Based on the monthly inflow data of Danjiangkou reservoir as well as the meteorological data of 41 weather stations in the upper basin during 1959-2016, the key meteorological factors and timing that affecting inflow were identified by correlation analysis, and then forecasting models for each month were established and systematic errors were also analyzed. The results show that the reservoir inflow is significantly and positively correlated with precipitation, with the highest correlation ranging from 0.45 to 0.85 for each month, and negatively correlated with temperature, with the highest correlation ranging from -0.26 to -0.54. generally, the rainy season has a higher correlation, and the influencing timing is in the latest 1-3 months, while the nonrainy season has a lower correlation, with a more longer influencing timing of 2-6 months. Using mean temperature and precipitation in the last several months as predictor variables, stepwise regression model with logarithmic transformations was chosen as the best fitted model, the models could explain 45%-88% of the total variance for each month, and more than 75% for rainy season months. 10-fold cross validation give a mean absolute percentage error within 20.5%-40.7% for each month, and 9.6% for whole year. The spatial and temporal distribution of precipitation can have a significant impact on forecast accuracy, with precipitation of the upper subbasin of hanzhong, shiquan, ankang and baihe above normal level, or precipitation in the edge of the basin below normal level, there may be a lower forecast discrepancy, while higher forecast discrepancy on the contrary, and a more concentrated or heavy precipitation results in a lower forecast discrepancy, a more even precipitation distribution leads to higher discrepancy.
Key words:Danjiangkou Reservoir; inflow; regression analysis; forecasting error