Work

An integrated framework for assessing the fidelity of climate reanalysis data for urban applications

Year: 2026
Type: dissertation
Abstract: Modern reanalysis products provide long, continuous, spatially consistent, and temporally resolved climate records, making them increasingly important for urban applications. So far, the suitability of reanalysis data in urban applications is commonly judged by goodness-of-fit measures that quantify the match between the reanalysis and reference climate time series. This approach, however, does not establish whether application-relevant information in urban contexts, such as trend, variability, and/or extreme values of intensity-, threshold-, and frequency-based climate indicators are preserved by reanalysis products. We propose a structured, fit-for-purpose benchmarking framework that evaluates the fidelity of reanalysis data through accuracy, robustness, and reliability across three urban application pillars related to model forcing, impact assessment, and design and risk analysis using extreme-value analysis. Rather than asking whether a reanalysis product is generally accurate, the framework asks whether the specific quantity required by an application is sufficiently accurate, robust, and reliable for its intended use. The proposed framework is demonstrated using ERA5-Land near surface air temperature and its bias-corrected versions across 40 major Canadian cities during 1950–2024. Raw ERA5-Land is compared with three commonly used bias-correction methods, Linear Direct Scaling, Non-Linear Scaling, and Equidistant Quantile Mapping, using a rotating, temporally blocked out-of-sample calibration and evaluation design. The performance is assessed over the complete record and through rolling 30-year windows, and spatial dependencies on latitude, longitude, and elevation are evaluated. Our results showed that ERA5-Land’s strongest fidelity is for daily temperature minimum, mean, and maximum at daily-to-seasonal scales, whereas annual aggregation and diurnal temperature range consistently showed weaker and less robust performance. The recurring weakness of diurnal temperature range further showed that good representation of daily minimum and maximum temperatures separately does not guarantee faithful representation of their same-day difference. Bias correction largely removed mean biases and often improved direct agreement, but these gains did not consistently propagate to climate indicators and/or extreme values. For impact assessment, raw ERA5-Land produced smaller errors in reconstructing trends and variability across nearly 60% of seasonal and annual comparisons, while errors in 30-year trends were generally unstable across cities. For extreme-value analysis, bias correction improved several extreme quantiles for annual minima and maxima of daily temperature, but provided little benefit for annual maximum diurnal temperature range, and did not consistently improve the reconstruction of associated 95% confidence intervals. Extreme-quantile and confidence-interval errors also became less stable toward longer return periods, for which extrapolation beyond the available record increased uncertainty. Spatially structured errors also persisted or emerged after bias correction. These findings reveal that reanalysis usability is application-contingent and that bias correction should be evaluated as a testable preprocessing hypothesis rather than a guaranteed pathway for universal improvement. The proposed framework provides a transferable basis for benchmarking reanalysis data across variables, scales, indicators, locations, and contexts relevant to urban applications.
Author: Zokaei, Amir
Language: English
Open Access status: green
OpenAlex ID: W7221499166
Cites: 168
Cited by:
Related to: 10