AEROSOL RETRIEVAL AND ATMOSPHERIC CORRECTION FOR MERIS DATA OVER LAKES Dana Floricioiu, Helmut Rott Institute of Meteorology and Geophysics, University of Innsbruck, Innrain, A-6 Innsbruck, Austria. Email: dana.floricioiu@uibk.ac.at ABSTRACT One of the objectives of the ENVISAT project AO-6 on Environmental Research in the Eastern Alps is the development of algorithms for retrieval of water quality parameters of lakes from MERIS data. In order to test and validate atmospheric correction algorithms and to provide basic data for the development of algorithms for retrieval of limnological parameters and aerosol loadings, several field campaigns were carried out in summer on the lakes Garda (Italy) and Mondsee (Austria) parallel to MERIS overflights. Field measurements of aerosol optical thickness () were used as input for atmospheric correction by means of the 6S model, and field spectra measured above the water surface were used to validate the at-surface reflectance derived from MERIS data. The agreement between field and MERIS reflectance spectra is in general good. Some differences are found at short wavelengths which can be attributed to insufficient knowledge of aerosol properties. The sensitivity of the radiative transfer model to changes in and the aerosol model was investigated. For a day with strong variability in the aerosol loading the spatial gradient of was estimated from MERIS data and compared with the temporal evolution of at a field measurement site. INTRODUCTION The at-surface reflectance is a key parameter for the retrieval of bio-geophysical parameters from spaceborne imaging spectrometry data. For deriving the at-surface radiance in the visible part of the spectrum, the main source of errors for the correction of the atmospheric effects is the uncertainty about the content and properties of atmospheric aerosols. As followon to preparatory studies [], [] for the development and validation of algorithms to retrieve water quality parameters in perialpine lakes from data of the Medium Resolution Imaging Spectrometer (MERIS), we carried out several field campaigns in summer on the lakes Garda (Italy) and Mondsee (Austria) parallel to MERIS overflights. In situ derived aerosol properties are used as input for the radiative transfer model, while spectroradiometric measurements above the water surface are used for the validation of the atmospheric correction. These investigations are also of relevance for studying algorithms for aerosol retrieval from MERIS data. MERIS DATA AND IN SITU MEASUREMENTS From June to September several field campaigns were carried out at lakes on the northern and southern rim of the Eastern Alps in connection with MERIS acquisitions. In this paper we report on field work, and on the analysis of Level b full resolution (FR) MERIS products from 7 August and August and of a reduced resolution (RR) image from June. The field measurements were carried out on Lake Garda, which covers an area of 6 km, and on Mondsee, covering. km (Fig. ). During the field campaign on Lake Garda on June water samples and reflectance spectra above the water surface were collected along several transects with two spectroradiometers, and SpectraScan. Averaged reflectance spectra were selected for the validation of the atmospheric correction. In addition, for the South basin of Lake Garda reflectance spectra are available for July, measured by IREA-CNR Milano. The spectral atmospheric transmittance was measured at Riva, located at the northern end of Lake Garda (. E,.6 N) at 6 m a.s.l. (site in Fig. ) and on the shore of the lake Mondsee in the Salzkammergut region of Austria ( E, 7 N) at m a.s.l (site ). Proc. MERIS User Workshop, Frascati, Italy, November (ESA SP-9, May )
Fig.. Location of the test areas Lake Garda and Mondsee in the MERIS scene from Aug.. The Precision Filter Radiometer was deployed at sites (Riva del Garda) and (Mondsee). MERIS bands, and are shown in red, green and blue, respectively. RADIOMETRIC MEASUREMENTS OF AEROSOL OPTICAL THICKNESS The atmospheric transmittance was measured with a Precision Filter Radiometer (PFR) [] at wavelengths: 6,, and 6 nm. The sun photometer, mounted on a sun tracker, was deployed in summer on eight days at site and on one day ( August ) at site. Every second minute the PFR starts a measuring cycle. Examples of aerosol optical thickness () at the four wavelengths and Ångström coefficients derived from PFR measurements on three cloudfree days are shown in Fig.. The measurements in Riva reflect the typical thermally induced daily wind circulation on Lake Garda and the Sarca valley in summer, when the wind direction changes from down-valley (north to south) to up-valley (south to north, called Òra) in the late morning due to differential heating of the atmosphere. In the morning of June very clear air conditions were observed at Riva, where low was measured at the wavelengths (Fig. a). After UTC the visibility towards south significantly degraded, leading to a strong increase in and Ångström turbidity coefficient due to advection of polluted air from the Po plain along the lake. The higher turbidity content of the air in the Po valley is also evident in the MERIS image (Fig. ), where the colors and features are more diffuse in the plain compared to the alpine area. The temporal change in observed at Riva suggests a spatial variation in the aerosol loading along the lake at the time of the MERIS overflight. In the morning of 7 August the visibility was lower than on June and a similar but less pronounced increase in the was observed during the day (Fig. b). Òra started at about 9: UTC at Riva. At site, Mondsee, stable conditions were observed on August (Fig. c). The Ångström coefficient shows little temporal variation during all of the days shown (Table ).
Site : Riva del Garda, June. 6 nm. nm nm 6 nm... Angström... Angström Site : Riva del Garda, 7 Aug..7.6..... Site : Mondsee, Aug..7.6..... Angström (c ). Angström.. Angström... Angström Fig.. Time series of at four wavelengths and Ångström coefficients derived from measurements with the PFR at the two sites. The MERIS overflights were between 9:7 and 9: UTC. Table. The and the Ångström - coefficients und (mean value and standard deviation of the specified time intervals) from PFR measurements, used for the atmospheric correction of MERIS data. Date Site UTC nm Ångström Ångström MERIS acq. UTC June (Riva) :-:.7.6±..±. 9: : :.9.6±.69.7±. 7 Aug. (Riva) 9:-:.6.77±..9±. 9:7 Aug. (Mondsee) 9:-:..69±.7.±. 9: RADIATIVE TRANSFER MODELING AND VALIDATION Because the concentrations of suspended sediments and of cholorophyll-a (Chl-a) are low in the observed lakes, the atsurface reflectance is low and the radiance measured at satellite altitude is dominated by the atmospheric contribution over a large part of the spectrum, as evident by comparing top of the atmosphere (TOA) reflectance and in situ measured spectra (Fig. ). These are very demanding requirements for atmospheric correction. The atmospheric correction procedure applied to the MERIS Lb data to calculate the surface reflectance is based on the 6S radiative transfer model []. The illumination and view parameters for the test areas are taken from the annotation data files delivered with the MERIS product. The troposphere and lower stratosphere (up to 7 - km) were described by radiosonde profiles of temperature, pressure and air humidity from the stations Milano and Munich for site and, respectively. The mid latitude summer atmospheric model was used for altitudes above those reached by the radiosonde. In 6S the aerosol properties are parameterized through at nm ( ), derived from in situ PFR measurements (Table ). 6S enables the selection of various aerosol size distributions. This capability is useful for testing effects of the size distribution on the atmospheric signal. For the narrow northern part of lake Garda and for Mondsee the MERIS
reflectance was corrected for adjacency effects with the method given in []. The size of the southern basin of Lake Garda makes it possible to select an area in the MERIS image where these effects could be avoided. The atmospheric correction was applied to all MERIS bands except the strong oxygen absorption band at 76 nm which, together with the water vapour absorption band at 9 nm (band ), is devoted to retrieve gaseous abundances and not surface reflectances [6]. 6 Lake Garda South June TOA Reflectance SpectraScan 6 7 9 Mondsee Aug. TOA Reflectance 6 6 7 9 Fig.. Top of the atmosphere (TOA) reflectance from MERIS Lb data compared to in situ measured spectra over an area in the south of Lake Garda on June and Mondsee on Aug.. For calculating the at-surface reflectance from MERIS FR Lb data over Mondsee (Fig. a) three different values of were used to study the sensitivity of the atmospheric correction to aerosol optical thickness. For these calculations the continental aerosol model was used. For a % variation of around the measured value ( =.) the change in reflectance is.6% or less at any wavelength. The effects of the aerosol model on the radiative transfer calculations were also studied. Among the aerosol models available in 6S the Junge power law with a Junge parameter, ν= +, derived from PFR measurements gives the best agreement between MERIS and field spectra (Fig. b). In addition to the measured wavelength exponent the following parameters of the Junge distribution were selected: minimum and maximum radii of the aerosol,.9 µm and.7 µm, respectively, and the index of refraction, n=. - i. The calculations show differences for the surface reflectance if different aerosol models are employed, in particular in the lower and upper wavelength sections of the observed spectral range. Though corrections for adjacency effects were applied, the slight overestimation of MERISderived reflectance in the near infrared might be an indication that these effects were not fully compensated as the lake is rather small. Mondsee Aug. 9 =. Mondsee Aug. Junge ν=.7 =. 9 Maritime =. Continental 7 7 6 6 7 9 6 6 7 9 Fig. At-surface reflectance spectra over Mondsee derived from MERIS FR Lb data with 6S and measured in situ. Reflectance retrieved with different values of, to study the sensitivity of the atmospheric correction to variations of. =. was measured with PFR. The continental aerosol model was used. Reflectance with different aerosol size distributions, to study effects of the aerosol model for atmospheric corrections.
For Lake Garda atmospheric radiative transfer calculations were carried out with the MERIS RR data from June. From the time sequence of measurements in Riva and the visual observations on the lake it can be concluded that during the MERIS acquisition the aerosol content was significantly higher in the southern part of the lake than in the northern part. In the northern section of the lake the field measurements and MERIS-retrieved at-surface spectra agree quite well at wavelengths nm if =.7 (the mean value measured in the morning at site ; Table ) and the Junge aerosol model with the measured wavelength exponent are used (Fig. a). For the southern part of the lake forward calculations were carried out with different values of to fit the fieldmeasured and MERIS-derived reflectance spectrum. Good agreement was obtained for =. which is close to the value measured in the early afternoon hours at Riva after increase of aerosol content due to advection from the south. Lake Garda North June Lake Garda South June =.7 Junge ν=. =. Junge ν=. =.7 Continental SpectraScan 6 7 9 6 7 9 Fig.. At-surface reflectance spectra over Lake Garda derived from MERIS RR Lb data of June and measured in situ. Area in the northern part with the assumption of two different aerosol models. Area in the southern part of the lake for which the was estimated by means of forward modeling. On 7 August the aerosol content was higher and the increase of at Riva during the day was less pronounced than on June. Therefore the same aerosol properties derived from PFR measurements (Table ) were used for the atmospheric correction all over the lake. Since no field spectra were measured on that day we used spectra acquired on other dates for validation. In the northern part of Lake Garda no significant changes in the optical properties of the water are expected during summer. Therefore the field spectrum acquired on June was taken as reference (Fig. 6a). For the southern part SpectraScan data acquired on July were taken as reference (Fig. 6b). Lake Garda North 7 Aug. Lake Garda South 7 Aug. =.6 Junge ν=.7 =.6 Junge ν=.7.6. SpectraScan.7. 6 7 9 6 7 9 Fig. 6. At-surface reflectance spectra over Lake Garda derived from MERIS FR Lb data of 7 Aug. and measured in situ on June and July, respectively. Area in the northern part. Area in the southern part of the lake. The reflectance of Lake Garda is in general very low due to the low concentrations of Chl-a (. µgl - ) and organic suspended matter ( mgl - ), while at Mondsee in August the concentrations were up to. µgl - and. mgl -, for
Chl-a and organic suspended matter, respectively. In June the reflectance values along the transects on the lake were below % throughout the observed spectra. Later in the summer some increase of reflectance was observed in the southern part of the lake. The differences between MERIS-derived and field spectra are below % at any wavelength. The main differences are found in the blue/green region, where the contribution of the atmospheric path radiance is largest. This has to be expected because the size distribution and absorption properties of the aerosols are not known, and the aerosol model is important for calculations of the atmospheric radiation fluxes. CONCLUSIONS Because of the low reflectance of perialpine lakes the requirements for deriving at-surface reflectance for retrieval of limnological parameters from satellite measurements are very demanding. Case studies, using in situ measured aerosol optical thickness and radiosonde data as input, showed good capabilities of the 6S model for the correction of atmospheric effects. Comparisons between in situ spectra and MERIS-derived at-surface reflectance revealed the importance of the aerosol model for atmospheric correction. Further investigations on the radiometric effects of aerosol properties are needed for optimizing the atmospheric correction over inland water and for improving aerosol retrievals over land surfaces from satellite spectrometric data. The analysis of a transect along Lake Garda with a spatial gradient of the aerosol loading suggests good capability for retrieval of from MERIS data over dark surfaces with known reflectance. Aknowledgements This work was supported by the Austrian Space Program of the Federal Ministry for Transportation, Innovation and Technology (BM:VIT) and Austrian Space Agency (ASA). The MERIS data were made available through the ESA AO project 6. We thank Dr. E. Zilioli from IREA-CNR Milano and Dr. C. Defrancesco from APPA Riva for supporting the field campaigns on Lake Garda and Prof. T. Weisse and Prof. M. Dokulil from the Institute for Limnology at Mondsee for supporting the field campaign on Mondsee. 6 REFERENCES. Floricioiu D., Rott E., Riedl C. and Rott H., Retrieval of water quality parameters of perialpine lakes in Austria by means of hyperspectral ROSIS data, Proc. of the First International Symposium on Recent Advances in Quantitative Remote Sensing, Spain, 6- Sept., Universitat de Valencia, pp. 6-.. Floricioiu D., Riedl C., Rott H. and Rott E., Envisat MERIS capabilities for monitoring the water quality of perialpine lakes, Proc. of IEEE IGARSS, - July, Toulouse, France.. Roth H. and Wehrli C., PFR-Precision Filter Radiometer Documentation version., prepared at PMOD/WRC Davos, 999.. Vermote E., Tanré D., Deuzé J.L., Herman M. and Morcrette J.J., Second Simulation of the Satellite Signal in the Solar Spectrum, 6S: An Overview, IEEE TGRS, (), 67 66, 997.. Richter R., Correction of atmospheric and topographic effects for high spatial resolution satellite imagery, Int. J. Remote Sensing, (), 99, 997. 6. Santer R., Carrère V., Dessailly D., Dubuisson P. and Roger J.-C., Atmospheric corrections over land, MERIS ATBD., Issue, Dec. 997.