![IconGems-Print](images/IconGems-Print.png)
264 – Contributed Oral Poster Presentations: Section on Statistics and the Environment
Bayesian and Transfer Function Estimation of a Tobit State-Space Model for Daily Precipitation Data
Sai Kumar Popuri
University of Maryland, Baltimore County
Nagaraj K. Neerchal
University of Maryland, Baltimore County
Amita Mehta
Joint Center for Earth Systems Technology
We analyze the daily precipitation time series data at a location in the upper Missouri River Basin (MRB) with prediction as the objective using two approaches: a. Bayesian estimation of a standard Tobit state space model and b. a transfer function approach with an Expectation-Maximization (EM)-like method to "fill in" the zero values (dry days) in the observed series. We use the daily precipitation data simulated by MIROC5, a Global Climate Model (GCM), as an exogeneous predictor. The prediction methods based on the two models can predict zero values as valid predictions, which is desirable for daily precipitation. While the prediction of intensities of precipitation (positive precipitation on wet days) from both the methods are similar on average, the transfer function method was more successful at correctly predicting zero precipitation on days when there was no rain (dry days). A few other relative strengths and weaknesses of the two methods are also discussed.