Forecasting global building internal heat gains to 2100: an applied XGBoost model driven by shared socioeconomic pathways

Accurate long-term building energy forecasting, essential for climate change mitigation, is often hindered by static internal heat gain assumptions that neglect dynamic socioeconomic shifts. This study addresses this gap in two distinct stages. First, we develop an XGBoost ensemble, leveraging real-world historical data and countrylevel indicators projected under five Shared Socioeconomic Pathways (SSPs), to forecast key internal heat gains—household sizes and energy intensities for major end-uses—globally through 2100. Validation against a hold-out set of historical data shows high performance metrics (global R2 ≥ 0.96), though interpretation requires caution due to data sparsity in some regions. Second, in an illustrative application, the generated dataset was integrated into building energy simulations for a prototype multi-apartment building across diverse locations. These simulations revealed that using dynamic, projected internal heat gains leads to substantial divergence from standard static assumptions: notably, up to 30 % higher energy demand in developing countries and up to 27 % lower demand in developed nations. These divergent outcomes fundamentally recompose the building’s energy balance, highlighting critical trade-offs between heating and cooling loads. This research exposes the limitations of using single, static values in international building codes and underscores the urgent need for context-specific, dynamic inputs. To this end, we provide an open dataset to support more robust future building energy analyses, inform the design of resilient, “future-proof” buildings, and enhance policy formulation.