Using antecedent soil moisture information for accurate prediction of subseasonal-to-seasonal hydroclimate variability over South America

Abstract:

This work uses remotely sensed and reanalysis data to demonstrate a proof-of-concept statistical model for skillful prediction of intraseasonal (30-75 day bandpass-filtered) hydroclimate anomalies over South America using antecedent soil moisture information, with a lead time of 30 days. Using extended empirical orthogonal function (EEOF) analysis, we establish a link between the dominant “see-saw” mode of intraseasonal hydroclimate variability (EEOF-1) and antecedent soil moisture anomalies. Composites of antecedent, filtered, land surface anomalies, between -35 day and -5 day of peak activity of EEOF-1 show the evolution of dry (wet) soil moisture anomalies underlying positive (negative) precipitation anomalies, in both phases of EEOF-1. We propose a continental scale surface-atmosphere feedback mechanism, as persistence of soil moisture anomalies in the root-zone can influence the local climate through modulation of surface heat fluxes. Results for terrestrial coupling metrics re-affirm the enhanced coupling between soil moisture and evapotranspiration in this antecedent period. Finally, we test the suitability of soil moisture information as a predictor for the dominant hydroclimate mode using a 4-layer artificial neural network, along with other previously known predictors such as El Nino Southern Oscillation (ENSO) and Madden Julian Oscillation (MJO), with a lead time of 30 days. Our model predicts the hydroclimate timeseries with >95% accuracy, and further tests suggest soil moisture as the most important predictor. Improved prediction of hydroclimate with a lead time of 30 days has immense agricultural and related benefits for the socio-economically crucial regions of South Atlantic Convergence Zone and Southeastern South America.