Spatially Explicit Correction of Simulated Urban Air Temperatures Using Crowdsourced Data

Area of Science:

• Urban climatology
• Environmental modeling
• Data science

Background:

• Urban climate models require robust observational data for accurate evaluation.
• Existing official weather stations offer limited spatial coverage in urban areas.
• Personal weather sensors (PWSs) present a dense, potential data source for urban climate research.

Purpose of the Study:

• To assess the utility of quality-checked PWS data for evaluating urban climate models.
• To develop and apply a novel bias correction method for urban air temperature simulations.
• To improve the accuracy of urban climate model outputs for impact studies.

Main Methods:

• Simulated near-surface air temperatures over London using the Weather Research and Forecasting (WRF) Model with building effect parameterization (BEP-BEM).
• Evaluated model performance against 402 urban PWSs, identifying heterogeneous cool biases.
• Employed a machine learning approach for spatially explicit bias correction of modeled air temperatures.

Main Results:

• PWS data revealed spatial biases in WRF model simulations not captured by official weather stations.
• The developed machine learning bias correction effectively adjusted daily minimum, mean, and maximum temperatures.
• The bias correction method accounts for nonlinear, spatially heterogeneous biases independent of urban fraction.

Conclusions:

• Quality-checked PWS data are crucial for enhancing urban climate model evaluation.
• Spatially explicit bias correction using PWS data significantly improves model accuracy.
• A framework for PWS-driven bias correction is recommended for future urban climate impact assessments.