Reconstruction of surface NO2 concentrations using a conservative downscaling of OMI, GOME-2, and CMAQ Hyun Cheol Kim1,2, Sang-Mi Lee3, Tianfeng Chai1,2, Barry Baker1,2, Fong Ngan1,2, Li Pan1,2, and Pius Lee1 1Air Resources Laboratory, National Oceanic and Atmospheric Administration, College Park, MD 2Cooperative Institute for Climate and Satellites, University of Maryland, College Park, MD 3South Coast Air Quality Management District, Diamond Bar, CA Air Resources Laboratory

Motivation Need for high-resolution and full coverage NO2 data For epidemiology, One of NAAQS pollutants Raised levels of nitrogen dioxide causes respiratory problems. Significant impacts on people with asthma. For emission estimation, Import precursor of ozone and particulate matter Detection of urban emission sources Fine-scale air quality simulations 위성자료를 이용한 배출량의 보정은 우선 위성에서 관측된 VCD와 모델에서 계산된 VCD를 정확하게 비교하는 과정이 필요합니다. 위성 VCD에 맞추어 모델에서 특정 시간을 맞추어 추출해야 하며, 수직적인 민감도를 고려하기 위해 averaging kernel 기법을 도입합니다. 위성 VCD의 경우 격자 해상도가 위성에 비해 성기기 때문에 이를 보완하기 위해 최근에 개발된 Conservative Downscaling 기법을 이용합니다. - 자세한 사항은 다음 슬라이드에 제시되어 있습니다. VCD는 대기 중 가스상 오염물질의 분포로, 배출량과 보정을 하기 위해 장기런을 이용한 배출량 – 농도 전환율을 고려합니다.

Topics to discuss Fair comparison of model and satellite Temporal sampling (diurnal) Spatial resolution (downscaling) Vertical sensitivity (averaging kernel) Reconstruction of surface NO2 concentration Data fusion Conversion to emission Machine learning (See Kim et al. poster [#4] tomorrow)

Top-down estimation of NOx emissions Satellite Model sVCD mVCD AK mVCD/AK sVCD/DS Local pass time (Temporal match) Conservative Downscaling (Horizontal match) Averaging Kernel (Vertical match) Comparison Emission Adjustment Emission-to-VCD Conversion 위성자료를 이용한 배출량의 보정은 우선 위성에서 관측된 VCD와 모델에서 계산된 VCD를 정확하게 비교하는 과정이 필요합니다. 위성 VCD에 맞추어 모델에서 특정 시간을 맞추어 추출해야 하며, 수직적인 민감도를 고려하기 위해 averaging kernel 기법을 도입합니다. 위성 VCD의 경우 격자 해상도가 위성에 비해 성기기 때문에 이를 보완하기 위해 최근에 개발된 Conservative Downscaling 기법을 이용합니다. - 자세한 사항은 다음 슬라이드에 제시되어 있습니다. VCD는 대기 중 가스상 오염물질의 분포로, 배출량과 보정을 하기 위해 장기런을 이용한 배출량 – 농도 전환율을 고려합니다. Air Resources Laboratory

Conservative Downscaling Problems Model uncertainties: Chemical & Meteorological (e.g. PBL) Retrieval uncertainties: Detection limit & Instrumental error Satellite Model sVCD mVCD AK mVCD/AK sVCD/DS Local pass time (Temporal match) Conservative Downscaling (Horizontal match) Averaging Kernel (Vertical match) Comparison Emission Adjustment Emission-to-VCD Conversion Diurnal profile Footprint resolution 위성자료를 이용한 배출량의 보정은 우선 위성에서 관측된 VCD와 모델에서 계산된 VCD를 정확하게 비교하는 과정이 필요합니다. 위성 VCD에 맞추어 모델에서 특정 시간을 맞추어 추출해야 하며, 수직적인 민감도를 고려하기 위해 averaging kernel 기법을 도입합니다. 위성 VCD의 경우 격자 해상도가 위성에 비해 성기기 때문에 이를 보완하기 위해 최근에 개발된 Conservative Downscaling 기법을 이용합니다. - 자세한 사항은 다음 슬라이드에 제시되어 있습니다. VCD는 대기 중 가스상 오염물질의 분포로, 배출량과 보정을 하기 위해 장기런을 이용한 배출량 – 농도 전환율을 고려합니다. AK quality & spatial resolution Conversion ratio Transport Air Resources Laboratory

How to compare? (A) OMI (B) CMAQ Air Resources Laboratory

Fair comparison (A) OMI (B) CMAQ Downscaling Upscaling (D) CAMQ-upscaled (Pseudo-OMI) (C) OMI-downscaled Air Resources Laboratory

Conservative downscaling of NO2 column density OMI CMAQ Spatial weighting kernel Reconstructed OMI Apply model’s spatial information to OMI pixels Model’s emission inventory has high uncertainty in its intensity, but has reliable accuracy in the location of emission sources. OMI’s original quantity is strictly preserved Air Resources Laboratory

OMI & P3 NO2 column density (CALNEX2010) (B) Original Downscaled Air Resources Laboratory [X10 15 #/cm2] [X10 15 #/cm2]

Model and observations 4-km WRF-CMAQ simulations over Southern California (SCAQMD) AQS surface observations OMI and GOME-2 NO2 column densities (KNMI, TEMIS)

Spatial distribution of OMI and CMAQ NO2 column densities Spatial distribution of OMI and CMAQ NO2 column densities. “DS” and “xDS” denote OMI NO2 column densities with and without the downscaling method, respectively, whereas “AK” and “xAK” denote CMAQ NO2 column densities with and without the averaging kernel, respectively. Air Resources Laboratory

AK DS+AK DS Scatter plot comparison of OMI and CMAQ NO2 column densities. “DS” and “xDS” denote OMI NO2 column densities with and without the downscaling method, respectively, whereas “AK” and “xAK” denote CMAQ NO2 column densities with and without the averaging kernel, respectively.

Spatial distribution of GOME-2 and CMAQ NO2 column densities Spatial distribution of GOME-2 and CMAQ NO2 column densities. “DS” and “xDS” denote GOME-2 NO2 column densities with and without the downscaling method, respectively, whereas “AK” and “xAK” denote CMAQ NO2 column densities with and without the averaging kernel, respectively.

What’s wrong with GOME-2 AK? DS DS+AK Scatter plot comparison of GOME-2 and CMAQ NO2 column densities. “DS” and “xDS” denote GOME-2 NO2 column densities with and without the downscaling method, respectively, whereas “AK” and “xAK” denote CMAQ NO2 column densities with and without the averaging kernel, respectively.

Averaging Kernel (bottom layer) GOME-2 OMI Daily snapshots Air Resources Laboratory

Reconstruction of surface NO2 concentration Satellite inferred surface NO2 concentrations where α is an additional adjustment due to satellite and model uncertainties. For perfect model and satellite, α=1. Air Resources Laboratory

Reconstructed NO2 [OMI] DS Bias = -3.03 RMSE=4.63 R=0.86 CMAQ Bias = -3.73 RMSE=5.03 R=0.86 DS & AK Bias = -2.59 RMSE=4.31 R=0.86 AK Bias = -3.59 RMSE=5.56 R=0.76

Reconstructed surface NO2 concentrations using OMI column densities and CMAQ column-to-surface ratios at 13:30 local time in 2008.

Reconstructed NO2 [GOME-2] DS Bias = 1.84 RMSE=6.62 R=0.87 CMAQ Bias = -3.66 RMSE=6.14 R=0.84 DS & AK Bias = 11.38 RMSE=18.48 R=0.86 AK Bias = 0.64 RMSE=7.18 R=0.82

Reconstructed surface NO2 concentrations using GOME-2 column densities and CMAQ column-to-surface ratios at 09:30 local time in 2008.

OMI with α=1.37 Bias = 0, RMSE=4.31, R=0.86 GOME-2 with α=0.88 Air Resources Laboratory

Conclusion A conservative downscaling was designed to enhance the spatial resolution of satellite measurements by applying the fine-scale spatial structure from the model, with strict mass conservation at each satellite footprint pixel level. With the downscaling approach, NO2 column densities from the OMI (13×24 km) and the GOME-2 (40×80 km) show excellent agreement with the CMAQ (4×4 km) NO2 column densities, with R = 0.96 for OMI and R = 0.97 for GOME-2. We further reconstructed surface NO2 concentrations by combining satellite column densities and simulated surface-to-column ratios from the model. Compared with AQS surface observations, the reconstructed surface concentrations show a good agreement; R = 0.86 for both OMI and GOME-2. Limitation in using current GOME-2 AK is also discussed.