| |
|
| Submitter's Name |
Surafel Abebe |
| Session Name |
Poster Session: Hydrologic, Biogeochemical and Ecological Processes 1 |
| Category |
Hydrologic, biogeochemical and ecological processes |
| Poster Number |
236 |
| |
| Author(s) |
Surafel Abebe, Environmental Management |
| |
Tirusew Abebe,
Hydrology |
| |
|
| |
Conditional Markov Mixture Models for Seasonal Rainfall Simulation |
| |
|
| |
Markov mixture models (also known as hidden Markov models) have been successfully used to simulate both stream flow and rainfalls at daily and annual time scales. The models simulate a time series by switching between different unobservable (hidden) climate states and sample the parameters of interest from the distribution that is specific to the state. Then the transition probability would be used to progress through time. At a monthly time scale, their applications is limited because seasonally presents another set of challenge. One of the approaches that was used to solve the seasonality problem is by fitting individual mixture models for each seasons and simulate the time series using the entire suits of models. Such a solution even though enable one to reproduce the long-term behavior of the time series, there is no mechanism to make a conditional (short-term) simulation as there is no explicit link between two time steps, say, Jan to Feb. Each time step belongs to different season, hence, different mixture model. Here we suggest a methodology that is based on a posterior that modifies emission probabilities for each season based on previous month’s observations. Previous month observation could be rainfall or some exogenous variable such as sea surface temperature anomaly. The results show a promising application of conditional Marko mixture models for short-term monthly rainfall simulation. Key words: Mixture Models, Conditional Simulation, Rainfall |
| |