A COMPARATIVE STUDY OF AEROSOL OPTICAL DEPTH OVER SINGAPORE FROM : RESULTS FROM AERONET AND MODIS DATA ANALYSIS.

Transcription

1 A COMPARATIVE STUDY OF AEROSOL OPTICAL DEPTH OVER SINGAPORE FROM : RESULTS FROM AERONET AND MODIS DATA ANALYSIS. Santo V. Salinas1, Astrid Muller2 and Tan Li1 Centre for Remote Imaging Sensing and Processing (CRISP), National University of Singapore, Singapore, 2 Institute of Low Temperature Science, Hokkaido University, Hokkaido, Japan muller@pop.lowtem.hokudai.ac.jp 1 KEY WORDS: Aerosols, AERONET, MODIS ABSTRACT: South East Asia (SEA) is heavily dependent on satellite remote sensing given its vast geographical distribution as well as its complex surface, sea and atmospheric environment. In recent years, trans-boundary smoke, mainly caused by forest and agriculture burning in Sumatra and Borneo has become a recurring event affecting the SEA region. Haze/smoke events are occurring particularly between the dry months of August and October of every year. In a previous work, we examined aerosol optical depth (AOD) trend over Singapore over a period of eight years spanning from AOD values, at 550nm, were obtained from our existing AERONET site. Smoke/non-smoke periods were analyzed and a multi-year trend showed a positive slope increase of average monthly AOD. In this work, and as a complement to our previous results, we have added AOD retrievals from the Moderate Resolution Imaging Spectro-radiometer (MODIS) sensor on board NASA s Terra and Aqua satellites for the same period of time. Monthly and seasonal variations of AOD are investigated and reanalyzed concurrently for both AERONET and MODIS data sets. In this study, we try to answer two questions; Does Terra/Aqua MODIS AOD values exhibit similar temporal variation patterns as those inferred from AERONET and whether there are common inter-annual or seasonal patterns noticeable in the AOD time series over the last 8 years. Monthly averaged level 3 MODIS gridded atmosphere global joint products and AERONET level 1.5/2.0 data sets (further processed for cloud contamination by the Spectral Deconvolution Algorithm) are exclusively used for this work. 1. INTRODUCTION Aerosols or small particulates suspended in the atmosphere, are major contributors to air pollution as well as to climate due to their interaction with solar radiation and by modifying the micro physical properties of clouds. Space borne instruments, specially those tracking environmental parameters i.e. AVHRR, MISR, MODIS, SeaWifS etc. have measured the so called aerosol optical depth (AOD) for several years. Since year 2000 for example, MODIS have yielded a wealth of reliable AOD data products of hight quality in both, area coverage and time scale. Trend estimates obtained from satellites are important to asses and investigate the impact of aerosols particles on climate and air quality. However, satellite remote sensing has its own uncertainties that can be attributed to complex surfaces, limited number of satellite passes, cloud contamination and the aerosol models used in the retrieval process etc.. A more accurate way of retrieving aerosol concentrations or AOD is via distributed networks of SunPhotometers such as AERONET(Holben et al, 1998). Photometric measurements are usually more reliable and accurate than satellite retrievals but they lack the spatial coverage that satellite observations can provide. For regions in which cloud cover is a severe hindrance, photometric observations can be used as a reliable tool not only to obtain aerosol concentrations but to validate satellite observations. Singapore being located at the Southern tip of the Malaysian peninsula, is continuously affected by episodes of biomass burning generated by land clearance in the near-by regions of Sumatra and Borneo. The yearly transition between the wet periods of the North-East monsoon (December March) and the dry period generated by the South-West monsoon (June September) has in impact on the transport pattern of aerosol pollutants from source areas such as Sumatra and to a less extend Borneo, to receptor sites such as Singapore for example and mostly during the South-West monsoon period. Previous studies of the local and regional aerosol environment have focused on the impact of smoke generated by widespread biomass burning. Several of these studies focus on aerosol properties and transport patterns specially during the so called dry season between August to October (e.g. Atwoood et. Al, 2013; Chew et al. 2013, Reid et al, 2013, Salinas et al, 2009 and 2013). In this work, we focus on the long time trends of aerosol optical depth by analyzing 8 years of MODIS (Aqua and Terra) and AERONET AOD time-series. Our focus is on whether both platforms can detect a similar inter-annual trend and at the same time investigate to what degree does the yearly occurring biomass burnings event affect this trend. For this work, we have used cloud screened and quality assured level 2.0 AERONET data as well as monthly averages level 3 MODIS gridded atmosphere global joint products.

2 2. INSTRUMENTATION, DATA PROCESSING AND METHODOLOGY The National University of Singapore host an AERONET linked sun-photometer since This instrument performs daily direct Sun measurements which are used to obtain AOD and other related by-products such as the Angstrom exponent number. Inversion products such as particle size distribution and refractive indexes, single scattering albedo, scattering phase function are also generated. For this study, we used level 2.0 cloud screened and quality assured photometric data. Months with less than 30 data points and less than 8 days of measurement within a month are classified as invalid sample and are excluded from the further analysis. In the case of MODIS, level 3 gridded atmosphere global joint products were used to obtain time series of area averaged values of AOD at 1X1 degree resolution ( a 2 x 1 degree box, enclosing Singapore was used for our trend analysis). Regular monthly mean AOD and deseasonalized (dry season months of AUG-OCT removed) monthly mean AOD was used for trend analysis comparison. Furthermore, to differentiate the influence of inter-seasonal variability, a boxcar averaging with a window width of 3 month is applied to the whole data set (Zhang et al., 2010). This means that a monthly data point is replaced by a 3 month average of this month and the two neighboring month. In order to generate interannual trends, a linear regression is applied to yearly time series of AOD for both AERONET and MODIS (Aqua and Terra). The significance of a non-zero regression coefficient is tested by application of a t-test of confidence. 3. TREND ANALYSIS 3.1 AERONET time-series Fig. 1: All seasons AERONET data trend from January 2007 to Data corresponds to AERONET level 2.0 product. In Fig. 1(a), we show the AERONET time series from January 2007 to October 2014 and including all months of the year. The data series correspond to fully cloud screened and quality assured level 2.0 data. The standard trend (black line) shows a positive AOD slope trend of 1.37x10^(-2)/year. The standard error is of the order of 7.04x10^(3) with a near 90% confidence level (p-value ~ 0.054). High AOD peaks are observed during third quarter of 2009, first and third quarter of 2010, third quarter 2012 and Applying box-car averaging, the trend curve is smoothed somewhat, specially for the 3rd quarter 2009, 2010 and In this case, the trend slope increases slightly to 1.41^(-2)/year and the standard error decreases to 4.83x10^(-3) at a 95% confidence level or better. Obviously, in both cases the inter-seasonal trend (the elevated AOD peaks) remains the driving factor even when box-car averaging is applied. This indicates that, the slope trend is strongly influenced by episodes in which AOD is high such as when events of severe smoke pollution occurs. Such a event are more likely to occur during the last two quarters of each year as this coincides with a dry seasonal period of the South-West monsoon. In order to reduce the influence of severe smoke pollution episodes on the inter-annual trend, we use the so called deseasonalized trend in which the influential months of August to October (fire season) are removed from the annual time series. The resulting trend and its corresponding box-car averaging are shown in Fig.(2). In this plot,

3 the standard trend slope is relatively smaller (6.42x10^(-3)) when compared with the full seasonal trend of Fig. 1(a). The standard error remains of the same magnitude but the t-test gives a low confidence for this case (p-value=0.25). Box-car smoothing however, provides a better view, with a similar yearly slope of 6.63x10^(-3) and a better than 95% confidence level. Fig. 2: Deseasonalized AERONET data trend from January 2007 to Data corresponds to AERONET level 2.0 product. 3.2 MODIS time-series In this section we focus on the time-series trend obtained from the MODIS instrument on board of both Aqua and Terra satellites for the study period mentioned in Sec AQUA-MODIS Fig. 3: Inter-annual AOD trend for MODIS-AQUA for all seasons during our study period. In fig.(3) we show AQUA-MODIS yearly trend that includes all seasons for the study period above mentioned. The AOD slope remains positive for both, the standard trend and the box-car averaged trend i.e. approx. 1.1x10^(2)/year. Similarly, the standard error are of the same order of magnitude for both trends and the degree of confidence is high (p-value < 0.02). This indicates that although the box car averaging helps on reducing the standard error, the seasonal spikes of high AOD values dominates the trend and there is no noticeable smoothing gained by performing box-car averaging. However, the most noticeable aspect of the Aqua data is that it shows elevated AOD activity during 2013, specially during the 2nd quarter time frame for which the June 2013 severe

4 smoke episode was recorded across this region. Such activity is not found in the case of the AERONET trend (Fig. 1, 2). This is due to the fact that we used AERONET lev. 2.0 data sets which is generated from a stringent cloud screening algorithm that sometimes removes episodes of heavy smoke/aerosols by confusing it with clouds. Fig. 4: Similar to Fig.(3) but excluding months of August to October of each year. In a similar fashion, Fig.(4) shows AQUA-MODIS yearly trend but this time with the period of August to October removed for each year of the study period ( See Sec. 3.1). In this deseasonalized trend, we observe that the AOD slope is weaker for both the standard and box-car averaged trends (approx. 7.2x10^(-3)/year). In this case, box averaging does help on halving the standard error in the trend as it smooths out some remaining AOD peaks most noticeably at the beginning of 2008 and also on the middle 2013 which is related to the June 2013 smoke episode. The confidence levels for both trends are relatively good however (p-values ~ 0.09) TERRA-MODIS Fig. 5: Inter-annual AOD trend for MODIS-AQUA for all seasons during our study period. As shown in Sec. 3.21, here in we present the inter-annual AOD trend obtained from the MODIS instrument on board the Terra satellite. For Terra, the AOD trend slope is of similar magnitude to what we obtained for Aqua i.e. 8.78x10^(-3)/year for the standard and 9.03x10^(-3)/year for box-car averaged trend. The standard errors are of similar magnitude for both curves although the confidence levels for the box-car averaged trend is higher than 90%. Box-car averaging has certainly reduced the impact of seasonal spikes of high AOD, most notable during the third quarter of 2009, 2010, 2011 and throughout all 2013 to early As noted in Sec. 3.21, the Terra data also shows elevated AOD activity during 2013, specially during the 2 nd quarter of that year. Furthermore, in Fig. 6 we

5 show the deseasonalized Terra-MODIS data trend. In this case, the AOD slope is slightly smaller than those found in the all-season trend with a value of 4.16x10^(-3)/year for the standard trend and 4.63x10^(-3) for the box-car averaged trend. The box-car averaging reduces the standard error of the trend by about half but both standard error on the trend are comparatively similar tho those found in the full seasonal trend shown in Sec Moreover, for the deseasonalized trend, the confidence levels are low as seen from their respective p-values. Fig. 6: Similar to Fig.(5) but excluding months of August to October of each year. In Table 1. we summarize all the AOD slope and standard error values derived from all the remotes sensing platforms we used in this work. These values correspond to the box-car averaging trend curve only as this type of trend smoothing clearly helps on removing some of the spikes observed in the inter-annual values of AOD. AERONET All Seasons Deseasonalized AQUA-MODIS TERRA-MODIS AOD slope=1.41x10^(-2) AOD slope=1.11x10^(-2) AOD slope=9.03x10^(-3) Std. Err=4.83x10^(-3) Std. Err=3.62x10^(-3) Std. Err=3.20x10^(-3) AOD slope=6.63x10^(-3) AOD slope=7.61x10^(-3) AOD slope=4.63x10^(-3) Std. Err=3.22x10^(-3) Std. Err=2.71x10^(-3) Std. Err=2.10x10^(-3) Table 1: Yearly AOD slope and Standard Error after box-car averaging for all platforms we studied. 4. SUMMARY AND CONCLUSIONS In this work we have presented a small but detailed study of the inter-annual AOD trend as observed by ground based platforms and satellite remote sensing platforms. Systematic trend differences exist over the study period and a strong seasonal variations are observed in both,in-situ ground data (AERONET) and satellite AOD retrievals. Seasonal variations were removed by excluding periods of high aerosol loading as well as by performing a simple box-car averaging with a three month window. For the standard trend i.e. including emissions from cyclical fire periods, both AERONET and AQUA-MODIS show similar inter-annual slope trend values. However, for TERRAMODIS the AOD trend slope is comparatively smaller by about 60%. Furthermore, and after removing months of high fire activity (August to October of each year), AERONET, AQUA and TERRA -MODIS all show a similar positive slope of AOD increase. Such is slope is comparable to other global/regional trend studies (e.g. [2]) which have shown (after removing biases) positive AOD trend of 8x10^(-3) for the case of MODIS. However, a word of caution must be said here as several artifacts could affect the observed trend, for example, micro-physical bias, cloud contamination, sampling bias, calibration drift among others.

6 5. REFERENCES Atwood, S. A., Reid, J. S., Kreidenweiss, S. M., Yu, L. E., Salinas, S. V., Chew, B.- N., Balasubramanian, B., Examination of Singapore as a receptor site for smoke in the Maritime Continent during the 2009 El Nin o fire season. Atmos. Env. 2013, doi: /j.atmosenv Boon Ning Chew, James R. Campbell, Santo V. Salinas, Chew Wai Chang, Jeffrey S. Reid, Ellsworth J. Welton, Brent N. Holben, Soo Chin Liew, Aerosol particle vertical distributions and optical properties over Singapore, Atmospheric Environment (2013), B. N. Holben, T. F. Eck, I. Slutsker, D. Tanre, J. P. Buis, A. Setzer, E. Vermote, J. A. Reagan, Y. A. Kaufman, T. Nakajima, F. Lavenu, I. Jankonwiak, and A. Smirnov, Aeronet - a federated instrument network and data archive for aerosol characterization, Rem. Sens. Environ., vol. 66, pp. 1 16, Jeffrey S. Reid, Edward J. Hyer, Randall Johnson, Brent N. Holben, Robert J. Yokelson, Jianglong Zhang, James R. Campbell, Sundar A. Christopher, Larry Di Girolamo, Louis Giglio, Robert E. Holz, Courtney Kearney, Jukka Miettinen, Elizabeth A. Reid, F. Joseph Turk, Jun Wang, Peng Xian, Guangyu Zhao, Rajasekhar Balasubramanian, Boon Ning Chew, Serm Janai, Nofel Lagrosas, Puji Lestari, Neng-Huei Lin, Mastura Mahmud, Anh X. Nguyen, Bethany Norris, Nguyen T.K. Oahn, Min Oo, Santo V. Salinas, E. Judd Welton, Soo Chin Liew, Observing and understanding the Southeast Asian aerosol system by remote sensing: An initial review and analysis for the Seven Southeast Asian Studies (7SEAS) program. Atmos. Res. 2012, doi: /j.atmosres Salinas, S. V., Chew, B.N., Liew, S.C., Retrievals of aerosol optical depth and Angstrom exponent from ground-based Sun-photometer data of Singapore, Applied Optics 48(8), , March (doi: /ao ) Salinas, S. V., Chew, B.N., Miettinen, J., ET AL., Physical and optical characteristics of the October 2010 haze event over Singapore: A photometric and lidar analysis, Atmospheric Research 122, , Mar (First published online 7 June 2012, doi: /j.atmosres Zhang, J. and Reid, J.-S., A decadal regional and global trend analysis of aerosol optical depth using a dataassimilation grade over water MODIS and Level 2 MISR aerosol products, Atmos. Chem. Phys., 10, , 2010.