% confidence intervals; significance at 5% reported. Analyses were conducted for the top-23 warmest countries (by CDDhum21 75p) and extended globally in supplementary tables when at least one parameter changed significantly. Aggregation sensitivity: Two approaches were compared: (a) national analysis using pre-aggregated, population-weighted country time series; (b) grid-point analysis computing thresholds and clustering metrics at each 0.25° grid cell, then averaging indicators with population weights. Agreement in magnitudes, trends, and significance was evaluated (Tables 2/5 vs 4/7).

- Global increase in cooling demand: Population-weighted CDD and CDDhum exhibit significant positive trends globally over 2000–2020, with stronger magnitudes when humidity is included. Tropical and densely populated countries (e.g., India, Bangladesh, Thailand) show pronounced increases. - Top-23 warmest (by CDDhum21 75p): Concentrated in the Northern Hemisphere tropics/subtropics (Central Africa, Arabian Peninsula, South Asia). Bahrain has the highest CDDhum21 levels, rising from ~3200 to ~3600 °C-year over 20 years. Bangladesh shows the steepest CDDhum21 trend (~22.31 °C-year per year) with large increases also in Thailand and India. - Intensification and clustering of extremes (CDDhum21): The total number of intense days per decade (N) increases in nearly all countries; in Thailand N rose from 70 (2000–2009) to 298 (2011–2020). N approximately doubled across countries, and tripled for Vietnam, India, Bangladesh, and Cambodia. Coefficient of Variation (CV) values over 2000–2020 exceed 2 in all 23 countries, with some >4 (Burkina Faso, Mali, Benin), evidencing clustering. CV rose markedly in Cambodia, Burkina Faso, Mali, Niger, India, Thailand, and Vietnam (e.g., Vietnam from 1.89 to 3.33). Ripley’s K function exceeded 10 days in the second decade for Bahrain, Kuwait, Saudi Arabia, Oman, and India, indicating denser clustering; K more than doubled in Thailand and Vietnam. In some cases (Mauritania, Bangladesh) K increased even when CV did not. - Cluster counts and sizes (CDDhum21): N_C and M_C generally increased in 2011–2020. Maximum cluster size (M_C) doubled in Vietnam and Thailand and tripled in Pakistan and Saudi Arabia (e.g., Saudi Arabia from 15 to 49 days). Despite large M_C, some Arabian countries (Bahrain, UAE, Gambia, Saudi Arabia) did not show significant N_C growth; most others saw N_C roughly double (e.g., Vietnam 22→60; India 13→47; Thailand 17→57; Bangladesh 21→71). - Dry vs humid differences: Patterns are broadly similar for CDD21 but less pronounced without humidity. Egypt emerges with the largest dry CDD21 trend (~14.01 °C-year per year), followed by Saudi Arabia, UAE, and Kuwait (~7.74 °C-year per year). Some countries show clustering increases only in one formulation: Saudi Arabia (dry only), India (humid only). Hong Kong’s prominence appears mainly with humidity; Egypt’s with dry. In dry conditions, K>10 days in the second decade occurs in UAE, Saudi Arabia, Chad, Benin, India; Saudi Arabia’s K more than doubled when humidity is ignored. Humid countries tend to have more but smaller clusters than dry regions; a record 37 consecutive dry CDD21 intense days occurred in Saudi Arabia in 2017. - Temporal evolution and variability: Peaks widen and cluster more in the latter decade (notably 2016–2017) in several countries (e.g., Bahrain, Iraq, Hong Kong, Pakistan). Arabian Peninsula countries show high interannual variability of intense-day statistics. - Aggregation robustness: Grid-point, population-weighted results (Tables 4 and 7) largely corroborate national pre-aggregated analyses (Tables 2 and 5) for N, N_C, and M_C, with N_C particularly robust to method choice. Some differences appear for specific countries (e.g., Iraq’s N trend differs between approaches; M_C magnitudes can vary slightly), but overall conclusions are consistent. - Policy-relevant implications: Increased frequency and temporal clustering of intense cooling demand episodes imply greater and more sustained peak loads, stressing grids especially in emerging economies with rising AC penetration and limited energy access. Humidity substantially alters both magnitude and clustering patterns, underscoring the need for humidity-aware planning.

The analysis demonstrates that country-level potential cooling demand has risen markedly over the past two decades and that intense cooling demand days are increasingly clustered in time. This directly addresses the research questions by showing both trend amplification (annual cumulative CDD/CDDhum) and statistically significant clustering (higher CV, larger K, more clusters, longer maximum clusters) across many heat-exposed countries. Incorporating humidity materially changes the assessment, revealing stronger warming signals and different clustering behavior, especially in tropical, humid regions (e.g., India, Hong Kong), and should be preferred when estimating human heat stress and cooling needs. These findings have practical significance for energy systems planning: temporal clustering of intense cooling demand elevates requirements for peak generation capacity, grid reliability, and storage, challenging countries where AC adoption and urbanization are accelerating. The results suggest traditional planning based on stationary 90th-percentile thresholds may underprepare systems for more frequent and persistent clusters. The demonstrated robustness of the results to aggregation method (national series vs grid-point analysis) supports use of readily available, pre-aggregated country indicators for monitoring and planning. Differences between humid and dry formulations indicate that local climate (humidity) must inform policy and infrastructure strategies, as ignoring humidity can misidentify priority regions or underestimate clustering-related risks.

Using a global, population-weighted dataset of ERA5-derived cooling indicators, the study shows that: (1) CDD and humidity-corrected CDD (CDDhum) increased significantly worldwide during 2000–2020; (2) intense cooling-demand days have become more temporally clustered across many of the warmest countries; and (3) humidity is crucial for both magnitude and clustering of cooling demand, especially in emerging, densely populated tropical regions. India, Cambodia, Thailand, and Vietnam stand out for strong increases in frequency and clustering of intense events; Bahrain and other Arabian Peninsula countries exhibit high magnitudes and variability. The work contributes a reproducible, globally consistent framework for tracking cooling demand and its clustering using both dry and humid CDD, evaluating decadal changes, and validating aggregation choices. Future research should: integrate socio-economic drivers (AC penetration, building stock, electrification, income, tariffs); assess grid and market impacts of clustering (capacity adequacy, storage sizing, demand response); explore urban heat island effects and land-use change; and extend analyses with climate projections to inform adaptation pathways under different emissions scenarios.

- Time span: Analyses focus on 2000–2020 (with dataset availability to 2022), a relatively short period for climate trend detection; decadal comparisons may be sensitive to interannual variability. - Reanalysis and population data: ERA5 and GPWv4 introduce uncertainties (e.g., reanalysis biases; interpolated population grids between census years). Population-weighted CDDs reflect exposure potential rather than realized consumption. - Indicator choices: Results depend on selected thresholds (base temperature 21 °C, intense-day threshold at 90th percentile, K window t=10 days) and warm-season definition (MJJASO/NDJFMA). Alternative thresholds or seasons could yield different magnitudes, though rankings were robust. - Socioeconomic and infrastructure factors: Actual cooling energy demand also depends on AC ownership, building efficiency, electricity access, prices, and behavior; these are not explicitly modeled. - Aggregation nuances: While overall robust, some differences in M_C and N trends between aggregated and grid-point methods suggest regional heterogeneity could affect specific metrics and warrants further investigation.