Many of the currently available satellite products for air quality applications tend to have a relatively coarse spatial resolution on the order of several kilometers to even tens of kilometers. While this kind of spatial resolution is useful for mapping air quality at the global, continental, and even country scales, the applications for local-scale applications, e.g. in urban areas where the majority of the air pollution issues occur, are somewhat limited.

For this reason, SAMIRA is developing algorithms for downscaling satellite products for air quality to a spatial resolution that is more suitable for urban-scale application (ca. 1 km). In order to do so, SAMIRA uses spatial proxy information that is available at a finer spatial resolution than the satellite data itself. This could be dynamic information from a chemistry transport model or any type of static information whose spatial patterns are correlated to some extent with the spatial patterns in the satellite data. The downscaling algorithm itself is based on geostatistical principles (area-to-point kriging with external drift).


The above Figure shows a demonstration of the downscaling methodology based on a simulated example (arbitrary units). The top two panels show the input data, namely the coarse-resolution satellite information (left) and the fine-resolution proxy information (right). Not the different scales of the two datasets, indicating that the proxy dataset does not have to show the exact same variable as the satellite parameters, as long as there is some correlation in spatial patterns. The bottom left panels shows the result using the SAMIRA downscaling methodology based on geostatistics on the two input datasets. It demonstrates how the satellite information is downscaled, i.e. spatially distributed, according to the dominant spatial patterns visible in the proxy information. The bottom right panel shows that a re-aggregation of the downscaled map results in the same information originally measured by the satellite, i.e. no bias is introduced as part of the downscaling process.


This figure demonstrates the initial downscaling algorithm with a real-world example. The left panel shows the average tropospheric column of nitrogen dioxide (NO2) measured the OMI instrument onboard the Aura platform for May through September 2014 over the greater Oslo (Norway) area [in units of 10^15 molec./cm^2]. A clear hotspot over the city of Oslo is visible with generally lower column amounts outside of the urban areas. The center panel shows the result of a traditional downscaling using simple bilinear interpolation, which produces a very smooth field with unrealistic spatial patterns. In contrast, the right panel shows the result of the SAMIRA downscaling methodology using proxy information on spatial patterns from the EMEP model, indicating a much more realistic spatial distribution with stronger spatial gradients. The satellite-derived tropospheric columns of NO2 are realistically distributed into two main hotspots, one over the downtown Oslo area and another slightly weaker one over the Drammen areas to the southwest of Oslo.  Initial validation experiments against station observations indicate that the SAMIRA downscaling methodology produces significantly more accurate estimates of local air quality than traditional methods.