Global climate models (GCMs) are important tools for simulating present and future climates, widely used to project potential climate scenarios and assess the impacts of climate change on a global scale (IPCC, 2023a). However, GCMs often operate at relatively coarse spatial resolutions (ranging from 0.5 to 2.5°) and temporal scales (e.g., daily or monthly, seasonally, or annually), which can limit their ability to accurately represent fine-scale regional climate variations (IPCC, 2023b; Keller et al., 2022; Tabari et al., 2021). This underscores the need for higher-resolution climate data to improve regional impact assessments. To address this, regional climate models (RCMs) are commonly employed to dynamically downscale GCM outputs, generating data with finer spatial resolutions (typically between 0.1 and 0.5°). RCMs provide several potential benefits over GCMs, such as the ability to better represent local topographical features, improve simulations of extreme weather events, and offer more refined regional climate projections, particularly in areas with complex terrain and climate variability (Chokkavarapu and Mandla, 2019; Kendon et al., 2010; Vichot-Llano et al., 2020). Convection-permitting models (CPMs) have become a valuable tool in short-range weather forecasting due to their enhanced ability to represent convective processes in detail, improve forecast accuracy, and capture localised high-impact rainfall that coarser models often miss (e.g., Done et al., 2004; Lean et al., 2008; Roberts and Lean, 2008; Weisman et al., 2008; Weusthoff et al., 2010). CPMs have also been widely applied in Europe (Berthou et al., 2020; Pichelli et al., 2021), Asia (Murata et al., 2017; Yun et al., 2020), and Africa (Kendon et al., 2019a; Maurer et al., 2017), as well as other regions(Trapp et al., 2011), to enhance the representation of convective processes and improve the accuracy of climate projections.

The UK Climate Projections 2018 (UKCP18) provide the latest generation of national climate projections for climate impact studies for the UK (Murphy et al., 2018). Several recent studies have employed the UKCP18-RCM 12 km regional perturbed parameter ensemble (PPE), which was released in November 2018, to assess the potential impacts of climate change. These applications include investigations into river and groundwater flow (Hannaford et al., 2023; Kay et al., 2023), extreme weather events (Hanlon et al., 2021), flood risks (Gudde et al., 2024), and droughts (Reyniers et al., 2023). In 2019, an additional UKCP18 toolkit was introduced, featuring the 2.2 km Convection-permitting model (UKCP18-CPM), which offers access to reliable climate information at local and hourly scales (Kendon et al., 2019b).

Despite their widespread use, climate model projections often show systematic biases relative to observations (Kotlarski et al., 2014; Vautard et al., 2021). For example, the UKCP18-RCM 12 km ensemble shows biases, including overestimation of winter precipitation, particularly in the mean and the 95th percentile of daily precipitation (representing heavy precipitation events) across most regions, and strong spatial variability in summer, with overestimations in the north and underestimations in the south (Reyniers et al., 2025). For temperature, the ensemble shows an overall cold bias at the annual scale in the mean and extremes. In winter, biases vary spatially and suggest overestimated variability, whereas in summer temperatures are more consistently underestimated across the UK (Reyniers et al., 2025). Building on the UKCP18-RCM, the UKCP18-CPM offers improvements, though notable issues persist. In winter, it remains too wet across most of the UK, while in summer it is too wet in the north and too dry in the south (Kendon et al., 2021). For temperature, the UKCP18-CPM slightly reduces UK-wide biases in summer compared to the UKCP18-RCM, although it remains cooler in the far north and exhibits minimal differences elsewhere (Kendon et al., 2021). In winter, the UKCP18-CPM introduces colder biases, particularly in northern regions, which are linked to its improved representation of snow processes, leading to greater snow accumulation and cooler temperatures (Kendon et al., 2021).

These biases, if left uncorrected, can severely affect the accuracy of climate impact assessments in sectors such as hydrology, ecology, and agriculture. The non-linear responses of impact models to these biases can amplify errors in projections, underscoring the importance of applying bias correction (BC) before using climate data in impact studies. In this paper, we use the term bias correction (BC) following common usage in the climate impact literature. However, these methods do not remove model biases in a physical sense; rather, they statistically adjust model outputs to better match an observational reference over a calibration period. We adopt this interpretation when discussing our results. Various BC methods, such as statistical downscaling, linear scaling, local intensity scaling, histogram equalizing, rank matching, and quantile mapping (QM), have been developed and evaluated (Gutmann et al., 2014; Maraun et al., 2019; Teutschbein and Seibert, 2012). Amongst these, QM has emerged as a particularly effective approach (Lafon et al., 2013; Shah et al., 2024; Themeßl et al., 2011), not only adjusting the mean but also correcting standard deviations and percentiles, making it suitable for temperature and precipitation adjustments in climate impact studies (Fang et al., 2015; Themeßl et al., 2012; Wilcke et al., 2013). For example, Cannon et al. (2015) evaluated the QM for bias correction of GCM precipitation outputs, emphasizing its effectiveness in adjusting mean values and precipitation extremes, while Sangelantoni et al. (2019) demonstrated its applicability for RCM temperature and precipitation corrections over Central Italy, particularly for seasonal and extreme climate variables.

In the UK, the UKCP18-RCM ensemble dataset has been widely used and bias-corrected for applications such as river flow, flood risk, and drought assessments (Hannaford et al., 2023; Robinson et al., 2023; Smith et al., 2025). In contrast, bias correction of the UKCP18-CPM ensemble remains limited. Kay and Brown (2023) applied a simple monthly scaling method to correct daily UKCP18-CPM precipitation data from the ensemble member 01 (EM01) against observations and kept temperature data uncorrected. However, this monthly scaling applied a uniform adjustment to all daily precipitation values within each month. It corrects monthly means but does not explicitly adjust the within-month distribution (e.g., quantiles and extremes). As a result, it may not accurately correct biases in extreme precipitation events. It also fails to correct the diurnal cycle of precipitation, which is a key feature of high-resolution climate models (Bannister et al., 2019; Scaff et al., 2019). This is consistent with Faghih et al. (2022), who compared two bias-corrected time series using a multivariate quantile mapping method, both with and without correction of the diurnal cycle, and found that bias correction of the diurnal cycle for sub-daily precipitation is preferable. Moreover, bias correction of sub-daily variables from CPMs is rare, largely due to two main challenges: (1) the substantial computational and memory demands associated with processing sub-daily data from high-resolution climate models, and (2) the relative scarcity of hourly observational datasets compared to daily observations, which poses challenges for the evaluation and correction of sub-daily variables from CPMs.

This study leverages a 1 km resolution gridded hourly observation-based rainfall dataset (CEH-GEAR1hr) for bias correcting the UKCP18-CPM hourly precipitation data. The CEH-GEAR1hr dataset (Lewis et al., 2022) integrates data from over 1900 quality-controlled rainfall gauges over Great Britain, providing highly accurate precipitation measurements and enabling more robust sub-daily evaluation and bias correction of precipitation timing and intensity. For temperature, there is currently no observation-based hourly temperature dataset across the UK. Therefore, bias correction was performed on the UKCP18-CPM daily temperature using the Met Office's 1 km daily HadUK-Grid dataset (Hollis et al., 2019; Met Office et al., 2022). Building on the findings of Faghih et al. (2022), we address a remaining gap: diurnal-cycle-aware bias correction is still rarely implemented and systematically evaluated for sub-daily precipitation in convection-permitting climate simulations, despite its importance for impact models that are sensitive to rainfall timing and intensity. The contribution of this paper is therefore application-driven evaluation in a CPM sub-daily setting, providing evidence on whether incorporating diurnal structure within quantile mapping improves (i) conventional climatological statistics and (ii) more challenging characteristics, including the diurnal cycle and sub-daily extremes.

The purpose of this paper is to present and assess a bias-correction method for UKCP18-CPM hourly precipitation and to quantify how closely the bias-adjusted UKCP18-CPM simulations align with observations relative to the raw simulations. We bias-corrected and evaluated the hourly precipitation and daily temperature variables from the UKCP18-CPM for England using the quantile mapping (QM) method. The UKCP18-CPM (Kendon et al., 2019b) consists of a 12-member convection-permitting model ensemble, which is nested within a 12-member RCM perturbed parameter ensemble (PPE), and further nested within a 12-member GCM PPE. Due to storage and computational constraints, we referenced the CHESS-SCAPE dataset (Robinson et al., 2023) and selected four ensemble members for this study (Sect. 2.3.1 Choice of sub-ensemble), including the default configuration EM01 (also the driest member), the wettest members (EM04 and EM07), and a moderate member (EM08). The bias-corrected dataset features a high spatial resolution of 1 km, with hourly precipitation and daily temperature data. Specifically, we address the following research questions:

What biases are contained in the UKCP18-CPM simulations?

Addressing these questions is crucial for understanding and improving the applicability of UKCP18-CPM data in climate impact assessments.

This study provides a practical and reproducible bias correction approach for UKCP18-CPM hourly precipitation and daily mean temperature over England and evaluates how the adjusted outputs compare with observational reference datasets across mean behaviour, diurnal characteristics, and extremes. A set of bias-corrected 1 km UKCP18-CPM temperature and precipitation projections for England is provided, using empirical quantile mapping (QM) for daily mean temperature and diurnal bias correction (DBC) for hourly precipitation. The methodological contribution is not the development of a new bias-correction method, but the transparent implementation and assessment of established quantile-mapping methods for a large CPM dataset at hourly resolution. This is relevant for impact applications where modelled rainfall timing and intensity can strongly influence simulated responses, especially for fast-response impact applications (e.g., short-duration rainfall-driven flooding). These results add evidence on the value of diurnal-cycle bias correction for sub-daily precipitation in convection-permitting simulations, a setting where such evaluations remain relatively limited.

We applied empirical QM to daily mean temperature to address seasonal biases and distributional differences using a reproducible procedure. It is widely used because it can correct not only mean biases but also biases in variability and quantiles (Fang et al., 2015; Themeßl et al., 2011; Wilcke et al., 2013). For hourly precipitation, we used DBC by hour of day, because applying a single mapping across all hours can overlook systematic diurnal-cycle biases in sub-daily precipitation. Figure 5 supports this choice, showing that the bias-corrected ensemble more closely reproduces the observed hour-of-day pattern. This hour-of-day treatment is intended to reduce discrepancies that matter for sub-daily impact applications, consistent with the broader argument that correcting the diurnal cycle can improve sub-daily bias-correction strategies (Faghih et al., 2022). Therefore, we recommend using DBC for sub-daily variables. However, in this study we applied a 3 h moving window in the DBC. By incorporating hourly data from the previous and following hours, this may have contributed to some of the remaining discrepancies. It may be worth exploring whether omitting the moving window gives better results when the training period sample size is sufficiently large.

Before bias correction, the UKCP18-CPM simulations show considerable biases in both precipitation and temperature across various temporal and spatial scales. In general, UKCP18-CPM simulations show wet precipitation biases (especially in winter) and cool biases in temperature. These broad patterns are consistent with previous evaluations of UKCP18-RCM simulations (Reyniers et al., 2025) and the UKCP18-CPM science report (Kendon et al., 2019b). These residual biases are consistent with the broader CPM literature, which shows that although kilometre-scale models can improve the representation of precipitation compared with convection-parameterised RCMs, systematic biases and model uncertainty may remain, particularly for sub-daily precipitation and across different regional settings (Ban et al., 2021; Correa-Sánchez et al., 2025; Soares et al., 2024). After bias correction, both precipitation and temperature show substantially improved agreement with observations at monthly and diurnal timescales (Figs. 4 and 5). This suggests that bias correction provides a more reliable basis for downstream impact assessments than raw UKCP18-CPM output.

Future changes were also analysed for both precipitation and temperature in Figs. 9–12. In general, the bias-corrected projections indicate a clear intensification of annual maximum 1 h precipitation (Rx1h) by both the 2030s and 2070s, while changes in annual total precipitation (Ptot) remain smaller and spatially variable (Figs. 9 and 12). This pattern (i.e., stronger changes in short-duration extremes than in totals) is broadly consistent with CPM-based studies that report amplified changes for sub-daily extremes and a strong dependence on dynamical factors and regional setting (Dallan et al., 2024; Pichelli et al., 2021). The projected temperature changes are consistent with findings from the UKCP18-CPM science report (Kendon et al., 2019b), which indicated that mean temperature is expected to increase across all regions and seasons. Our findings also align well with existing literature, such as Robinson et al. (2023), who reported similar projected changes for UKCP18-RCM. This suggests that the bias-corrected projections remain broadly consistent with previously reported climate change signals.

Due to limitations in computational resources and time, we focused our analysis on four ensemble members: EM01, EM04, EM07, and EM08. These were selected to represent a diverse range of climate outcomes, from the driest to the wettest scenarios, allowing the bias-corrected dataset to effectively capture the range of possible climate responses in England. This sub-ensemble was chosen to balance computational efficiency with representativeness while ensuring that the selection captures a wide range of precipitation responses. The reduced number of ensemble members may influence the ensemble mean and spread compared to using all 12 members. Future studies could address this by expanding the bias correction to include more ensemble members and broader spatial coverage. Additionally, the integration of multivariate bias correction methods (Cannon, 2018; Faghih et al., 2022) could offer the advantage of preserving inter-variable dependencies, ensuring that precipitation and temperature are corrected consistently without disrupting their natural relationship.

This high-resolution bias-corrected dataset offers enhanced reliability, making it suitable for a wide range of climate change impact simulations and studies. The dataset's fine spatial and temporal resolution enables its application in various fields, such as flood forecasting, agricultural planning, and natural resource management, providing valuable insights for decision-making and long-term adaptation strategies. Future research could extend this analysis to other regions and incorporate additional climate models, providing a broader understanding of the impacts of climate change and helping to assess the applicability of these methods in different contexts. This would not only improve the robustness of climate projections but also enhance the relevance and utility of bias-corrected datasets in informing climate adaptation strategies.

In this study, we applied bias correction to the UKCP18 Convection-permitting model (CPM) data to produce 1 km high-resolution precipitation and temperature projections for England, UK. Using quantile mapping (QM) as our bias correction method, we corrected biases in hourly precipitation and daily temperature data for four ensemble members (EM01, EM04, EM07, and EM08) selected to represent the spread of the full ensemble. Our results demonstrate that bias correction improved the alignment of UKCP18-CPM simulations with observational datasets, effectively reducing discrepancies (biases) and enhancing the model's performance across various metrics. The bias-corrected dataset provides a more reliable foundation for assessing future climatic changes and regional impacts across England. Although the bias-correction approaches applied here are established, this study provides new evidence of their performance for sub-daily precipitation in convection-permitting UK climate projections, including improvements in diurnal-cycle behaviour and sub-daily extremes relevant to impact modelling. The key findings are as follows:

Raw UKCP18-CPM simulations showed consistent wet biases for precipitation and cooler biases for temperature. For precipitation, the annual average biases ranged from 4.6 % (EM07) to 18.3 % (EM04), with the largest biases occurring in winter (DJF) ranging from 20.4 % (EM01) to 43.5 % (EM08). Temperature biases show a clear seasonal dependence, with winter (DJF) exhibiting larger spatial contrasts and summer (JJA) showing a more consistent cool bias across all members. On an annual mean basis, the temperature bias ranged from -0.87 °C (EM04) to +0.02 °C (EM08), with three of the four members showing a cool bias.

After bias correction, both precipitation and temperature simulations aligned well with observational data. Monthly precipitation and temperature biases were substantially reduced, with bias corrected outputs closely following observed monthly patterns. The diurnal cycle of bias corrected precipitation captured the variability and magnitude of observed precipitation across most hours of the day. The 95th percentile (P95) of hourly precipitation and temperature extremes (T95) showed markedly improved agreement with observations after bias correction for both mean behaviour and high-percentile extremes (P95 and T95).

Future projections for the 2030s and 2070s show significant increases in both precipitation and temperature. By the 2030s, annual maximum 1 h precipitation (Rx1h) is projected to increase by 19.9 % (EM04) to 23.9 % (EM08), while annual total precipitation (Ptot) shows more moderate increases, ranging from 3 % (EM08) to 9.4 % (EM01). In terms of temperature, annual mean temperature (Tmean) increases range from 1.2 °C (EM07) to 1.7 °C (EM08) by the 2030s. By the 2070s, these trends become more pronounced, with Rx1h increasing by 25.1 % (EM07) to 39.1 % (EM04), and Tmean projected to rise by 3.1 °C (EM07) to 4.5 °C (EM04). These projections indicate a potential for more intense extreme weather events, particularly for precipitation and temperature extremes, as the century progresses.