Does LCZ map accuracy matter? Sensitivity of WRF urban climate simulations during a heatwave in Amsterdam

Urban climate simulations increasingly rely on Local Climate Zone (LCZ) classifications to represent urban morphology and land–atmosphere interactions in numerical weather prediction models such as the Weather Research and Forecasting (WRF) model. Most studies assume that using LCZ maps with higher classification accuracy leads to the most realistic urban climate simulations; however, the sensitivity of modeled intra-urban temperatures to the accuracy of LCZ classification maps remains insufficiently explored. In this study, we investigate how the classification accuracy of manually generated LCZ maps, produced using the LCZ Generator and evaluated through cross-validation accuracy metrics, influences simulated urban temperature fields during a heatwave event in Amsterdam, the Netherlands. We performed ten WRF simulations for a heatwave period in early September 2022, including one default land-use simulation and nine simulations using LCZ maps with classification accuracies from 0.20 to 0.84. Modeled 2-m air temperature outputs were evaluated against observations from the Amsterdam Atmospheric Monitoring Supersite urban meteorological network.

Results show that LCZ map selection substantially affects simulated urban air temperature, with modeled temperatures consistently underestimating observations by 1.2 to 2.4 °C across the urban network, depending on the LCZ map used. The highest-accuracy LCZ map (0.84) produced the best overall temperature performance (1.2 °C mean absolute bias), followed by the default land-use map. However, several low- to medium-accuracy LCZ maps (e.g., accuracies from 0.31 to 0.48; mean absolute bias ≈ 1.9 °C) performed similarly to or even better than some higher-accuracy maps (e.g., accuracies from 0.75 to 0.81; mean absolute bias ≈ 2.2 °C), indicating that LCZ classification accuracy alone does not fully determine model performance. These findings demonstrate that LCZ map selection can substantially influence modeled intra-urban heat distribution, but increasing LCZ map classification accuracy does not necessarily translate into proportionally improved urban climate simulations. The results highlight the importance of evaluating LCZ datasets within modeling frameworks rather than relying solely on classification accuracy metrics, with implications for urban climate modeling, heat risk assessment, and prioritization of urban mapping efforts.

Acknowledgements. The authors acknowledge support from the 4TU-program HERITAGE (HEat Robustness In relation To AGEing cities), funded by the High Tech for a Sustainable Future (HTSF) program of 4TU, the federation of the four technical universities in the Netherlands.