基于气象雷达反演和云图外推法的临近期降雨预报方法研究

Research of approaching rainfall forecast method based on weather radar inversion and cloud image extrapolation

开展精准化临近期降水预报是开展实时洪涝预测预警的基础。传统的临近期降水预报一般基于气候模式进行数值预报,计算时间较长、空间分辨率低,预报精度仅为15%~20%,难以满足洪涝灾害预测预警需求。为快速、高精度、高空间分辨率预报降水过程,本研究提出了一种基于气象雷达反演和云图外推法的临近期降雨预报方法。主要包括:(1)采用金字塔光流法(Pyramid Lucas Kanade Optical Flow,简称LK光流法)进行云的运动矢量计算,根据运动矢量计算云团的运移位置;(2)提出基于回波强度和梯度特征的云层类型自动识别方法;(3)利用回波反射率与降水强度间的关系进行临近期降水量的反演及预报;(4)对预报结果进行落区精度评估和量级精度评估。2019-2020年四场降水实验研究结果表明:该方法可以实现0~1 h近实时1km高空间分辨率预测,而且预测期内的降水落区预测精度较高,平均精度可达60%;对落区内雨量站的量级精度进行对比分析得出四场降水的偏差BIAS分别为0.23 mm、3.56 mm、-7.12 mm、-5.15 mm,均方根误差RMSE分别为5.09 mm、8.45 mm、21.23 mm、14.63 mm,四场降水的相对误差绝对值均值为37.21%。预测精度均随时间增加呈下降趋势,下降程度与降水系统特征密切相关。总体来说,该方法对大范围高强度的降水预测精度较高,可为实时洪涝过程模拟分析提供重要支撑。

Precisely carrying out near-period precipitation forecasting is the basis for real-time flood forecasting and warning.Generally,forecasting is carried out by means of climate model simulation forecast.This method takes a long time to calculate,which is also low in sptial resolution.The current forecast accuracy is only 15%~20%.In response to this problem,this study proposes a method of near-period rainfall forecast based on weather radar inversion and cloud image extrapolation.The main steps include:(1)Use the pyramid optical flow method(Pyramid Lucas Kanade Optical Flow,LK optical flow method for short)to calculate the motion vector of the cloud,and calculate the movement position of the cloud based on the motion vector;(2)Propose an echo-based method automatic identification method of cloud layer type based on intensity and gradient characteristics;(3)Use the relationship between echo reflectivity and precipitation intensity to invert the forecast of precipitation in the imminent period;(4)Carry out the accuracy assessment and magnitude accuracy of the forecast results.Four precipitetion prediction experiments are carried out in 2019 and 2020.The forecast accuracy of the precipitation area during the 0~1 h forecast period with high spatial resolution of 1 km is relatively high,with an average accuracy of up to 60%.A comparative analysis of the magnitude accuracy shows that the deviation BIAS of the four precipitation fields are 0.23 mm,3.56 mm,-7.12 mm,-5.15 mm,and the root mean square error RMSE are 5.09 mm,8.45 mm,21.23 mm,and 14.63 mm,respectively.The average absolute value of the relative error of field precipitation is 37.21%.The prediction accuracy decreas with time,and the decreasing degree is closely related to the characteristics of the precirltation system.The accuracy of precipitation prediction is low during local rainfall,and the accuracy of precipitation prediction is high during large-scale rainfall.The results can provide important support for real-time flood process simulation analysis.