LSTM Neural Networks to Forecast Urban Heat Island Mitigation with Urban Improvements

An analysis of summer Urban Heat Island (UHI) is underway at the University of Modena and Reggio Emilia for the Municipality of Carpi, a town of seventy-two thousand inhabitants located in the central Po Valley, Italy. A work in progress is described here for the application of Long Short-Term Memory Neural Networks (LSTM NN) to a range of data typologies wider than previous research. A network of IoT control units hourly provides air temperature, pressure, relative humidity, wind, solar radiation, and rain. The study to better integrate the LSTM NN meteorological datasets with those of urban characteristics is now underway. The surface data come from WV3 satellite imagery, aggregated energy consumption counters and traffic control units. The predictive model of the trained LSTM NN will forecast, in the same areas now monitored by every climate control unit, the reduction of air temperature following improvements in surrounding urban fabric: more green, gray asphalts, energy retrofitting of buildings, etc. The simulations will provide hourly forecasts of air temperature under different weather conditions, represented by data that extends back to the previous days. Different combinations of urban improvements will be tested to obtain the best cost-effective temperature behavior.