10th International Aerosol Conference September 2 - September 7, 2018 America's Center Convention Complex St. Louis, Missouri, USA
Abstract View
Roadmap For Statistical Calibration Model Development and Maintenance: Prediction of Organic and Elemental Carbon Composition in PM2.5 with Infrared Spectroscopy
SATOSHI TAKAHAMA, Matteo Reggente, Adele Kuzmiakova, Ann Dillner, Andrew Weakley, Bruno Debus, EPFL
Abstract Number: 1310 Working Group: Instrumentation
Abstract Machine learning algorithms are becoming increasingly important for calibration of instruments (or suite of instruments) that require mathematical separation of analyte signals from analytical or electronic interferences. When physically-based models are not available for signal interpretation, data-driven models can learn discriminating features from a training set of samples to make quantitative predictions in new observations.
We summarize procedures for data preprocessing, data selection, model training, and model evaluation to determine a suitable calibration model among a suite of candidate models that can be generated. The final calibration model can include several models combined together with a classification algorithm (i.e., multilevel modeling framework) when a single model is not suitable for all sample types.
We further discuss methods for identifying limitations in model applicability. Model understanding in the form of variable importance assessment can identify the most discriminating features and how their relationships are used by the calibration model. In the operation phase of the model when reference (and possible other auxiliary measurements) are not available, methods for anticipating precision errors and large prediction errors due to biases that arise from differences in sample composition must be estimated from the primary measurement to monitor model performance. These topics are presented alongside strategies for model updating.
This life cycle of a statistical calibration model is demonstrated for calibration of thermal optical reflectance organic and elemental carbon using collocated Fourier Transform Infrared (FT-IR) spectra from several sites in the IMPROVE and Chemical Speciation Networks. This work presents a synthesis of the recent published efforts to extend FT-IR in this domain of analysis (Dillner et al. 2015a, Dillner et al. 2015b, Kuzmiakova et al. 2016, Reggente et al. 2016, Takahama et al. 2016, Weakley et al. 2016), and also includes new results.
[1] Dillner, A.M., Takahama, S., 2015a. Predicting ambient aerosol thermal–optical reflectance measurements from infrared spectra: elemental carbon. Atmos. Meas. Tech. 8, 4013–4023. https://doi.org/10.5194/amt-8-4013-2015. [2] Dillner, A.M., Takahama, S., 2015b. Predicting ambient aerosol thermal-optical reflectance (TOR) measurements from infrared spectra: organic carbon. Atmos. Meas. Tech. 8, 1097–1109. https://doi.org/10.5194/amt-8-1097-2015. [3] Kuzmiakova, A., Dillner, A.M., Takahama, S., 2016. An automated baseline correction protocol for infrared spectra of atmospheric aerosols collected on polytetrafluoroethylene (Teflon) filters. Atmos. Meas. Tech. 9, 2615–2631. https://doi.org/10.5194/amt-9-2615-2016. [4] Reggente, M., Dillner, A.M., Takahama, S., 2016. Predicting ambient aerosol thermal–optical reflectance (TOR) measurements from infrared spectra: extending the predictions to different years and different sites. Atmos. Meas. Tech. 9, 441–454. https://doi.org/10.5194/amt-9-441-2016. [5] Takahama, S., Ruggeri, G., Dillner, A.M., 2016. Analysis of functional groups in atmospheric aerosols by infrared spectroscopy: sparse methods for statistical selection of relevant absorption bands. Atmos. Meas. Tech. 9, 3429–3454. https://doi.org/10.5194/amt-9-3429-2016. [6] Weakley, A.T., Takahama, S., Dillner, A.M., 2016. Ambient aerosol composition by infrared spectroscopy and partial least-squares in the chemical speciation network: Organic carbon with functional group identification. Aerosol Science and Technology 0, 1–19. https://doi.org/10.1080/02786826.2016.1217389.