| 引用本文: | 刘小强,代智光,吴立峰.GPR、XGBoost和CatBoost模拟江西地区参考作物蒸散量的适应性研究[J].灌溉排水学报,0,():-. |
| Liu Xiaoqiang,Daizhiguang,Wulifeng.GPR、XGBoost和CatBoost模拟江西地区参考作物蒸散量的适应性研究[J].灌溉排水学报,0,():-. |
|
| 摘要: |
| 【背景】江西的旱涝急转需合理的灌溉制度才能保证作物的稳产。【目的】为提高机器学习模型模拟参考作物蒸散量在江西省适应性和精度。【方法】基于江西南昌等15个站2001—2015年日值气象数据（最高气温、最低气温、地表辐射、大气顶层辐射、相对湿度和2 m高风速），以FAO-56 Penman-Monteith（P-M）公式的计算结果作为对照，建立了计算ET0的高斯过程回归（GPR）、极限梯度增强（XGBoost）和梯度增强决策树（CatBoost）模型，并分别与经验模型进行比较。【结果】结果表明：各气象参数对机器学习模型模拟ET0的精度影响由大到小依次为：Rs、Tmax和Tmin、RH、U2，且采用Tmax、Tmin、Rs和RH气象参数组合的机器学习模型（RMSE<0.2 mm/d）模拟ET0精度高。此外，3种机器学习模型在有限的气象数据时具有较好的适用性，且优于传统经验模型，其中GPR和CatBoost模型的预测精度高，但GPR模型稳定性最好。【结论】考虑到所研究模型调参的复杂性、预测精度和稳定性，GPR模型可作为江西地区参考作物蒸散量模拟的推荐方法。 |
| 关键词: 参考作物蒸散量；高斯过程回归；极限梯度增强；梯度增强决策树；经验模型 |
| DOI： |
| 分类号:S274.1；S274.4 |
| 基金项目:江西省教育厅研究项目青年基金，江西省科技厅自然科学基金资助项目 |
|
| Study on adaptability of GPR,XGBoost and CatBoost to simulate reference crop evaporation in Jiangxi Province |
|
Liu Xiaoqiang, Daizhiguang, Wulifeng
|
|
Nanchang Institute of Technology
|
| Abstract: |
| 【Background】The rapid turn of drought and waterlogging in Jiangxi need a reasonable irrigation regime to ensure the steady production of crops.【Objective】In order to improve the adaptability and accuracy of machine learning model simulation reference crop evapotranspiration in Jiangxi.【Method】meteorological data from 2001 to 2015 at 15 stations of Jiangxi（i.e.,daily maximum and minimum ambient temperature，global solar radiation，extra-terrestrial solar radiation, relative humidity and 2m high wind speed）was used to train and text the proposed model of Gaussian process regression、 extreme gradient enhancement and gradient-enhanced decision tree.【Result】The results show thatthe influence of each meteorological element on the accuracy of machine learning model simulation ET0 from large to small is: Rs, Tmax and Tmin, RH, U2; the machine learning model with combination of Tmax, Tmin, Rs and U2 meteorological elements（RMSE<0.2mm/d）obtains the high ET0accuracy. In addition, three machine learning models have good applicability in limited meteorological data, and are superior to the traditional empirical model, The prediction accuracy of GPR and CatBoost model is high, but the stability of GPR model is good.【Conclusion】 In view of the complexity, accuracy and stability of the model, the GPR model can be used as a recommended method for the simulation of evapotranspiration in Jiangxi province. |
| Key words: reference crop evapotranspiration; gaussian process regression; extreme gradient boosting; gradient boosting with categorical features support; empirical model |
