研究报告
胡占占,陈传法,胡保健.基于时空XGBoost的中国区域PM2.5浓度遥感反演[J].环境科学学报,2021,41(10):4228-4237
基于时空XGBoost的中国区域PM2.5浓度遥感反演
- Estimating PM2.5 concentrations across China based on space-time XGBoost approach
- 基金项目:山东省自然科学基金项目(No.ZR2020YQ26,ZR2019MD007,ZR2019BD006);山东省高等学校青创科技支持计划(No.2019KJH007)
- 胡占占
- 山东科技大学测绘与空间信息学院, 青岛 266590
- 陈传法
- 山东科技大学测绘与空间信息学院, 青岛 266590
- 胡保健
- 山东科技大学测绘与空间信息学院, 青岛 266590
- 摘要:为了提高PM2.5估算精度,获得连续的PM2.5浓度空间分布,本文提出了一种时空XGBoost模型(STXGB).STXGB模型引入克里金法,将地理信息和时间信息融合到XGBoost算法体系中,通过集成遥感数据、气象数据和地理信息数据建立了基于STXGB模型的PM2.5质量浓度空间估算方法.最后,以2019年中国区域PM2.5质量浓度月数据为例,采用基于样本、站点和时间的十折交叉验证法,评估了STXGB模型的性能,并与BP神经网络(BPNN)、随机森林(RF)、XGBoost、反距离加权XGBoost (XGBIDW)模型结果进行对比.结果表明,STXGB模型的预测精度优于其它模型,其中,STXGB模型验证的决定系数为0.92,均方根误差为6.51 μg·m-3,平均预测误差为4.26 μg·m-3,利用该模型生成的中国区域PM2.5浓度空间分布更为合理.
- Abstract:In order to improve the accuracy of PM2.5 estimation and obtain continuous spatial distribution of PM2.5 concentration, this paper proposes a spatiotemporal XGBoost model (STXGB).The STXGB model introduces the Kriging method, integrates geographic information and time information into the XGBoost algorithm system, and establishes a spatial estimation method of PM2.5 mass concentration based on the STXGB model by integrating remote sensing data, meteorological data and geographic information data. Taking the monthly data of China in 2019 as an example, the performance of the STXGB model was evaluated using a 10-fold cross-validation method based on samples, sites, and time, and it was compared with BP neural network (BPNN), random forest (RF), XGBOOST, inverse distance weighted XGBoost (XGBIDW) model results. The experimental results show that the prediction accuracy of the STXGB model is better than other models. Among them, the R2 verified by the STXGB model is 0.92, the root mean square error is 6.51 μg·m-3, and the average prediction error is 4.26 μg·m-3. The spatial distribution of PM2.5 concentration in China generated by this model is more reasonable.