基于机器学习算法和能量闭合理论的高标准稻田蒸散估算  

Simulation of Evapotranspiration in a Well-facilitated Paddy Field Based on Machine Learning Algorithms and Energy Balance Closure

作　　者：邰久 王伟[1] 徐敏 胡凝[1] 陈上 徐敬争 胡小旭 吕恒 祝子涵 赖宇婧 TAI Jiu;WANG Wei;XU Min;HU Ning;CHEN Shang;XU Jing-zheng;HU Xiao-xu;LV Heng;ZHU Zi-han;LAI Yu-jing(Key Laboratory of Meteorological Disaster,Ministry of Education/Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters/Key Laboratory of Ecosystem Carbon Source and Sink,China Meteorological Administration/Jiangsu Key Laboratory of Agricultural Meteorology,Nanjing University of Information Science&Technology,Nanjing 210044,China;Climate Center of Jiangsu Province,Nanjing 210008;Aerospace New Weather Technology Co.,Ltd,Wuxi 214028)

机构地区：[1]南京信息工程大学气象灾害教育部重点实验室/气象灾害预报预警与评估协同创新中心/中国气象局生态系统碳源汇重点开放实验室/江苏省农业气象重点实验室,南京210044 [2]江苏省气候中心,南京210008 [3]航天新气象科技有限公司,无锡214028

出　　处：《中国农业气象》2024年第10期1109-1122,共14页Chinese Journal of Agrometeorology

基　　金：江苏省“333人才”领军型人才团队(BRA2022023);江苏省气象局揭榜挂帅科研项目(KZ202302);江苏省农业气象重点实验室开放基金(JKLAM2305)。

摘　　要：为选出模拟长江中下游高标准稻田各生育期实际蒸散(ETa)的最优机器学习模型,并探究能量闭合对机器学习模型模拟ETa的影响,基于南通市2020年高标准水稻田小气候、土壤和通量观测数据,分析各生育期稻田ETa及相关因子的时间变化特征,利用BP神经网络和随机森林两种算法估算各生育期ETa,评估基于波文比的强迫能量闭合对机器学习模型模拟ETa精度的影响。结果表明:(1)在不同生育期,气象和土壤因子对稻田ETa的重要性不同,入射短波辐射(K↓)始终是稻田ETa的主控因子。(2)加入K↓可显著提高机器学习模型对ETa的模拟精度,相关系数提高了14.9%,RMSE降低51.1%。5种变量组合中,包含饱和水汽压差(VPD)、风速(U)、气温(Ta)、相对湿度(RH)和入射短波辐射(K↓)的BP1模型是模拟分蘖期前稻田ETa的最佳模型,包含Ta、RH和K↓的BP3模型更适于模拟分蘖期后的稻田ETa。(3)强迫能量闭合能改善BP神经网络模型对ETa的模拟效果,在分蘖期前更为明显,5种变量组合中,BP2模型(U、Ta、RH和K↓)在能量闭合后的模拟效果提升最明显,相关系数提高了3.5%,RMSE降低25.7%。To find the optimal machine learning model for simulating the actual paddy evapotranspiration(ETa)in each growth stages,and to quantify the impacts of forcing energy balance closure on simulation results,authors firstly analyzed the temporal variations in ETa and its influencing factors(air temperature Ta,relative humidity RH,wind speed U,vapor pressure deficit VPD,soil moisture at 5cm depth SWC and incident shortwave radiation K↓)with in-situ observations in a well-facilitated paddy filed in Nantong city,Jiangsu in 2020.Then the ETa of each growth stages were simulated by two machine learning algorithms:back propagation(BP)neural network and random forest.Finally,the impact of forcing energy balance closure on ETa simulation by BP model was quantified.The results showed that the relative importance of influencing factors to paddy ETa differed among growth stages.K↓was the most important influencing factor for ETa,while SWC had negligible effect on ETa.Therefore,including K↓significantly improved the ETa simulation by BP neural network algorithm with correlation coefficient(R)increased by 14.9%and root mean square error(RMSE)reduced by 51.1%.BP1 model containing the five meteorological factors(Ta,RH,U,VPD and K↓)ranked the best for simulating ETa before tillering stage,while the BP3 model including Ta,RH,and K↓was more suitable after tillering stage.Forcing energy balance closure had improved the simulation performance of the BP neural network algorithm especially before tillering stage.After forcing energy balance closure,the simulation of BP2 model(Ta,RH,U and K↓)had been improved most obviously among five variable combinations.The R between BP2 simulations and field observations increased by 3.5%and the RMSE reduced by 25.7%.

关 键 词：蒸散 机器学习算法 能量闭合 水稻 生育期 

分 类 号：S511[农业科学—作物学] S161.4[自动化与计算机技术—控制理论与控制工程] TP181[自动化与计算机技术—控制科学与工程]