Comparison of aerosol optical depth from satellite (MODIS), sun photometer and broadband pyrheliometer ground-based observations in Cuba

In the present study, we report the first comparison between the aerosol optical depth (AOD) and angstrom ngstrom exponent (AE) of the Moderate Resolution Imaging Spectroradiometer (MODIS) instruments on the Terra (AOD(t)) and Aqua (AOD(a)) satellites and those measured using a sun photometer (AOD(SP)) at Camaguey, Cuba, for the period 2008 to 2014. The comparison of Terra and Aqua data includes AOD derived with both deep blue (DB) and dark target (DT) algorithms from MODIS Collection 6. Combined Terra and Aqua (AOD(ta)) data were also considered. Assuming an interval of +/- 30 min around the overpass time and an area of 25 km around the sun photometer site, two coincidence criteria were considered: individual pairs of observations and both spatial and temporal mean values, which we call collocated daily means. The usual statistics (root mean square error, RMSE; mean absolute error, MAE; median bias, BIAS), together with linear regression analysis, are used for this comparison. Results show very similar values for both coincidence criteria: the DT algorithm generally displays better statistics and higher homogeneity than the DB algorithm in the behaviour of AOD(t), AOD(a), AOD(ta) compared to AOD(SP). For collocated daily means, (a) RMSEs of 0.060 and 0.062 were obtained for Terra and Aqua with the DT algorithm and 0.084 and 0.065 for the DB algorithm, (b) MAE follows the same patterns, (c) BIAS for both Terra and Aqua presents positive and negative values but its absolute values are lower for the DT algorithm; (d) combined AOD(ta) data also give lower values of these three statistical indicators for the DT algorithm; (e) both algorithms present good correlations for comparing AOD(t) , AOD(a), AOD(ta) vs. AOD(SP), with a slight overestimation of satellite data compared to AOD(SP), (f). The DT algorithm yields better figures with slopes of 0.96 (Terra), 0.96 (Aqua) and 0.96 (Terra C Aqua) compared to the DB algorithm (1.07, 0.90, 0.99), which displays greater variability. Multi-annual monthly means of AOD(ta) establish a first climatology that is more comparable to that given by the sun photometer and their statistical evaluation reveals better agreement with AOD(SP) for the DT algorithm. Results of the AE comparison showed similar results to those reported in the literature concerning the two algorithms' capacity for retrieval. A comparison between broadband aerosol optical depth (BAOD), derived from broadband pyrheliometer observations at the Camaguey site and three other meteorological stations in Cuba, and AOD observations from MODIS on board Terra and Aqua show a poor correlation with slopes below 0.4 for both algorithms. Aqua (Terra) showed RMSE values of 0.073 (0.080) and 0.088 (0.087) for the DB and DT algorithms. As expected, RMSE values are higher than those from the MODIS-sun photometer comparison, but within the same order of magnitude. Results from the BAOD derived from solar radiation measurements demonstrate its reliability in describing climatological AOD series estimates.