【24h】

Towards a National Scale Spatio-Temporal Model for Ambient PM2.5 in India

机译:迈向印度环境PM2.5的全国规模时空模式

获取原文

摘要

We report the results of a nation-wide effort to develop a PM2.5 model on a fine spatial and temporal resolution. We will combine WRF-Chem estimates of particle concentrations on a finer spatial scale than GEOS-Chem, updated fire emissions inventories, satellite remote sensing AOD data on a 1 km grid, land use data, and weather data (including PBL height) and train on ground based monitoring using multiple machine learning techniques. Parameter choices will be made using 10-fold cross validation. PM10 data was converted into PM2.5 using a support vector machine on collocated monitors. Nearby monitoring is included as a predictor and missing AOD values will be filled in so that predictions are available for all days and locations. Final models are ensemble weighted averages of the individual models: support vector machine, neural network, gradient boosting, and random forests). Within grid cell land use regression for address specific estimates are done similarly on monitor deviations from cell averages. In preliminary analyses for the greater Dehli area, we find cross validated R2 for annual average PM2.5 of greater than 0.91 except for one year (2011, CV R2=0.72).
机译:我们报告了全国范围内的结果,以在精细的空间和时间分辨率下开发PM2.5模型。将WRF-Chem估计与Geos-Chem的更精细的空间级,更新的火灾排放库存,卫星遥感AOD数据在1 km网格,土地利用数据(包括PBL高度)和火车上基于地面的监控,使用多机学习技术。参数选择将使用10倍交叉验证进行。 PM10数据在并置监控显示器上使用支持向量机转换为PM2.5。附近的监视包含作为预测仪,缺少AOD值将填写,以便所有日期和位置都可以使用预测。最终模型是各个模型的合奏加权平均值:支持向量机,神经网络,渐变升压和随机林)。在网格单元格中,地址特定估计的使用回归是同样的关于从细胞平均值的监测偏差进行完成。在更大的Dehli地区的初步分析中,除了一年(2011年,CV R2 = 0.72),我们发现年平均PM2.5的交叉验证的R2大于0.91。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号