Prediction of Photochemical Properties of Dissolved Organic Matter Using Machine Learning

Apparent quantum yields (phi) of photochemically produced reactive intermediates (PPRIs) formed by dissolved organic matter (DOM) are vital to element cycles and contaminant fates in surface water. Simultaneous determination of phi PPRI values from numerous water samples through existing experimental methods is time consuming and ineffective. Herein, machine learning models were developed with a systematic data set including 1329 data points to predict the values of three phi PPRIs (phi 3DOM*, phi 1O2, and phi center dot OH) based on DOM spectral parameters, experimental conditions, and calculation parameters. The best predictive performances for phi 3DOM*, phi 1O2, and phi center dot OH were achieved using the CatBoost model, which outperformed the traditional linear regression models. The significances of the wavelength range and spectral parameters on the three phi PPRI predictions were revealed, suggesting that DOM with lower molecular weight, lower aromatic content, and a more autochthonous portion possessed higher phi PPRIs. Chain models were constructed by adding the predicted phi 3DOM* as a new feature into the phi 1O2 and phi center dot OH models, which consequently improved the predictive performance of phi 1O2 but worsened the phi center dot OH prediction likely due to the complex formation pathways of center dot OH. Overall, this study offered robust phi PPRI prediction across interlaboratory differences and provided new insights into the relationship between PPRIs formation and DOM properties.

Automated summary

Machine learning models were developed to predict the values of three photochemical reactive intermediates (PPRIs) apparent quantum yields (phi) based on dissolved organic matter (DOM) spectral parameters, experimental conditions, and calculation parameters. The CatBoost model showed the best predictive performance, outperforming traditional linear regression models. The significance of wavelength range and spectral parameters on phi PPRIs predictions was revealed, indicating that DOM with lower molecular weight, lower aromatic content, and a more autochthonous portion possessed higher phi PPRIs.