![Thumbnail](/bitstream/handle/20.500.11850/457358/1-s2.0-S2212095520301061-main.pdf.jpg?sequence=6&isAllowed=y)
Open access
Date
2021-01Type
- Journal Article
Abstract
Understanding the patterns of urban temperature a high spatial and temporal resolution is of large importance for urban heat adaptation and mitigation. Machine learning offers promising tools for high-resolution modeling of urban heat, but it requires large amounts of data. Measurements from official weather stations are too sparse but could be complemented by crowd-sensed measurements from citizen weather stations (CWS). Here we present an approach to model urban temperature using the quantile regression forest algorithm and CWS, open government and remote sensing data. The analysis is based on data from 691 sensors in the city of Zurich (Switzerland) during a heat wave using data from for 25-30th June 2019. We trained the model using hourly data from for 25-29th June (n = 71,837) and evaluate the model using data from June 30th (n = 14,105). Based on the model, spatiotemporal temperature maps of 10 × 10 m resolution were produced. We demonstrate that our approach can accurately map urban heat at high spatial and temporal resolution without additional measurement infrastructure. We furthermore critically discuss and spatially map estimated prediction and extrapolation uncertainty. Our approach is able to inform highly localized urban policy and decision-making. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000457358Publication status
publishedExternal links
Journal / series
Urban ClimateVolume
Pages / Article No.
Publisher
ElsevierSubject
Random forest; Urban heat; Low-cost sensors; Crowd-sensing; Machine learningOrganisational unit
09576 - Bresch, David Niklaus / Bresch, David Niklaus
03777 - Knutti, Reto / Knutti, Reto
Funding
167215 - Combining theory with Big Data? The case of uncertainty in prediction of trends in extreme weather and impacts (SNF)
More
Show all metadata