LSTM模型在京津冀干旱预测应用中的研究  

Evaluating the application of LSTM model for droughtforecasting in Beijing-Tianjin-Hebei region

作　　者：胡小枫 赵安周 相恺政 张向蕊 HU Xiaofeng;ZHAO Anzhou;XIANG Kaizheng;ZHANG Xiangrui(School of Mining and Geomatics,Hebei University of Engineering,Handan 056038,China;Key Laboratory of Natural Resources and Spatial Information,Handan 056038,China;Laboratory of Resources and Environmental Information System,Institute of Geographic Sciences and Natural Resources Research,Chinese Academy of Science,Beijing 100101,China)

机构地区：[1]河北工程大学矿业与测绘工程学院,河北邯郸056038 [2]邯郸市自然资源空间信息重点实验室,河北邯郸056038 [3]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101

出　　处：《西安理工大学学报》2022年第3期356-365,432,共11页Journal of Xi'an University of Technology

基　　金：国家自然科学基金资助项目(42171212,42071246);河北省普通高等学校青年拔尖人才计划资助项目(BJ2018043);2021年河北省硕士研究生创新资助项目(CXZZSS2021089);资源与环境信息系统国家重点实验室开放基金资助项目。

摘　　要：在多时间尺度上对京津冀的旱情进行准确地预测可为当地抗旱提供有效支撑。基于1961—2019年京津冀22个气象站点的降水、气温、平均湿度等多个气象因子数据,计算标准化降水蒸散发指数(standardized precipitation evapotranspiration index,SPEI),并构建长短时记忆神经网络模型(long short-term memory model,LSTM)对多时间尺度的SPEI(SEPI-3、SPEI-6、SPEI-9、SPEI-12和SPEI-24)进行时空预测。采用平均绝对误差(MAE)、均方根误差(RMSE)和决定系数(R 2)对预测模型进行精度评估。结果表明在唐山气象站上,LSTM模型对多时间尺度SPEI(SEPI-3、SPEI-6、SPEI-9、SPEI-12和SPEI-24)值的预测效果较好。在时间序列预测方面,LSTM模型的预测精度随着SPEI的时间尺度增加而逐渐提高,其中LSTM模型在3个月和24个月SPEI时间尺度上的MAE分别为0.473和0.197,RMSE分别为0.627和0.260,R 2分别为0.604和0.935。在空间分布预测上,LSTM模型对2019年京津冀季节和年时间尺度上的SPEI预测值与实际值在空间分布上相似,说明LSTM模型能够较为精确地预测京津冀干旱的空间分布。Accurate prediction of drought condition in Beijing-Tianjin-Hebei on multiple time scales can provide effective support for local drought resistance. Based on the data multiple meteorological factors such as precipitation, temperature and mean humidity from 22 meteorological stations in Beijing-Tianjin-Hebei region from 1961 to 2019, the standardized precipitation evapotranspiration index(SPEI) is calculated, the long short-term memory models(LSTM) constructed to forecast the time and space of SPEI(including SPEI-3, SPEI-6, SPEI-9, SPEI-12 and SPEI-24) at multi-temporal scales. The accuracy of all prediction models are determined by mean absolute error(MAE), root mean square error(RMSE) and decision coefficient(R~2). The results show that LSTM model has a well predictive effect on multi-time scales SPEI(SPEI-3, SPEI-6, SPEI-9, SPEI-12 and SPEI-24) values of meteorological station in Tangshan. In terms of time series forecast, the prediction accuracy of LSTM model for SPEI is increased with the increase of time scale. MAE values of LSTM model in SPEI-3 and SPEI-24 are 0.473 and 0.197, RMSE values are 0.627 and 0.260, R~2 values are 0.604 and 0.935 respectively. In terms of spatial distribution forecast, the predicted value of LSTM model on seasonal and annual time scales of Beijing-Tianjin-Hebei region in 2019 is very similar to the actual value of SPEI, indicat that LSTM model can accurately forecast the spatial distribution of drought in Beijing-Tianjin-Hebei region.

关 键 词：干旱预测 LSTM模型 SPEI 京津冀 

分 类 号：P426.616[天文地球—大气科学及气象学]