Journal Article

Causal Discovery Analysis Reveals Global Sources of Predictability for Regional Flash Droughts

Tian, Di, Kumar, Sudhanshu

2024

Overview

Detecting and quantifying the global teleconnections with flash droughts (FDs) and understanding their causal relationships is crucial to improve their predictability. This study employs causal effect networks (CENs) to explore the global predictability sources of subseasonal soil moisture FDs in three regions of the United States (US): upper Mississippi, South Atlantic Gulf (SAG), and upper and lower Colorado river basins. We analyzed the causal relationships of FD events with global 2‐m air temperature, sea surface temperature, water deficit (precipitation minus evaporation), and geopotential height at 500 hPa at the weekly timescale over the warm season (April to September) from 1982 to 2018. CENs revealed that the Indian Ocean Dipole, Pacific North Atlantic patterns, Bermuda high‐pressure system, and teleconnection patterns via Rossby wave train and jet streams strongly influence FDs in these regions. Moreover, a strong link from South America suggests that atmospheric circulation forcings could affect the SAG through the low‐level atmospheric flow, reducing inland moisture transport, and leading to a precipitation deficit. Machine learning utilizing the identified causal regions and factors can well predict major FD events up to 4 weeks in advance, providing useful insights for improved subseasonal forecasting and early warnings. Plain Language Summary This study investigates the global factors affecting flash droughts (FDs) in the upper Mississippi, South Atlantic Gulf, and upper and lower Colorado river basins in the United States. Using causal effect networks, the research identifies key global influences, such as the Indian Ocean Dipole, Pacific North Atlantic patterns, and Bermuda high‐pressure system, which affect FDs through atmospheric circulation and jet streams. Findings show that these factors, along with local and remote climate processes, can predict FDs up to 4 weeks in advance. Machine learning models utilizing these climate variables effectively forecast FDs events. The study highlights the importance of large‐scale climate oscillations and teleconnections in predicting FDs, offering useful insights for improving early warnings and climate risk management. Key Points Causal discovery framework identified local and remote regions over the globe causing U.S. Flash droughts (FDs) Machine learning with climate variables over the discovered regions can well predict U.S. FDs up to 4 weeks in advance The discovered regions are consistent with the known sources of predictability but also reveal potentially new sources of predictability

• Publisher: John Wiley & Sons, Inc,Wiley
• Subject: Air temperature / Atmospheric circulation / Bermuda / causal discovery analysis / Climate / Climate change / Climate models / Climate prediction / Colorado River / Dipoles / Drought / Dynamic height / Environmental risk / Evaporation / flash drought / Geopotential / Geopotential height / Indian Ocean / Jet stream / Jet streams (meteorology) / Learning algorithms / Machine learning / Mississippi / Moisture content / moisture diffusivity / Ocean basins / Oceans / Oscillations / Planetary waves / Precipitation / Regional analysis / Risk management / River basins / Rivers / Rossby waves / Sea surface temperature / Soil moisture / Soil temperature / soil water / South America / surface water temperature / teleconnection / Teleconnection patterns / Teleconnections / warm season / Warm seasons / Water deficit / Wave trains / Weather forecasting