Completion of the Central Italy daily precipitation instrumental data series from 1951 to 2019

This method combines the outcome of 15 strategies based on four various gap‐filling techniques (quantile mapping, spatial interpolation, machine learning, and multi‐strategy merging). These strategies employ the daily dataset of the neighbouring stations and the matched ERA5 data to estimate missing values at the target stations. Both raw data and the final serially complete station datasets (SCDs) underwent comprehensive quality control.

Many accuracy indicators have been utilized to evaluate the performance of the strategies' estimations and the final SCD, such as Correlation Coefficient (CC), Root mean square error (RMSE), Relative bias (Bias %), and Kling‐Gupta efficiency (KGE″). Multi‐strategy merging strategy based on the Modified Kling‐Gupta efficiency (MS 1 ) shows the highest performance as an individual precipitation gap‐filling strategy. However, the machine learning strategy using random forest (ML 3 ) has the most outstanding share in the final estimates among all other strategies.

In the end, the temporal–spatial performance of the final SCD is promising and depends on the pattern of the missing values (MV%). The mean values of KGE″, CC, variability ( α ), and bias term ( β ) are 0.9, 0.93, 1.064, and 4.98 × 10 −7 , respectively.