| 21 Sep 2026
PRECISi: A High-Resolution Daily Gridded Precipitation Dataset for Sicily (1951–2022) Derived from a Doubly Conditional Geostatistical Framework
Abstract. Long-term, high-resolution precipitation datasets are indispensable for understanding hydroclimatic variability and supporting water-resource and climate-impact studies. Yet, the reconstruction of spatially consistent rainfall fields over extended historical periods remains a major challenge, particularly in data-sparse Mediterranean regions characterized by complex terrain and pronounced precipitation heterogeneity. This study introduces the first long-term, high-resolution gridded precipitation dataset for Sicily, delivering a continuous historical reconstruction since 1951 from sparse gauge observations through a novel two-phase, doubly conditional spatial interpolation framework. Calibration was performed on a dense modern network of automated rain gauges; a Transfer-Informed Modelling strategy, whereby all geostatistical parameters calibrated on this high-density contemporary period are transferred without modification to epochs of substantially lower network density, ensures temporal coherence across the full historical record. A merged archive of historical stations was assembled from two independent networks following rigorous homogeneity assessment of inter-network spatial coherence prior to reconstruction. Three principal methodological innovations distinguish the framework: (i) a double classification of daily events by intermittency and magnitude into distinct hydrometeorological regimes having unique spatial correlation structures; (ii) a comparison of regime-specific intensity-modelling paradigms based on geostatistics or regression; and (iii) a contrast of binary masking against probabilistic hurdle-weighting for occurrence-conditional intensity. Leave-one-out cross-validation demonstrated high rainfall detection capability across all intermittency classes and established the consistent superiority of the geostatistical framework over its regression-kriging counterpart across all hydrometeorological regimes. The superior optimal framework was consequently selected for operational historical reconstruction. Validated against an independent monthly climatological benchmark over the full study period, the bias-corrected primary dataset reproduces the reference climatology with high fidelity, affirming its fitness for long-term hydroclimatic analysis across topographically complex Mediterranean terrain (data is available at https://doi.org/10.5281/zenodo.19228150; BEIKAHMADI et al., 2026).
I thank the authors for presenting this long-term, high-resolution daily precipitation dataset for Sicily. The dataset has clear potential value, and the manuscript includes extensive methodological development and validation. However, I have concerns regarding the robustness of the historical reconstruction and the uncertainty of the final product.
1. A key assumption is that the spatial dependence structures derived from the recent dense station network can be transferred to the much sparser historical period. The current sensitivity analyses are useful, but they do not directly show how reconstruction skill changes as station density decreases. I suggest adding a network-thinning experiment using the modern observations, with reconstruction performance evaluated against withheld stations. This would provide more direct support for the historical reconstruction.
2. The monthly benchmark provides useful evidence of consistency, but it is not fully independent because it relies on much of the same underlying observational archive. The manuscript should clarify this point and avoid overstating the independence of this validation. If possible, recent stations not involved in the reconstruction should be used for additional out-of-sample validation.
3. The uncertainty characterization remains limited for an ESSD data product. Reconstruction uncertainty is likely to vary substantially with station density, precipitation regime, topography, and historical period. The manuscript should provide more quantitative guidance on where and when the reconstruction is less reliable. Even a practical uncertainty or quality indicator linked to station density and reconstruction skill would considerably improve the usability of the dataset.
Overall, I consider the dataset potentially valuable. However, stronger evidence for the transferability of the modern reconstruction framework to the historical sparse network, together with clearer uncertainty information, would be necessary to support its broad use as a long-term precipitation product.