Trend estimation and forecasting of atmospheric pollutants in the Mexico City Metropolitan Area through a non-parametric perspective

Abstract

Trends and forecasts of the main atmospheric pollutants (O3, SO2, NO2, CO, PM10, PM2.5, NO and NOX) are estimated by regions of the Metropolitan Zone of Mexico City (MZMC), with maximum daily data from 2008 to 2018. A nonparametric statistical smoothing controlled technique based on the Hodrick and Prescott filter and estimated through the Kalman filter, is used. Both point and interval estimates, as well as their respective forecasts are generated. The estimates against the Environmental Standard for Mexico City (NADF-009-AIRE-2017) are compared, and it is evident that, in general, they are still distant from good air quality in the MZMC, as opposed to CO and NO2. The remaining pollutants have trends and forecasts, which are far from the permissible limits.

Trend detection of atmospheric time series

“…Except for simple situations, such as a turnaround or a leveling off of the trend, it is generally difficult to interpret highly nonlinear behavior through an explicit parametric representation or a deterministic model (Chandler and Scott, 2011). Many adaptive nonlinear trend fitting techniques are available, such as state-space modeling (and its variant, dynamical linear modeling; Petris et al, 2009;Durbin and Koopman, 2012;Laine et al, 2014), vector autoregressive modeling (Holt and Tera ¨svirta, 2020), empirical mode decomposition (Wu et al, 2007), signal filter technique (Thoning et al, 1989), the Gasser-Mu ¨ller kernel smoothing (Gasser and Mu ¨ller, 1984), the Kalman filter (Harvey, 1990;Ramos-Ibarra and Silva, 2020), and the Kolmogorov-Zurbenko filter (Rao et al, 1997;Yang and Zurbenko, 2010). It should be emphasized that even though the above techniques are able to capture the nonlinearity in the time series, not all the curve features can be considered to be a change point of the trends or having interpretable information (see Figure 9).…”