Global-scale spatiotemporal clustering of multivariate drought events using 3D DBSCAN

Drought is one of the most significant natural hazards, impacting ecosystems, water resources, and human livelihoods worldwide. Traditional drought analysis often focuses on specific types or limited geographical regions, leaving a critical gap in understanding the global evolution and interconnection of drought events across different timescales and dimensions.
This study aims to address this gap by employing DBSCAN (Density-Based Spatial Clustering of Applications with Noise, e.g., Camalieri and Toreti, 2023) algorithm to identify, and quantify diverse characteristics of meteorological, hydrological, and agricultural droughts on a global scale. Specifically, we focus on the sensitivity of the DBCAN parameters, which are crucial for distinguishing meaningful drought clusters from noise in large, complex datasets. Our objective is to develop and validate a robust framework for detecting and assessing the spatiotemporal evolution of drought in different compartments of hydrological cycle, enabling a more comprehensive evaluation of entire drought dynamics.
Using a global hydrological dataset forced with ERA5 meteorologic dataset (Řehoř et al, 2024), we implement a 3D DBSCAN method, integrating spatial and temporal dimensions. The dataset provides key outputs of a hydrological model, including soil moisture, precipitation, potential evapotranspiration, and discharge, which are used to calculate drought metrics and identify large clusters with a total area exceeding 150,000 km² and lasting at least 30 days. At this stage, we work with historical data from 1980 to 2022, providing a robust platform to assess spatiotemporal drought patterns. This historical dataset will serve as a foundation for a future comparison with projected climate scenarios from 2025 to the end of the 21st century, enabling insights into potential changes in drought characteristics.
Our findings reveal that 3D DBSCAN is highly effective in capturing the spatiotemporal evolution of drought events, with parameter sensitivity playing a pivotal role in cluster detection. Small adjustments of algorithm’s inputs significantly influence the size, shape, and distribution of clusters, highlighting the need for careful calibration. This framework provides new insights into the relationships between drought events across regions and temporal scales, highlighting their potential to inform water resource management and climate adaptation strategies.

Cammalleri, C. and Toreti, A., 2023. A generalized density-based algorithm for the spatiotemporal tracking of drought events. Journal of Hydrometeorology, 24(3), pp.537-548.
Řehoř, J., Brázdil, R., Rakovec, O., Hanel, M., Fischer, M., Kumar, R., Balek, J., Poděbradská, M., Moravec, V., Samaniego, L. and Trnka, M., 2024. Global catalog of soil moisture droughts over the past four decades. EGUsphere, 2024, pp.1-34.

We acknowledge the Czech Science Foundation grant 23-08056S.