Global spatiotemporal estimation of daily high-resolution surface carbon monoxide concentrations using Deep Forest

Abstract

Surface carbon monoxide (CO) is a ubiquitous atmospheric pollutant, which is risky to human health and sustainable developments. To date, data-driven fusion methods have been proposed as new approaches to estimate surface CO concentrations. However, existing related works only focus on national scales or local areas, which cannot contribute to global-scale researches on surface CO. To solve this problem, this study develops a brand-new framework to estimate daily globally distributed surface CO concentrations (2019–2020) at a high spatial resolution (0.05°) through data-driven fusion. Considering the complex relationships between surface CO and multi-source data, an advanced ensemble learning technique, i.e., Deep Forest (DF), is exploited to train the estimation model. Evaluation results (historical) show that the proposed framework presents a desired estimation accuracy over the globe, with the Rs/RMSEs of 0.73/0.273 ppm and 0.77/0.215 ppm at daily and monthly scales, respectively. Meanwhile, the model performance of DF exceeds some frequently-used machine learning techniques, e.g., Light Gradient Boosting Machine and Deep Neural Network. Compared to GEOS Composition Forecasting replay CO product, the proposed framework also performs distinctly better. In addition, seasonal evaluation and uncertainty analyses based on land cover fractions are adopted to fully evaluate the estimated surface CO concentrations. As for the spatial patterns, the estimated results accurately provide the seasonal changes of global high-spatial-resolution surface CO concentrations. At last, the health-related application of ambient CO exposure is discussed in our study. It is discovered that the population-weighted surface CO concentrations of Asia are the most severe in the world, with the annual averaged value ± standard deviation of 0.58 ± 0.13 ppm. The daily surface CO concentrations (0.05°) from our study can help acquire the quantification of global ambient CO exposure and support relevant sustainable researches (e.g., CO-related health effects). The produced global surface CO dataset is available at https://doi.org/10.5281/zenodo.5558611.