Analysing spatio-temporal patterns of the global NO 2 -distribution retrieved from GOME satellite observations using a generalized additive model

M. Hayn, Steffen Beirle, Fred A. Hamprecht, U. Platt, Bjoern Menze, Thomas Wagner

Abstract

With the increasing availability of observational data from different sources at a global level, joint analysis of these data is becoming especially attractive. For such an analysis – oftentimes with little prior knowledge about local and global interactions between the different observational variables at hand – an exploratory, data-driven analysis of the data may be of particular relevance. In the present work we used generalized additive models (GAM) in an exemplary study of spatio-temporal patterns in the tropospheric $NO_2$-distribution derived from GOME satellite observations (1996 to 2001) at global scale. We focused on identifying correlations between $NO_2$ and local wind fields, a quantity which is of particular interest in the analysis of spatio-temporal interactions. Formulating general functional, parametric relationships between the observed $NO_2$ distribution and local wind fields, however, is difficult – if not impossible. So, rather than following a model-based analysis testing the data for predefined hypotheses (assuming, for example, sinusoidal seasonal trends), we used a GAM with non-parametric model terms to learn this functional relationship between $NO_2$ and wind directly from the data. The $NO_2$ observations showed to be affected by wind-dominated processes over large areas. We estimated the extent of areas affected by specific $NO_2$ emission sources, and were able to highlight likely atmospheric transport "pathways". General temporal trends which were also part of our model – weekly, seasonal and linear changes – showed to be in good agreement with previous studies and alternative ways of analysing the time series. Overall, using a non-parametric model provided favorable means for a rapid inspection of this large spatio-temporal $NO_2$ data set, with less bias than parametric approaches, and allowing to visualize dynamical processes of the $NO_2$ distribution at a global scale.

Introduction

Nitrogen dioxide ($NO_2$) is a critical atmospheric pollutant linked to respiratory issues and ozone formation. While satellites like GOME (Global Ozone Monitoring Experiment) provide global observations, interpreting this data is challenging because $NO_2$ levels are influenced by a complex overlap of factors: human activity (weekly cycles), climate (seasonal cycles), economic growth (linear trends), and meteorology (wind transport).

Traditional "parametric" models often assume these relationships follow a specific shape—for example, assuming seasonal changes always look like a simple sine wave. However, real-world interactions, especially the way wind "smears" pollution away from cities, are far too complex for simple equations. This paper addresses the need for a more flexible, data-driven approach to explore these spatio-temporal patterns without making restrictive prior assumptions.

Methodology

The study employs a Generalized Additive Model (GAM). Unlike standard linear regression, a GAM models the response variable as a sum of smooth, non-parametric functions (splines) of the explanatory variables. This allows the model to "learn" the shape of the relationship directly from the data. The model decomposes $NO_2$ levels into four distinct additive terms: a linear long-term trend, a 365-day periodic seasonal cycle, a discrete weekly cycle, and a cyclic spline for wind direction.

Figure 5: Decomposition of the NO2 VCD time series into trend, annual cycle, weekly cycle, and wind component for four distinct global regions using the GAM approach.

Datasets

  • GOME TVCD: Tropospheric $NO_2$ Vertical Column Density retrieved from the ERS-2 satellite (1996–2001).
  • ECMWF Wind Data: Global surface wind data sampled every 6 hours and averaged to match satellite overpass times.
  • Spatial Resolution: Data was re-sampled to a $0.5^\circ \times 0.5^\circ$ global grid.

Key Insights and Results

Wind as a Diagnostic Tool

By correlating $NO_2$ with wind direction, the authors could "point" back to major pollution sources. They identified clear "pathways" of pollution extending over 1,000 km from industrial centers. The model effectively uncouples local emissions from short-term wind transport, which is vital for improving atmospheric chemistry models.

Figure 4: Visualization of NO2 distribution and the impact of wind-dominated transport processes, highlighting pollution pathways from major sources.

Spatio-Temporal Patterns

  • The "Sunday Effect" Travels: In industrialized regions like Europe, the typical Sunday drop in $NO_2$ (due to reduced traffic/industry) was observed to shift to Monday or Tuesday in Eastern Europe, physically tracking the eastward transport of air masses.
  • Complex Seasons: In the Indian Ocean, the GAM captured a "double peak" in $NO_2$ levels due to monsoon transitions, a feature that rigid parametric models often miss.
  • Regional Trends: The study confirmed massive increases in $NO_2$ over China (up to +15% per year) and the Middle East (+11% per year), while identifying significant decreases in parts of Europe (up to -12% per year).

Conclusion and Future Work

The use of non-parametric GAMs provided a rapid, low-bias inspection of large spatio-temporal datasets. While the coarse resolution of GOME ($320 \times 40$ km) and reliance on surface winds are limitations, the methodology proves robust. Future work will apply these techniques to higher-resolution instruments like OMI or SCIAMACHY and incorporate vertical wind profiles to better account for transport at higher altitudes.

Summary updated

Abstract

With the increasing availability of observational data from different sources at a global level, joint analysis of these data is becoming especially attractive. For such an analysis – oftentimes with little prior knowledge about local and global interactions between the different observational variables at hand – an exploratory, data-driven analysis of the data may be of particular relevance. In the present work we used generalized additive models (GAM) in an exemplary study of spatio-temporal patterns in the tropospheric NO2-distribution derived from GOME satellite observations (1996 to 2001) at global scale. We focused on identifying correlations between NO2 and local wind fields, a quantity which is of particular interest in the analysis of spatio-temporal interactions. Formulating general functional, parametric relationships between the observed NO2 distribution and local wind fields, however, is difficult – if not impossible. So, rather than following a model-based analysis testing the data for predefined hypotheses (assuming, for example, sinusoidal seasonal trends), we used a GAM with non-parametric model terms to learn this functional relationship between NO2 and wind directly from the data. The NO2 observations showed to be affected by wind-dominated processes over large areas. We estimated the extent of areas affected by specific NO2 emission sources, and were able to highlight likely atmospheric transport "pathways". General temporal trends which were also part of our model – weekly, seasonal and linear changes – showed to be in good agreement with previous studies and alternative ways of analysing the time series. Overall, using a non-parametric model provided favorable means for a rapid inspection of this large spatio-temporal NO2 data set, with less bias than parametric approaches, and allowing to visualize dynamical processes of the NO2 distribution at a global scale.
BibTeX
@inproceedings{hayn2009analysing,
  author = {M. Hayn and Steffen Beirle and Fred A. Hamprecht and U. Platt and Bjoern Menze and Thomas Wagner},
  title = {{Analysing spatio-temporal patterns of the global NO <sub>2</sub> -distribution retrieved from GOME satellite observations using a generalized additive model}},
  booktitle = {Atmospheric},
  year = {2009},
  doi = {10.5194/acp-9-6459-2009},
}
Figure