Dealing with spatial heterogeneity in pointwise-to-gridded- data comparisons

Abstract. Most studies on validation of satellite trace gas retrievals or atmospheric chemical transport models assume that pointwise measurements, which roughly represent the element space, should compare well with (model) pixels (grid box). This assumption implies field interest must possess a high degree spatial homogeneity within box), may not hold true for species short lifetimes in proximity plumes. Results this often lead to perception nonphysical discrepancy between data, resulting from different scales, potentially making comparisons prone overinterpretation. Semivariogram is mathematical expression variability discrete data. Modeling semivariogram behavior permits carrying out optimal linear prediction random process using kriging. Kriging can extract information (variance) pertaining specific scale, turn translates data gridded space quantified uncertainty such grid-to-grid comparison be made. Here, both theoretical and real-world experiments, we demonstrate classical geostatistical approach adapted solving problems evaluating model-predicted satellite-derived gases. study suggests procedures present method take kriging variance response functions into account. We Ozone Monitoring Instrument (OMI) tropospheric NO2 columns against 11 Pandora spectrometer instrument (PSI) systems during DISCOVER-AQ campaign over Houston. The least-squares fit paired shows low slope (OMI=0.76×PSI+1.18×1015 molecules cm?2, r2=0.66), indicative varying biases OMI. perceived slope, induced by problem disappears convolved kriged PSI OMI (0.96×PSI+0.66×1015 r2=0.72), illustrating possibly has constant systematic bias area. To avoid gross errors made vs. argue concept (or autocorrelation) taken consideration, particularly if exhibits strong heterogeneity at scale and/or model footprints.