Advancing Global Drought Monitoring with Novel High-Resolution Datasets of Soil Moisture and Flux Dynamics
Qianqian Han is a PhD student in the Department of Water Resources. (Co)Promotors are prof.dr. Z. Su and dr. Y. Zeng from the Faculty ITC.
Drought is one of the most devastating natural disasters, significantly impacting agriculture, ecosystems, and water resources. It is both a consequence of climate change and a major driver of global environmental and socio-economic instability. Effective drought monitoring is crucial for early warning systems, emergency responses, and policy-making, particularly in drought-prone regions. Soil moisture and land surface fluxes, such as evapotranspiration (ET), play a central role in assessing drought conditions. Soil moisture, in particular, is a key variable in land-atmosphere interactions, regulating surface energy balance, runoff, and plant water uptake. Meanwhile, flux such as ET provide insight into the water cycle by reflecting the balance between soil water availability and atmospheric demand. Given their critical role in drought dynamics, accurate and high-resolution soil moisture and flux datasets are essential for comprehensive drought analysis; however, existing global datasets lack the necessary spatial and temporal resolution to capture high-frequency variations in soil moisture and fluxes.
This dissertation aims to improve global drought monitoring by employing physics-guided machine learning (ML) to generate high-resolution soil moisture and land surface flux datasets. The research integrates in situ measurements, remote sensing, meteorological data, and physical process understanding to achieve this goal.
As a foundational step, an ensemble of optimized ML algorithms was developed to predict surface soil moisture (SSM) across climate zones of the globe. By systematically integrating diverse data sources, this study assessed the performance of multiple ML algorithms, demonstrating that ensemble approaches significantly enhance prediction accuracy across different climate zones. The results highlight the capability of ML-based methods in generating robust global soil moisture datasets for hydrological and agricultural applications.
Building on this, a physics-informed ML framework was implemented to generate a global, long-term (2000–2020), high-resolution (1 km, daily) soil moisture dataset. By incorporating physical constraints alongside ML techniques, this approach ensured more realistic and consistent soil moisture estimates. The dataset was extensively validated against independent observations and demonstrated strong agreement with existing soil moisture products, supporting its use in climate trend analysis and extreme event detection.
Leveraging the high-resolution soil moisture dataset we generated, a global, long-term, high-resolution dataset of terrestrial water-energy-carbon fluxes was developed by integrating process-based land surface modeling, ML, and optimal interpolation techniques. The dataset provides hourly 9 km estimates of key flux components, including net radiation, latent heat flux, sensible heat flux, soil heat flux, gross primary productivity, and solar-induced fluorescence at 685 nm and 740 nm wavelengths. It demonstrated that incorporating both physical modeling and ML significantly improves prediction accuracy, with key environmental drivers such as radiation, soil moisture, and vegetation properties shaping the model performance. This dataset enables critical insights into land-atmosphere interactions and ecosystem responses to climate variability.
Finally, these high-resolution datasets were applied to a comprehensive global drought analysis. By jointly employing cumulative precipitation deficit Σ(ET–P), the Standardized Soil Moisture Index (SSMI), and the GRACE-based Drought Severity Index (GRACE-DSI), this study revealed four key findings. (1) Global prevalence and intensification: near-normal to abnormal droughts (5–25-year return periods) were globally widespread, and both drought extent and severity intensified significantly after 2014. (2) Severe hotspots: extreme droughts (≥100-year return periods) emerged in localized hotspots including high-latitude areas, semi-arid regions, and unexpectedly humid zones previously not regarded as drought-prone. (3) Frequency and duration shifts: moderate drought frequency peaks shifted toward fewer annual drought days, while the most extreme events increased in frequency. Drought duration rose with severity, exceeding 100–120 days in hotspots such as central Eurasia, eastern Russia, and southeastern South America. (4) Ecosystem vulnerabilities: case studies of the 2018 Netherlands drought and the 2015–2016 Amazon drought highlighted ecosystem-specific responses. GPP analysis confirmed that energy limitation dominated high-latitude impacts, while pronounced water limitation prevailed in drylands. Land energy–water partitioning further emphasized climate-driven disparities, consistent with global hydroclimatic regimes.
Overall, this dissertation developed a novel framework to advance global drought monitoring. By capturing the spatial heterogeneity of drought drivers and ecosystem responses, this work provides a more reliable foundation for climate risk assessment, adaptation strategies, and sustainable water resource management in a warming climate.
