The Kaya identity is a powerful index displaying the influence of individual carbon dioxide (CO₂) sources on CO₂ emissions. The sources are disaggregated into representative factors such as population, gross domestic product (GDP) per capita, energy intensity of the GDP, and carbon footprint of energy. However, the Kaya identity has limitations as it is merely an accounting equation and does not allow for an examination of the hidden causalities among the factors. Analyzing the causal relationships between the individual Kaya identity factors and their respective subcomponents is necessary to identify the real and relevant drivers of CO₂ emissions. In this study we evaluated these causal relationships by conducting a parallel multiple mediation analysis, whereby we used the fossil fuel CO₂ flux based on the Open-Source Data Inventory of Anthropogenic CO2 emissions (ODIAC). We found out that the indirect effects from the decomposed variables on the CO₂ flux are significant. However, the Kaya identity factors show neither strong nor even significant mediating effects. This demonstrates that the influence individual Kaya identity factors have on CO₂ directly emitted to the atmosphere is not primarily due to changes in their input factors, namely the decomposed variables.

Urban electric vehicle (EV) charging data display strong station-level heterogeneity, nonstationary patterns, and sharp peaks with long-tail surges. These characteristics undermine conventional forecasting: zone-specific models face data scarcity and unstable scales; global models dilute cross-zone differences; and graph or sequence architectures handle abrupt surges and drift under distribution shifts poorly. Therefore, we propose Peak-Aware Residual LLM (PAR-LLM) framework decoupling dense numerical learning from context-conditioned adjustments. A global long short-term memory (LSTM) with zone embeddings provides a stable baseline by learning short-term dynamics from normalized station data. A risk localizer identifies critical periods including sudden surges or daily peaks by analyzing historical volatility. For flagged pairs, a selective large language model (LLM) adjusts residuals based on recent statistics and calendar cues. These adjustments are merged with the numeric anchor through a discrepancy-aware weighting scheme, while guardrails constrain temporal variation and apply a soft reversion toward hour-of-day baselines to maintain physical plausibility. Evaluated on the Urban EV hourly dataset (120 zones) across three forecast horizons, the hybrid method outperforms graph-, sequence-, and tree-based models while keeping interventions sparse and latency controlled. The design addresses scale mismatch, cross-zone heterogeneity, and peak sensitivity by limiting language-driven corrections to high-risk hours and relying on the numeric anchor for routine behavior.