961463 - High Spatiotemporal NO₂ Estimates by Land Use Random Forests Integrated with WRF/CMAQ

An accurate spatial and temporal prediction of ambient Nitrogen Dioxide (NO2) concentrations is crucial for an appropriate exposure assessment. Here, we develop a spatiotemporal land use random forests model (LURF) by combining random forests (RF) and the Community Multiscale Air Quality modeling system (CMAQ), to predict daily NO2 concentrations for the year 2014 in the Kansai region of Japan. The CMAQ model is used for the simulation of NO2 concentrations; then the CMAQ-simulated NO2 concentrations are used as the input variable for the LURF model. The RF technique is used to capture the nonlinearity among the predicted air pollutants and the predictor variables, including land-use variables, meteorological variables, and CMAQ-estimated NO2 concentration. We compare the predicting performance between a model built with the CMAQ variable (with-CMAQ LURF) and that of a model built without the CMAQ variable (without-CMAQ LURF) for investigating whether the CMAQ-simulated NO2 concentrations can proficiently enhance predicting performance. A cross-validation (CV) technique is performed to evaluate the prediction performance in spatial and temporal aspects and obtained the Coefficient of Determination (R2) and Root Mean Square Error (RMSE). We discover that the CV-R2 of both spatial and temporal CVs from the with-CMAQ LURF model, CV-R2 = 0.69 and 0.72, are higher than that from the without-CMAQ model, CV-R2 = 0.53 and 0.53. In addition, the CV-RMSE values in both aspects from the with-CMAQ LURF model, CV-RMSE = 4.2 and 4.0, are smaller than that from the without-CMAQ LURF model, CV-RMSE = 5.2 and 5.2, respectively. Our findings demonstrate that the with-CMAQ LURF model is advantageous to accurately estimate ambient NO2 concentration in Kansai region of Japan.