A Data-driven Diagnostic Framework for Small Watersheds based on Markov Chain-Monte Carlo

Understanding dynamics of hydrological properties is essential in producing skillful runoff forecast. This can be quantitatively done by tracking changes in parameters of hydrology models that represent physical characteristics. In this study, we implemented a Bayesian estimation method for small watersheds in continuously estimating hydrology model parameters given observations of precipitation and runoff. The method was coupled with a conceptual hydrology model of Instantaneous Unit Hydrograph model based on a modified Gamma distribution. The whole diagnostic framework was tested using simulated data as well as observational data from the Fall Creek watershed. Both analyses showed good consistency between Bayesian parameter estimations and true values or maximum likelihood estimations. Also for the case study using observational data, a systematic shift in local precipitation-runoff response was observed in 1943, which could not be learned by looking at times series of precipitation, runoff, and runoff coefficients. Our results demonstrated potential of the Bayesian estimation method in monitoring hydrological dynamics and rapidly detecting changes in hidden physical processes for small watersheds.