Confounding factors in determining causal soil moisture-precipitation feedback

Samuel E. Tuttle, Guido D. Salvucci

Research output: Contribution to journalArticlepeer-review

38 Scopus citations


Identification of causal links in the land-atmosphere system is important for construction and testing of land surface and general circulation models. However, the land and atmosphere are highly coupled and linked by a vast number of complex, interdependent processes. Statistical methods, such as Granger causality, can help to identify feedbacks from observational data, independent of the different parameterizations of physical processes and spatiotemporal resolution effects that influence feedbacks in models. However, statistical causal identification methods can easily be misapplied, leading to erroneous conclusions about feedback strength and sign. Here, we discuss three factors that must be accounted for in determination of causal soil moisture-precipitation feedback in observations and model output: seasonal and interannual variability, precipitation persistence, and endogeneity. The effect of neglecting these factors is demonstrated in simulated and observational data. The results show that long-timescale variability and precipitation persistence can have a substantial effect on detected soil moisture-precipitation feedback strength, while endogeneity has a smaller effect that is often masked by measurement error and thus is more likely to be an issue when analyzing model data or highly accurate observational data.

Original languageEnglish (US)
Pages (from-to)5531-5544
Number of pages14
JournalWater Resources Research
Issue number7
StatePublished - Jul 2017
Externally publishedYes


  • Granger causality
  • confounding factors
  • feedback
  • land-atmosphere interactions
  • precipitation
  • soil moisture

ASJC Scopus subject areas

  • Water Science and Technology


Dive into the research topics of 'Confounding factors in determining causal soil moisture-precipitation feedback'. Together they form a unique fingerprint.

Cite this