基于LME/BME的珠江三角洲PM2.5星地融合技术研究

收集并处理了遥感反演的气溶胶光学厚度（AOD）、归一化植被指数（NDVI）和气象数据，采用贝叶斯最大熵（BME）结合线性混合模型（LME）估算了2015年10月~2016年3月珠江三角洲地区近地表旬平均PM_2.5质量浓度.结果表明，LME+BME模型的预测精度比LME模型有较大提升，LME+BME模型的交叉验证结果R²为0.751，RMSE为6.886μg/m³，MAE为4.52μg/m³，而LME模型的交叉验证结果R²为0.703，RMSE为7.546μg/m³，MAE为4.927μg/m³.空间分布看，PM_2.5高浓度地区主要集中在广州、佛山、东莞等地区，低浓度地区主要集中在肇庆、惠州、江门的南部等地区；时间变化看，PM_2.5污染比较严重的时间为2015年10月中旬、2015年11月下旬以及2016年3月下旬，而2015年10月上旬、2015年12月上旬和2016年1月下旬污染则相对较低.

Fusion of satellite data and ground observed PM2.5 in Pearl River Delta region with linear mixed effect and Bayesian maximum entropy method

By combining Linear Mixed Effect (LME) model and Bayesian Maximum Entropy (BME) method, ground-level PM_2.5 from October 2015 to March 2016 in Pearl River Delta region were estimated in this paper by AOD, NDVI and meteorological data. The results showed that the prediction accuracy of LME+BME method were greatly improved compared with that of the LME method. The cross-validation R² of LME+BME model was 0.751, and root mean squared prediction error (RMSE) was 6.886μg/m³, the mean prediction error (MPE) was 4.52μg/m³, while R²=0.703, RMSE=7.546μg/m³, and MAE=4.927μg/m³ for the LME method. The high PM_2.5 concentration was mainly located in Guangzhou, Foshan, Dongguan, and the low PM_2.5 concentration was mainly distributed in Zhaoqing, Huizhou, Jiangmen. In terms of seasonal variation, PM_2.5 pollution was more serious in mid-October in 2015, late November in 2015 and late March in 2016, while it was relatively low in early October in 2015, early December in 2015 and late January in 2016.