Fig. 1
Frequency (left vertical axis) and PDFs (right vertical axis) of coastal AODs in 2002–2011. Plots are derived from AODs at 62 coastal AERONET sites and collocated MODIS retrievals over those sites. µ is the log-normal location parameter and σ is the log-normal scale parameter, and the mean is the average AOD over the whole time period. The subplots show the frequency of quality-filtered (A) AERONET AODs, (B) MODIS Land_And_Ocean AODs, (C) AERONET AODs only where a paired MODIS AOD from the Dark Land algorithm exists, (D) MODIS AOD from Dark Land algorithm, (E) AERONET AODs only where a paired MODIS AOD from the Dark Ocean algorithm exists and (F) MODIS AODs from Dark Ocean algorithm.
Fig. 2
Frequency (left vertical axis) and PDFs (right vertical axis) of the coastal AODs from (A) AERONET, (B) MODIS Land_And_Ocean, (C) MODIS Dark Ocean algorithm and (D) MODIS Dark Land algorithm. All MODIS AODs were filtered with quality flag for the span of 2002–2011. The p-values indicate statistical significance of fit between frequency distributions derived from the lognormal PDFs (with corresponding parameters shown in Fig. 1) and actual frequency distribution (e.g. the bars in red). See text for details. (A) Shows only those AERONET AODs, that correspond to valid MODIS AOD retrievals.
Fig. 3
Scatter plot of AERONET AOD (x-axis) and the quality flag filtered MODIS AOD (y-axis) for 2002–2011. In (A), (B) and (C), AODs in y-axis are respectively derived from MODIS Land_And_Ocean, Land and Ocean products over the non-coastal AERONET stations. (D), (E) and (F) are respectively the same as (A), (B) and (C) but over the coastal AERONET stations. In each scatter plot, also shown is the coefficient of determination (R 2), mean bias, the number of MODIS-AERONET collocated data points (N), the density of points (color bar), the best-fit linear regression equation (solid black line), the 1:1 line (dashed black line), and the expected error (EE) envelope (red dashed line) for MODIS AOD explained in Section 3.2.
Table 1. Regression statistics for the MODIS AOD products with respect to AERONET
Table 2. MODIS AOD mean bias for all AERONET coastal stations
[i] 62 coastal AERONET sites were identified and the results are an average of all the sites. Each of the MODIS aerosol algorithms is shown with the recommended quality filtering except for the Land_And_Ocean dataset, which is shown without any quality filtering (default MODIS dataset) and the results of our quality filtering technique described in Section 4. Bias results are separated into Low AOD and High AOD events as classified based on the AERONET measurements with the cutoff at AOD = 0.25.
Fig. 4
Frequency (left y-axis) and PDFs (right y-axis) of MODIS AOD biases from the (A) Dark Land algorithm, (B) Land_And_Ocean dataset, (C) Dark Ocean algorithm and (D) the corrected Dark Ocean algorithm. The pictured data are for all coastal AERONET sites in 2002–2011. The thin black and green lines respectively show zero bias and mean bias for each panel. (E) Regression of bias at 550 nm in MODIS AOD at 550 nm for Dark Land and (F) for Dark Ocean.
Fig. 5
(A) Map of the location of all 62 coastal AERONET sites analysed in the study. Also shown are the maps of MODIS AOD bias (with respect to AERONET AOD) at each of these coastal sites respectively for: (B) MODIS Land AODs dataset filtered with quality flag, (C) MODIS Ocean AODs dataset filtered with quality flag; (D) MODIS Land_And_Ocean AODs without any quality filtering; (E) MODIS Land_And_Ocean AODs after using the quality filtering method described in Section 4. Bias calculations are based on ~9 years (2002–2011) of collocated MODIS and AERONET AOD data. Blue indicates an underestimation (e.g. negative bias) in MODIS AOD and red indicates overestimation (positive bias). Common legend for (B)–(E) is shown on the left of (A).
Fig. 6
Taylor diagram comparing ~2002–2011 quality-flag-filtered MODIS AOD retrievals and AERONET AOD observations. Coastal MODIS AOD retrievals are listed with a 1 and Non-Coastal AODs are shown with a 2. The MODIS Ocean, Land, Land_And_Ocean and empirically corrected Ocean (Section 5) AODs are represented by blue, red, green and purple, respectively. The arrow represents the effect of the empirical correction on the MODIS Ocean dataset.
Fig. 7
Cumulative distribution functions (CDF) of AOD derived from AERONET (black), and corresponding paired MODIS AODs derived respectively from MODIS Land (red), Ocean (blue) and Land_And_Ocean (green) AODs after filtering with quality flag. Maximum differences (Δmax) between the AERONET CDF and MODIS CDFs are shown by the two dashed horizontal lines and their values are denoted by the labels in their respective colors. Statistics are based upon MODIS aerosol observations in 2002–2011 over the coastal regions. Critical values for the K–S test are also denoted in the top left of the figure and are described in the text (Section 3.1).
Table 3. MODIS AOD bias with respect to AERONET AOD for both coastal and open-ocean sites
[i] The bias is listed for three categories based on how MODIS AOD is used in the evaluation. The first is the filtering of the data using quality control flag; the second builds upon the first but also removes MODIS AOD data with cloud fraction larger than 80%; the third is the same as the second except the threshold for cloud fraction is now decreased to 70%. The number of AOD retrievals used in the different analyses (last row) is also shown to display the reduction in the data volume associated with each category. In each category, the bias is further analysed in terms of low AOD conditions (AOD < 0.25) and high AOD conditions. In addition, the relative change of the bias due to the filtering of data with cloud fraction is shown in parentheses, negative percentages indicate an increase in bias. See the text for further details.
Fig. 8
(A) Scatter plot of horizontal wind speeds that are 2 meters above the surface from MERRA (x-axis) and the biases in the quality-flag-filtered AODs. (B) The frequency (left y-axis) and CDF (right y-axis) of coastal wind speeds during MODIS overpass times. (C): same as (A) but for the bias of MODIS AOD after 70% cloud fraction filter. (D) Scatter plot of the wind speed and cloud fraction pairs for each AOD retrieval from MODIS Dark Ocean algorithm. The analysis is for all coastal sites (62 AERONET sites) and for the years ~2002–2011. R is the Pearson linear correlation coefficient (R), N is the number of retrievals and Y is the regression equation. In (A) and (C), red dots show the MODIS biases binned to 1 ms−1 intervals, and their corresponding regression lines and correlation coefficient are denoted in red as well; the blue dotted line is a reference of zero bias.
Fig. 9
(A) Location of each coastal AERONET site, (B) the correlation between sea-surface wind speed and the biases in quality-flag-filtered AODs from MODIS Dark Ocean algorithm, (C) and (D) the y-intercept and slope in the linear regression equation between the MODIS AOD bias and wind speed. Blue color represents statistically significant values in (B) and negative intercepts and slopes in (C) and (D), respectively. Red represents statistically insignificant values in (B) and positive intercepts and slopes in (C) and (D). Magnitude scales are shown by size of the circles, and are provided in each panel for clarity. (C) and (D) show only sites with p-value ≤ 0.05 (46 out of the possible 62 sites). Results are for the date record of 2002–2011.
Fig. 10
Frequency distribution of quality assured (A) AERONET AOD over coastal regions that have an MODIS Dark Ocean algorithm collocated retrieval, (B) AOD from MODIS Dark Ocean algorithm after cloud fraction and quality flag filtering only, (C) AOD from MODIS Dark Ocean algorithm after cloud fraction filtering (70%), wind speed bias correction and quality flag filtering. (D): same as Fig. 7 except the cumulative distribution function (CDF) is derived from the frequency distributions, respectively, in (A)–(C) and Fig. 2c, and shown correspondingly as black, red and blue color, respectively.
Fig. 11
Scatter plot of the bias for coastal MODIS AOD (retrieved by the Dark Ocean algorithm) as a function of the Rrs (Sr−1) at 645 nm. Also shown is the best linear fit equation, the statistical significance (p-value) of the fit and number of data points (N).
Fig. 12
(A) Spatial distribution of the trend of annual AOD at different AERONET sites that have at least 6 yr of data during 2002–2010. Blue indicates negative AOD trends while red indicates positive AOD trends. The size of the circle is relatively proportional to the absolute value of the trend. (B) The relative difference (in%) between annual AOD trends computed with MODIS before and after the empirical correction, defined as the (∣Trend_modis_corrected − Trend_aeronet∣ − ∣Trend_modis − Trend_aeronet∣)/∣Trend_aeronet∣; negative value is shown in blue, and indicates that Trend_modis_corrected is closer to Trend_aeronet than Trend_modis; positive value is shown in red, and indicates that Trend_modis_corrected is further away from Trend_aeronet than Trend_modis. See Section 5 for details.