HYADES - A Global Archive of Annual Maxima Daily Precipitation

Abstract

Time series of annual maxima daily precipitation are crucial for understanding extreme precipitation behavior and its shifts toward nonstationarity with global warming. Extreme precipitation insight assists hydraulic infrastructure design, water resource management, natural hazard prevention, and climate change adaptation. However, not even a third of the records are of sufficient length, and the number of active stations keeps decreasing. Herein, we present HYADES: archive of yearly maxima of daily precipitation records, a global dataset derived from the Global Historical Climatology Network database of daily records (GHCN-Daily). The HYADES dataset contains records from 39 206 stations (heterogeneously distributed worldwide) with record lengths varying from 16 to 200 years between 1805 and 2023. HYADES was extracted through a methodology designed to accurately capture the true maxima even in the presence of missing values within the records. The method's thresholds were determined and evaluated through Monte Carlo simulations. Our approach demonstrates a 96.73% success rate in detecting the true maxima while preserving time series statistical properties of interest (L-moments and temporal monotonic trend).

Fig. 1

Time series of the number of active ground stations with total daily precipitation data as extracted from the Global Historical Climatology Network. The dashed lined denotes the year 2001 when the number of active ground stations had a trend inversion that lasted for a decade before reverting to an overall decreasing trend.

Fig. 2

Spatial distribution of 43 585 ground stations with daily precipitation records of at least 20 years on a 2.5° × 2.5° regular grid. (a) Average record length of ground stations. (b) Number of ground stations per grid cell.

Fig. 3

Missing value profile of the 43 585 selected stations. (a) Distribution of the record length in years. (b) Percentage of missing values in the records. (c) Percentage of years with missing values. (d) Number of years with missing values. (e) Number of missing days within incomplete years. (f) Number of moths with missing values within incomplete years. (g) Number of missing days within incomplete months. (h) Missing value structure within incomplete years.

Fig. 4

Monte Carlo simulation scheme. (a) extraction of all available complete (0% missing values and 0% quality flags) records from the GHCN-Daily database and generate a time series of annual maxima. (b) extraction of binary missing value profiles from the rest of the records. (c) generation of randomly altered records using the binary profiles to introduce missing values in the complete records. (d) generation of time series of annual maxima from the artificially created records for multiple threshold criteria values. (e) evaluation of the fidelity of the artificially generated annual maxima time series generated in (d) with the real maxima series created in (a).

Fig. 5

Explanatory plot of the maxima extraction method. The annual maxima of daily precipitation is extracted regardless of the quantity of missing values or quality flags as long as the record length is at least 20 years. Subsequently, the annual maxima is considered invalid (red crosses in (c)) if its probability of exceedance (PE%) is above 60% (red dashed line in (a)) and the missing-value percentage (MV%) of the year it belongs to is larger than 33% (red dashed line in (b)).

Fig. 6

Scatter plot summary results of the 100 000 Monte Carlo simulations with artificially generated time series in the y-axis and the real time series in the x-axis. In addition, error histograms with the percentage of points in the y-axis and the error (artificially generated time series minus the real time series) in the x-axis. (a) Annual maxima of daily precipitation. (b) Mean of the annual maxima of daily precipitation. (c) L-Variation of the annual maxima of daily precipitation. (d) L-Skewness of the annual maxima of daily precipitation. (e) L-Kurtosis of the annual maxima of daily precipitation. (f) Trend of the annual maxima of daily precipitation.

Fig. 7

L-moments ratio diagram for the 39 206 HYADES records and the theoretical line of the Generalized Extreme Value distribution. L-Kurtosis versus L-Skewness points in matte yellow. Average point in red.