基于183 GHz机载微波辐射计探测水汽和云中液态水反演算法研究

Retrieval Method of Precipitation Water Vapor and Cloud Liquid Water Path Based on 183 GHz Airborne Microwave Radiometer

基于机载微波辐射计GVR,以北京探空资料作为训练数据,机载对空微波辐射传输方程作为前向算子,建立基于BP神经网络和Decker模型基础上的积分水汽含量和液态水路径反演算法。与GVR自带算法相比,新算法增加观测高度作为输入变量,考虑温度对云水相态的影响建立新的云模型,对历史探空资料进行补全,增加积分高度至30 km。利用数值模拟检验、外场观测试验和观测误差传递分析对两种算法反演的积分水汽含量反演误差进行验证,结果表明:新算法和自带算法模拟计算的反演值与探空计算值相关系数分别为0.9988、0.9929,自带算法反演值普遍低于探空计算值;新算法和GVR自带算法反演的积分水汽含量统计均方根绝对误差分别为0.05~1.30 mm、0.2~3.0 mm,相对误差分别为1%~10%、4%~65%,新算法在6 km以下计算的积分水汽含量相对误差最小值为1%,整层高度上约75%的相对误差低于5%,而自带算法仅有不到1%的相对误差低于5%;以2016年11月20日观测个例为例,GVR仪器观测误差经过新算法和自带算法反演公式传递后造成的误差在3000 m高度分别为0.05 mm和0.06 mm,随水汽探测值增大(飞机高度下降),自带算法传递误差增大幅度明显高于新算法。选择2021年6月30日飞行个例,对两种算法反演的液态水路径反演误差进行验证,结果表明:在飞机上空无云时段,新算法和自带算法反演的液态水路径范围分别为0 mm、0~0.006 mm。误差分析表明,新算法可提升本地积分水汽含量和液态水路径反演精度。

Based on the airborne microwave radiometer(G-band water vapor radiometer,named GVR),a retrieval algorithm of precipitation water vapor and liquid water path based on BP neural network and Decker model is established by using the Beijing sounding data as the training data,and the airborne microwave radiation transfer equation as the forward algorithm.Compared to the GVR own algorithm,the observation height is added as an input variable.A new cloud model considering the influence of temperature on the cloud water phase,and the historical sounding data completed to increase the integral height to 30 km are applied in the new algorithm.The precipitation water vapor errors calculated by two algorithms are verified by using numerical simulation test,external field observation test and observation error transfer analysis.The results show that the correlation coefficient between the retrieval and probe values of the new and GVR own algorithms are 0.9988 and 0.9929 respectively.The retrieval value of the GVR own algorithm is generally lower than the sounding calculation value.The absolute errors of precipitation water vapor calculated by the new and GVR own algorithms are 0.05-1.30 mm and 0.2-3.0 mm,relative errors are 1%-10%and 4%-65%,the minimum relative error calculated by the new algorithm is 1%below 6 km,about 75%of new algorithm results are below 5%while only less than 1%of GVR own results are below 5%.Take the data observed on 20 November 2016 as an example.The errors caused by the GVR instrument observation errors that are calculated by the new algorithm and the GVR own algorithm retrieval formula are 0.05 mm and 0.06 mm at 3000 m height respectively.After of retrieval formula,with the increase of the water vapor detection value(the aircraft altitude drops),the transfer error increase of the GVR own algorithm is significantly higher than that of the new algorithm.The flight case on 30 June 2021 is selected to verify the liquid water path retrieval error of the two algorithms.The results show that the liquid water path ranges of the new algorithm and the own algorithm are 0 mm and 0-0.006 mm during cloudless periods over the plane.Error analysis shows that the new algorithm is effective for improving the retrieval accuracy of local precipitation water vapor and liquid water path.