Implementing Q Learning and Multimethod Data Analysis to Determine Land Surface Temperature and Heat Indexed Routes at the Neighborhood-Scale

Abstract:

Initially developed in response to social distancing protocols mandated by the COVID-19 pandemic, open streets have sprouted in several locations within New York City. Although well-intentioned, their placement has been pursued with little scientific validity and with little consideration to their wide-reaching effects. For example, while open streets can alleviate urban heat by halting anthropogenic activity such as motor vehicle traffic, the program lacks representation in low-income communities that may benefit from it the most. As a result of its sparsity but high effectiveness, the open street initiative should be accompanied by complementary efforts. For instance, routing tools responsible for plotting cool paths may be implemented. Such tools would allow for greater public heat resilience and help New Yorkers adapt to the impacts of the urban heat island effect (UHIE), particularly as the climate warms. Routing tools and methodologies could also be used to: a) identify appropriate streets for the open streets initiative; b) determine cool corridor pathways; c) provide scientific guidance to policymakers as they contemplate expanding the open streets program; d) protect heat vulnerable populations, and e) help in identifying and ameliorating the root causes of the UHIE. Pedestrian pathways and roads in Bedford-Stuyvesant, Brooklyn were analyzed using summer average and single day Landsat 8 satellite observations from 2013 to 2020 to determine land surface temperature hotspots, cool spots, and areas of high and low heat indices for preliminary use in a deep Q learning network. Additionally, using conventional Q learning, an agent responsible for determining the coolest and hottest pathways by land surface temperature from some geographic starting point to some geographic ending point was created. Preliminary results show that pathways bordered by low residential buildings, few walls, and much vegetation are preferred. Therefore, given adequate computing power, required data, and further development, these heat-indexed routing systems may be expanded beyond neighborhood-scale applications and used as urban climate science tools for global cities as they strive for climate change resilience and robustness.