Global-scale massive feature extraction from monthly hydroclimatic time series: Statistical characterizations,...

Area of Science:

• Earth and Environmental Sciences
• Data Science
• Hydrology

Background:

• Traditional hydroclimatic time series analysis uses limited features, capturing only a fraction of observational information.
• This limits the reliability of analyses, particularly in complex applications like time series clustering.

Purpose of the Study:

• To develop and apply a novel big data framework for comprehensive hydroclimatic time series analysis through massive feature extraction.
• To enhance the reliability and depth of insights derived from hydroclimatic data.
• To characterize global hydroclimatic variables and identify spatial patterns.

Main Methods:

• Developed a big data framework for automatic, massive feature extraction (approx. 60 features) from hydroclimatic time series.
• Applied the framework to global datasets of temperature, precipitation, and river flow (40 years, >13,000 stations).
• Utilized Breiman's random forests for a new hydroclimatic time series clustering methodology.

Main Results:

• Extracted interpretable knowledge on seasonality, trends, autocorrelation, long-range dependence, and entropy.
• Identified global patterns in feature variability and spatial characteristics across different hydroclimatic variables.
• Developed spatially coherent clusters of hydroclimatic time series, demonstrating the methodology's utility.

Conclusions:

• The massive feature extraction framework significantly enhances understanding of hydroclimatic variables compared to traditional methods.
• The proposed clustering methodology, based on random forests, yields spatially coherent and reliable results.
• The scale-independent features and spatially coherent clusters offer potential benefits for regionalization, forecasting, and simulation in hydrology.