A distinction is made between data rescue (i.e., copying, digitizing and archiving) and data recovery that implies deciphering, interpreting and transforming early instrumental readings and their metadata to obtain high-quality datasets in modern units. This requires a multidisciplinary approach that includes: palaeography and knowledge of Latin and other languages to read the handwritten logs and additional documents; history of science to interpret the original text, data e metadata within the cultural frame of the 17th, 18th and early 19th century; physics and technology to recognize bias of early instruments or calibrations, or to correct for observational bias; astronomy to calculate and transform the original time in canonical hours that started from twilight. The liquid-in-glass thermometer was invented in 1641 and the earliest temperature records started in 1654. Since then, different types of thermometers were invented, based on the thermal expansion of air or selected thermometric liquids with deviation from linearity. Reference points, thermometric scales, calibration methodologies were not comparable, and not always adequately described. Thermometers had various locations and exposures, e.g., indoor, outdoor, on windows, gardens or roofs, facing different directions. Readings were made only one or a few times a day, not necessarily respecting a precise time schedule: this bias is analysed for the most popular combinations of reading times. The time was based on sundials and local Sun, but the hours were counted starting from twilight. In 1789-90 Italy changed system and all cities counted hours from their lower culmination (i.e., local midnight), so that every city had its local time; in 1866, all the Italian cities followed the local time of Rome; in 1893, the whole Italy adopted the present-day system, based on the Coordinated Universal Time and the time zones. In 1873, when the International Meteorological Committee (IMO) was founded, later transformed in World Meteorological Organization (WMO), a standardization of instruments and observational protocols was established, and all data became fully comparable. In the early instrumental period, from 1654 to 1873, the comparison, correction and homogenization of records is quite difficult, mainly because of the scarcity or even absence of metadata. This paper deals about this confused situation, discussing the main problems, but also the methodologies to recognize missing metadata, distinguish indoor from outdoor readings; correct and transform early datasets in unknown or arbitrary units into modern units; finally, in which cases it is possible to reach the quality level required by WMO. The focus is to explain the methodology needed to recover early instrumental records, i.e., the operations that should be performed to interpret, correct, and transform the original raw data into a high-quality dataset of temperature, usable for climate studies.

Meteorological observations over the last four decades are of paramount importance to investigate the ongoing climate change. The assessment of the reliability of any climatic time series is thus mandatory to draw correct conclusions. This evaluation involves homogeneity tests to detect artificial discontinuities whose identification is facilitated by metadata availability. In this work, daily minimum and maximum temperature measurements collected in Padua, Italy, between 1980 and 2022, are examined. Hourly observations began in 1993 and since the aim is to study long term behavior of the temperature, the focus is on daily averages and extremes. Over this period, the weather station of Padua center underwent many changes, in location or instrument; therefore, some tests have been used to identify and remove the effects of these variations and obtain a homogeneous time series. The homogeneity tests applied must be able to identify change-points both in the middle and at the extremes of the series. Some well-known absolute tests have been applied to investigate shift in the mean value: Standard Normal Homogeneity test (SNH), Buishand U and range tests, Pettitt test, F-test, STARS. Some relative tests have been applied too, which are generally more reliable than absolute tests, because they consider the information from neighboring stations. As relative tests rely on the homogeneity and quality of the reference series, several nearby stations and two reanalysis datasets (ERA5 and MERIDA) have been considered, to enhance the picture of the local situation and provide more robust conclusions. The applied tests identify change-points in the years in which a change in instrument or location of the station have occurred, confirming that these changes have compromised the homogeneity of the series. The sub-series obtained splitting the observations in correspondence of these change-points have been homogenized with respect a selected period; corrections must be applied also to future measurements to extend the time series properly.

The study of precipitation amount at different time scales constitutes an important issue in climate research and risk assessment. When dealing with daily totals, a frequent but sometimes underestimated problem is at what time the observation day begins. The choice of different starting times may lead to incompatibility between stations and incorrect identification of extreme events. In this work, the problem of temporal misalignment between precipitation datasets characterized by different starting time of the observation day is analyzed. The most widely used adjustment methods (1day and uniform shift) and two methods based on reanalysis (NOAA and ERA5) are evaluated in terms of temporal alignment, precipitation statistics and percentile distributions. As test series, the precipitation amount collected from 9 a.m. local time (09 LT) on the previous day to 09 LT on the target day (9-9 datasets) of the Padua and nearby stations in the period 1993-2022 have been selected. Results show that the reanalysis-based methods, in particular ERA5, outperform the others in temporal alignment, regardless the station. But, for the periods in which reanalysis data are not available, “1day” method, which shifts the daily amount back one calendar day, and “unif” method, which distributes uniformly the daily total from a 2-day moving window surrounding the target date, can be considered valid alternatives. On the other side, concerning the precipitation statistics, the reanalysis-based methods are not the best option, as they increase the precipitation frequency and reduce the mean value over wet days, NOAA much more than ERA5. Nevertheless, the uniform method provides a larger deviation from the original daily series. The use of the series of a station nearby the target one, which is mandatory in case of missing data, gives similar or better results than applying any adjustment method to the 9-9 series. General conclusions can hardly be drawn as they depend on the method and station. For the Padua dataset, the analysis was repeated at monthly and seasonal resolution. In general, the adjustment series show the most relevant changes in the precipitation statistics in summer and less temporal alignment with the original series in summer and autumn, the two seasons mainly affected by heavy rains in Padua. Finally, the percentiles distribution, analyzed for all the methods and stations, indicates that any adjustment method underestimates the percentile values, except ERA5. Only Legnaro, the station most correlated with Padua, gives results like ERA5.