We demonstrate one potential method of using a Non-Homogeneous Hidden Markov Model (NHMM) and a First-Order Auto Regressive Model (AR(1)) to stochastically generate two future 50-year precipitation series: a baseline scenario derived from TRMM Observations (2000-2009) and IPCC AR4 SRES A2 (2070-2099) experiment trend adjusted scenario for the water-stressed Tien Shan region in Central Asia. The four major components are 1) A NHMM training period based on 10 years TRMM data, 2) accurate representation of regional precipitation seasonality using a defined predictor based on a 60 day low pass filter of the TRMM precipitation data 3) GCM trend adjusted predictor (delta adjusted) of the baseline predictor for a projected future climate 4) Coupling with an AR1 model for layering of the low frequency variability shown within river streamflow gauge data.
18 IPCC AR4 GCMs simulations of historical climate of the 20th century (20c3m) for 1950-1999 are used to determine the GCMs that produce broadly realistic precipitation mean monthly values for our region. Future precipitation trend projections are extracted from 13 GCMs, SRES A2 experiments for the 2070-2099 periods. We compare the mean monthly precipitation values from IPCC AR4 20c3m data and TRMM observations from 2000 to 2009 for GCM accuracy and suitability for future projections of climate for our region of interest. By comparing historical (20c3m) mean monthly values and simulations (SRES A2) of 2070-2099, a delta change is derived and used to adjust the baseline predictor for stochastically generating a NHMM 50-year GCM adjusted precipitation series.
Coupling the 8 State Non-Homogeneous Hidden Markov Model (NHMM) stochastic simulations with a First Order Auto Regressive model (AR1) based on river streamflow data allows for a layered approach combining seasonal precipitation and multi-decadal variability. These 50-year simulations were then used in a coupled climate-land-ice-hydrological model for enhancing the understanding of water resource management and planning in the region.