Seasonal Precipitation Forecasts Over China Using Multiple Statistical And Dynamical Models And Application

Seasonal precipitation forecasts can be highly valuable for different socioeconomic sectors, such as flood and drought disaster detection and mitigation, water resources planning and management, the energy sector, rain-fed agriculture, human settlements, environment, tourism and many other purposes. This is especially the case for the countries which are more prone to natural hazards such as China. However, it remains a formidable challenge to produce forecasts of precipitation at the seasonal time scale that are of sufficient accuracy to meet public demands, due to the chaotic characteristic and high complexity of the climate system. This P.h.D dissertation studied the potential of using mutltiple statistical and dynamical models to forecast seasonal precipitation over China, the main contents, results and conclusions are summarized as follows:(1) The temporal-spatial distribution and trend of seasonal precipitation over China was analyzed over the period1950-2011. Eight climate indices that have connections with the inter-annual variability of seasonal precipitation over China were identified. Seasonal precipitation over China exhibits great temporal and spatial variability. In general, precipitation increases from northwest to southeast. The overall trend of seasonal precipitation totals averaged across the184grid cells over China is not significant at the5%confidence level according to a modified Mann-Kendall test. Averaged across all12seasons,4.25%of the grid cells in which precipitation exhibit significantly increase and3.44%exhibit significantly decrease.(2) A Bayesian probabilistic forecast framework was proposed for forecasting seasonal precipitation over the whole China and for all12overlapping seasons. For each grid cell and season, we established24forecast models using eight large-scale oceanic-atmospheric indices at lag times of1-3months as predictors based on a Bayesian joint probability modelling approach. Each model has one predictor and one predictand, so that the models were independent from each other. We then merged forecasts using Bayesian model averaging to combine the strengths of the individual models. Forecast accuracy and reliability were assessed through leave-one-year-out cross validation. The merged forecasts were most skillful in spring and late summer to early winter periods. Positive forecast skill was mostly retained when forecast lead time is increased from0to2months. Forecast distributions were found to reliably represent forecast uncertainty. (3) Skill of two lastest CGCMs-ECMWF SYS4and Austrilian BoM POAMA2.4as well as other five CGCMs used in the ENSEMBLES project in forecasting seasonal precipation over China was evaluated. The results suggested that all the seven CGCM suffer from system deficiency:the ensembles show system bias and exhibit insufficient spread. More importantly, seasonal precipitation forecast skill is still too low. Forecast skill of the two lastest CGCMs was higher than the other five in terms of accuracy, reliability, sharpness and resolution.(4) A statistical post-processing method that use multiple outputs from CGCM was proposed to improve the skill of the raw CGCM.To ameliorate systematic deficiencies in precipitation forecasts from raw CGCM, a modified Bayesian joint probability modelling approach was applied to calibrate the ensemble mean of the raw seasonal forecasts. To improve the skill of precipitation forecasts, six large-scale climate indices, calculated from CGCM SST forecasts, were used as predictors to establish a set of BJP statistical bridging forecasts. The calibration forecasts and bridging forecasts were merged through Bayesian model averaging to combine strengths from different models. Results suggest that the BJP calibration models effectively removed biases and improved statistical reliability of the raw forecasts. Forecast accuracy was further improved by inclusion of bridging model forecasts.(5) A merged forecast framework from multiple statistical and dynamical models was proposed for seasonal forecasting of preicipation over China. This framework combines the strength of individual statistical model forecasts that using multiple lagged climate indices as predictors, with outputs from multiple dynamical models, thus has the potential to maximize the overall forecast skill.(6) A seasonal streamfiow forecast model that use multi-souces information was proposed. This model use hydrological model predictions, antecedent streamflow and catchment rainfall as predictors to represent initial catchment condition; and use dynamical seasonal forecast system output and large-scale climate indices as predictors to represent climateduring the forecast period. Results suggest that this model was rather accurate and reliable in forecasting summer inflow into Fengman reservoir and Huanren reservoir. Seasonal inflow forecast information was then used in the optimal operation of Huanren reservoir, results suggest that power generation benefit was improved compared to the current used operation rules.