Research on medium-to long-term runoff prediction for the northwest arid region considering astronomical influences
摘要
针对西北干旱区冰雪融水补给型河流月径流中长期预测中低频背景信息不足问题,以新疆提孜那甫河江卡水文站 1961-2019 年月径流资料为基础,开展月径流量中长期预测研究.整合历史径流、降水、气温、雪水当量、大气环流指数,以及太阳活动、行星距离、相对速度和潮汐势等天文相关变量,采用互信息滞后寻优、层次聚类和簇内互信息优选筛选候选因子,比较 Base-LSTM、Met-LSTM 和 Astro-LSTM 三类模型在 T+1、T+2、T+3 预见期下的预测效果.结果表明,太阳_相对速度_Lag6 被选为最终的代表性天文相关变量,Astro-LSTM 在 T+1、T+2、T+3 预见期下纳什效率系数分别为 0.890、0.878、0.869,均高于 Base-LSTM 和 Met-LSTM的,均方根误差和平均绝对误差同步降低.筛选后的天文相关变量可为江卡站月径流量预测提供常规气象和环流因子之外的低频背景信息,但其作用主要体现为改善预测效果.
To address the insufficient low-and medium-frequency background information for medium-to long-term prediction of monthly run-off in meltwater-supplied rivers in the arid northwest region,this study used monthly runoff data from the Jiangka Hydrological Station along the Tizi Napu River in Xinjiang(1961-2019)as the prediction dataset.Historical runoff,precipitation,temperature,snow water equivalent,atmospheric circulation indices,as well as astronomical variables such as solar activity,planetary distances,relative velocities,and tidal potential were integrated.Candidate factors were selected using mutual information lag optimization,hierarchical clustering,and intra-cluster mutual information selection.The predictive performance of three models—Base-LSTM,Met-LSTM,and Astro-LSTM—was compared for the T+1,T+2,and T+3 forecast periods.Results demonstrated that:Solar Relative Velocity Lag 6 was identified as the repre-sentative astronomical variable.Astro-LSTM achieved NSE values of 0.890,0.878,and 0.869 for the T+1,T+2,and T+3 periods,respec-tively,all exceeding those of Base-LSTM and Met-LSTM,with both RMSE and MAE values decreasing simultaneously.The filtered astronom-ical variables provide low-frequency background information beyond conventional meteorological and circulation factors for monthly runoff pre-diction at the Jiangka Station,primarily enhancing prediction accuracy.