Reliable Interpretation of Extreme-Heat Energy Response for Sustainable Campus Management: Baseline, Calendar, and Chilled-Water Metering Controls

Abstract

Open hourly meter data are often used to infer how buildings respond to extreme heat, but the interpretation can depend on the analytical baseline, institutional calendar, and meter configuration. We examined 60 education and office buildings in the Building Data Genome Project 2 using repeated whole-week validation, temperature-matched placebo tests, 13 academic-calendar definitions, and same-building, common-hour carrier comparisons. A leakage-resistant change-point model that incorporated a previous-day, loadderived operating-regime proxy reduced blocked-validation error relative to a conventional model in 58 of 60 buildings; temperature binning provided no additional predictive benefit after regime control. Recovery-related signals were uncommon: rebound was significant in 2 of 49 eligible buildings, load-shape disruption in 1 of 49, and the 72 h duration diagnostic in 3 of 60. At Fox, the BASE calendar supported term and break comparisons through 39–42 ◦C, but neither building-use group differed significantly after false-discovery-rate adjustment; estimates above 42 ◦C were inconclusive. In the matched sample of 13 education and 10 office buildings, no electricity-only, chilled-water-only, or weighted specification provided robust evidence of a between-group difference. Across 99 combinations of carrier weight, normalization, and temperature bin, every 95% interval included zero. Reliable campus benchmarking therefore requires out-of-sample baseline assessment, explicit calendar sensitivity, common-hour meter matching, and carrier-specific uncertainty reporting before building-use rankings or resilience-related interpretations are made.