“Weather forecasting” of indoor air temperatures in Amsterdam (The Netherlands) to facilitate early heat warnings.

The ongoing climate change results in increasing frequency of warm episodes and heat waves. While a lot of studies focuses on understanding and predicting outdoor air temperature and human comfort, less is known about indoor temperatures that are reached during these warm episodes. This is surprising since people generally spend the majority of their time indoors, e.g. at work or in their living or bed room. This study develops and evaluates a relatively simple forecasting system that aims to make five day forecasts for real world room temperatures. The system builds upon a relatively simple physical-statistical model for the heat budget of a room that is forced by outdoor weather variables like solar radiation, wind speed, air temperature. In addition, room temperature observations from ~60 houses in Amsterdam are used to train this physical-statistical model. After training the calibrated model is used for forecasting and driven by ECMWF operational forecasts of outdoor meteorological variables for the city of Amsterdam (The Netherlands) for the summer of 2023. Room temperatures are initialized daily by room temperature observations from these ~60 houses. We evaluate the forecasting system on the summer of 2023. We find the model system is well to produce a meaningful room temperature forecast for most of the houses. Forecasts for the daily mean temperature of the living room outperforms the forecast for the daily mean bed room temperatures. The median RMSE (over all 60 houses) for the living room forecast increases from 0.38 K for the one-day forecast to 0.95 K for the five-day forecast. Forecasts for daily mean temperatures are better than for daily maximum temperatures. With this forecast tool we aim to study whether citizens adapt their behaviour in protecting their homes from indoor heat when they receive the forecast information.