Image-based Atmospheric Correction Research Articles

Remote Retrieval of Suspended Particulate Matter in Inland Waters: Image-Based or Physical Atmospheric Correction Models?

The objective of this paper was to compare the limits of three image-based atmospheric correction models (top of the atmosphere (ToA), dark object subtraction (DOS), and cosine of the sun zenith angle (COST)), and three physical models (atmospheric correction for flat terrain (ATCOR), fast line-of-sight atmospheric analysis of spectral hypercubes (FLAASH)), and ACOLITE) for retrieving suspended particulate matter (SPM) concentrations in inland water bodies using Landsat imagery. For SPM concentration estimates, all possible combinations of 2-band normalized ratios (2bNR) were computed, and a stepwise regression was applied. The correlation analysis allowed highlighting that the red/blue 2bNR was the best spectral index to retrieve SPM concentrations in the case of image-based models, while the red/green 2bNR was the best in the case of physical models. Contrary to expectations, image-based atmospheric models outperformed the accuracy of physical models. The cross-validation results underlined the good performance of the DOS and COST models, with R2 > 0.83, NASH-criterion (Nash) > 0.83, bias = −0.01 mg/L, and RMSE < 0.27 mg/L. This outperformance was confirmed using blind test validation data, with an R2 > 0.86 and Nash > 0.58 for the DOS and COST models. The challenges and limitations involved in the remote monitoring of SPM spatial distribution in turbid productive waters using satellite data are discussed at the end of the paper.

Assessing the Potential of Geostationary Himawari-8 for Mapping Surface Total Suspended Solids and Its Diurnal Changes

Ocean color sensors, typically installed on polar-orbiting satellites, have been used to monitor oceanic processes for last three decades. However, their temporal resolution is not considered to be adequate for monitoring highly dynamic oceanic processes, especially when considering data gaps due to cloud contamination. The Advanced Himawari Imager (AHI) onboard the Himawari-8, a geostationary satellite operated by the Japan Meteorological Agency (JMA), acquires imagery every 10 min at 500 m to 2000 m spatial resolution. The AHI sensor with three visible, one near-infrared (NIR), and two shortwave-infrared (SWIR) bands displays good potential in monitoring oceanic processes at high temporal resolution. This study investigated and identified an appropriate atmospheric correction method for AHI data; developed a model for Total Suspended Solids (TSS) concentrations estimation using hyperspectral data and in-situ measurements of TSS; validated the model; and assessed its potential to capture diurnal changes using AHI imagery. Two image-based atmospheric correction methods, the NIR-SWIR method and the SWIR method were tested for correcting the AHI data. Then, the new model was applied to the atmospherically corrected AHI data to map TSS and its diurnal changes in the Pearl River Estuary (PRE) and neighboring coastal areas. The results indicated that the SWIR method outperformed the NIR-SWIR method, when compared to in-situ water-leaving reflectance data. The results showed a good agreement between the AHI-derived TSS and in-situ measured data with a coefficient of determination (R²) of 0.85, mean absolute error (MAE) of 3.1 mg/L, a root mean square error (RMSE) of 3.9 mg/L, and average percentage difference (APD) of 30% (TSS range 1–40 mg/L). Moreover, the diurnal variation in the turbidity front, using the Normalized Suspended Material Index (NSMI), showed the capability of AHI data to track diurnal variation in turbidity fronts, due to high TSS concentrations at high temporal frequency. The present study indicates that AHI data with high image capturing frequency can be used to map surface TSS concentrations. These TSS measurements at high frequency are not only important for monitoring the sensitive coastal areas but also for scientific understanding of the spatial and temporal variation of TSS.

Assessment of the image-based atmospheric correction of multispectral satellite images for geological mapping in arid and semi-arid regions

This paper aims to assess the image-based and physical atmospheric corrections of Landsat 8 images for geological mapping. This study was carried out in the central Jebilet inlier in Moroccan anti atlas, which is characterized by its geological diversity, its mining potential and an arid climate. Two physical atmospheric corrections FLAASH (Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes) and ATCOR (Atmospheric & Topographic Correction) were compared to the DOS1 (Dark Object Subtraction) image-based method. The assessment of the results was based on the ASD (Analytical Spectral Devices) spectroradiometric data as well as the SVM (Support Vector Machine) classifications issued from the atmospherically corrected images. The FLAASH provided the most accurate Bottom Of Atmosphere BOA reflectance estimation (R2 = 0.95 and RMSE = 0.033) and slightly outperformed the DOS1 method (R2 = 0.94 and RMSE = 0.034). For the SVM classification outputs, DOS1 gave results almost similar to those obtained by FLAASH and ATCOR. The similarity between the FLAASH and DOS1 SVM classifications was 97.77%. The results of this work demonstrate the effectiveness of the image-based method in atmospheric correction of multispectral data for geological and mineral mapping over the arid and semi-arid regions. The image-based atmospheric correction method is proved as an accurate, simple way and very straightforward to apply.

A simple model for PIFs extraction at digital change detection approach

The radiometric normalization using pseudo-invariant features (PIFs) is one of the best methods of image-based atmospheric correction. However, one of the main challenges of this approach is the correct selection of PIFs. So far many efforts have been made to automate radiometric correction using PIFs, although the capability of all these automated methods has been proven in various studies, their application requires intermediate to advanced statistical and computational knowledge and their application is beyond the reach of most conventional remote sensing users. Therefore, in this research, a straightforward simple method is proposed based on the definition of PIFs that automatically identifies these areas and uses them on a regression procedure for automatic normalization. The study area is Hara Protection Area (the Hara international wetland), which has sub-tropical climate and due to surface homogeneity of its mangrove forests can be used as a pilot for natural, and homogeneous ecosystems such as forested areas. To create an automatic PIFs extraction in this area, initially, we masked areas that should not be in the PIFs by applying some thresholds. At this stage, water, vegetation and mountainous areas were filtered. Then, the radiometric resolution of the two different sensor images was identical, and using the differential histogram shape of the two images, unvaried areas were extracted. To validate the accuracy of the proposed method, absolute radiometric correction using atmospheric and topographic correction (ATCOR), fast line-of-sight atmospheric analysis of hypercubes (FLAASH) and atmospheric correction (ATMOSC) methods, and relative radiometric correction using both empirical line calibration method and dark object subtraction method and automatic radiometric correction using QUick atmospheric correction (QUAC) and autonomous atmospheric correction (IMAGINE AAIC) were applied on the data. The output of all atmospheric correction methods and the proposed method was applied in the image algebra change detection in the form of a difference and with a threshold of twice the standard deviation from the mean to be checked by 219 points. The results of validation and qualitative studies of the histogram comparisons proved the proper functioning of the proposed method (κ greater than 0.8), and using cross tables, the performance of the proposed method is similar to that of the empirical line calibration method (more than 76%). Finally, a few unique features in the current research proposal, including simplicity, automation, negligible systematic error, the possibility of using in a biomarker degradation warning system, the independence on the type of sensor used, distinguish it from other radiometric correction methods.

Assessment of Total Suspended Sediment Distribution under Varying Tidal Conditions in Deep Bay: Initial Results from HJ-1A/1B Satellite CCD Images

Using Deep Bay in China as an example, an effective method for the retrieval of total suspended sediment (TSS) concentration using HJ-1A/1B satellite images is proposed. The factors driving the variation of the TSS spatial distribution are also discussed. Two field surveys, conducted on August 29 and October 26, 2012, showed that there was a strong linear relationship (R2 = 0.9623) between field-surveyed OBS (optical backscatter) measurements (5-31NTU) and laboratory-analyzed TSS concentrations (9.89–35.58 mg/L). The COST image-based atmospheric correction procedure and the pseudo-invariant features (PIF) method were combined to remove the atmospheric effects from the total radiance measurements obtained with different CCDs onboard the HJ-1A/1B satellites. Then, a simple and practical retrieval model was established based on the relationship between the satellite-corrected reflectance band ratio of band 3 and band 2 (Rrs3/Rrs2) and in-situ TSS measurements. The R2 of the regression relationship was 0.807, and the mean relative error (MRE) was 12.78%, as determined through in-situ data validation. Finally, the influences of tide cycles, wind factors (direction and speed) and other factors on the variation of the TSS spatial pattern observed from HJ-1A/1B satellite images from September through November of 2008 are discussed. The results show that HJ-1A/1B satellite CCD images can be used to estimate TSS concentrations under different tides in the study area over synoptic scales without using simultaneous in-situ atmospheric parameters and spectrum data. These findings provide strong informational support for numerical simulation studies on the combined influence of tide cycles and other associated hydrologic elements in Deep Bay.

Development of a new image based atmospheric correction algorithm for aerosol optical thickness retrieval using the darkest pixel method

The problem of atmospheric intervention has received considerable attention from researchers in remote sensing who have developed a range of methods, either simple or sophisticated. The sophisticated methods require auxiliary information about the state of the atmosphere which is obtained either from standard databases or from simultaneous in-situ field measurements or by iterative techniques. It has been found that the darkest pixel atmospheric correction (DP) is one of the most effective atmospheric correction methods especially for visible spectral bands. The DP is the simplest and fully image-based correction method. The integrated use of the DP basic theory and the radiative transfer equation is implemented in this study. Indeed, this leads to the development of the proposed 'image-based atmospheric correction algorithm.' The proposed algorithm retrieves the aerosol optical thickness (AOT) only for areas with urban and maritime aerosols. The effectiveness of this algorithm is assessed by comparing the AOT values retrieved from the proposed 'image-based atmospheric correction algorithm' after applied to Landsat TM/ETM+ images with those measured in-situ both from MICROTOPS II hand-held sun photometer and the CIMEL sun photometer (AERONET). It has been found that the AOT values retrieved from the proposed algorithm were very close with those measured from the CIMEL sun photometer for the Limassol area in Cyprus.

Study of the surface water circulation in San Blas channel (Argentina) using landsat imagery

This paper deals with the application of satellite images to study turbidity and water circulation patterns in San Blas channel during a theoretical tidal cycle. Eight Landsat TM and ETM images acquired under clear-sky conditions and representing different tidal stages were selected from a pool of Landsat images provided by the argentinean National Commission of Space Activities (CONAE) and the US Geological Survey. Standard digital image processing techniques were used to perform geometric and radiometric corrections on the visible and near-infrared bands. An image-based atmospheric correction (COST method by CHAVEZ, 1996) was applied. An ISODATA unsupervised classification was performed in order to identify different turbidity levels throughout the channel and adjacent areas. The results suggest that suspended sediment transport towards the channel mouth by ebb currents occurs along both flanks. These currents carry suspended sediment into the open sea, generating an ebb tidal delta which tends to rotate in a clockwise direction. Flood currents trigger turbidity mostly over the southern flank of the channel, generating a flood tidal delta with elongated banks extending in the direction of the tidal currents. From the elongated shape of the turbidity plumes, general tidal circulation patterns were identified.

Image-based reflectance conversion of ASTER and IKONOS imagery as precursor to structural assessment of plantation forests in KwaZulu-Natal, South Africa

Reflectance-converted imagery is a requirement for establishing temporally robust remote sensing algorithms, given the reduction of time-specific atmospheric effects. Thus, in this study image-based atmospheric correction methods for ASTER and IKONOS imagery for retrieving surface reflectance of plantation forests in KwaZulu-Natal, South Africa were evaluated. This effort formed part of a larger initiative that focused on retrieval of forest structural attributes from resultant reflectance imagery. Atmospheric correction methods in this study included the apparent reflectance model (AR), dark object subtraction model (DOS), and the cosine approximation model (COST). Spectral signatures derived from different image-based models for ASTER and IKONOS were inspected visually as first departure. This was followed by comparison of the total accuracy and Kappa index computed from supervised classification of images that were derived from different image-based atmospheric correction of ASTER and IKONOS imagery. The classification accuracy of DOS images derived from ASTER and IKONOS imagery exhibited percentages of 93.3% and 94.7%, respectively. Classification accuracies for images from AR and COST, on the other hand, resulted in lower accuracy values of 87.9% and 83.6% for ASTER and 90.5% and 92.8% for IKONOS, respectively. We concluded that the image-based DOS model was better suited to atmospheric correction for ASTER and IKONOS imagery in this study area and for the purpose of forest structural assessment. This has important implications for the operational use of similar imagery types for forest inventory approaches.

New atmospheric correction technique to retrieve the ocean colour from SeaWiFS imagery in complex coastal waters

There exists a large demand for an accurate atmospheric correction of satellite ocean colour data over highly turbid coastal waters, where the standard atmospheric correction (SAC) algorithms designed for open ocean water turn out to be unsuccessful because of eventual interference of elevated radiance from suspended materials and perhaps the shallow bottom with the corrections based on the two near-infrared bands at 765 and 865 nm in which the water-leaving radiances are discarded (or modelled) in order to estimate aerosol radiative properties and extrapolate these into the visible spectrum in the atmospheric correction of the imagery. Furthermore, in the presence of strongly absorbing aerosols (e.g. Asian dust and Sahara dust) the SAC algorithms often underestimate water-leaving radiance values in the violet and blue spectrum or completely fail to deliver the desired biogeochemical products for coastal regions. To make the satellite ocean colour data offer unrivaled utility in monitoring and quantifying the components of ecologically important coastal waters, this study presents a more realistic and cost-effective image-based atmospheric correction method to accurately retrieve water-leaving radiances and chlorophyll concentrations from SeaWiFS imagery in the presence of strongly absorbing aerosols over highly turbid Northwest Pacific coastal waters. This method is a modified version of the spectral shape matching method (SSMM) previously developed by Ahn and Shanmugam (2004 Korean J. Remote Sens. 20 289–305), re-treating the assumption of spatial homogeneity of the atmosphere using simple models for assessing the contributions of aerosol and molecular scattering. Because of the difficulties in making atmospheric measurements concurrently with each overpass of SeaWiFS the atmospheric diffuse transmittance values are dependent on a standard method with the SAC scheme designed for processing SeaWiFS ocean colour data. The new method is extensively tested under the presence of various atmospheric conditions using SeaWiFS imagery and the results are compared with in situ (ship-borne) measurements in highly turbid coastal waters of the Korean Southwest Sea (KSWS). Such comparison demonstrates the efficiency of SSMM in terms of removing the effects of strongly absorbing aerosols (Asian dust) and improving the accuracy of water-leaving radiance retrieval with an RMSE deviation of 0.076, in contrast with 0.326 for the SAC algorithm which masked most of the sediment-laden and aerosol-dominated coastal areas. Further comparison in the Yellow Sea waters representing a massive phytoplankton bloom on 27 March 2002 revealed that the SAC algorithm caused an excessive correction for the visible bands, with the 412 nm band being affected the most, leading to severe overestimation of chlorophyll concentrations in the bloom-contained waters. In contrast, the SSMM remained very effective in terms of reducing errors of both water-leaving radiance and chlorophyll concentration estimates.

A comparison of radiometric normalization methods when filling cloud gaps in Landsat imagery

Mapping persistently cloudy tropical landscapes with optical satellite imagery usually requires assembling the clear imagery from several dates. This study compares methods for normalizing image data when filling cloud gaps in Landsat imagery with imagery from other dates. Over a complex tropical island landscape, namely St. Kitts and Nevis and the island of St. Eustatius, all of the methods tested reduce interdate image differences for ETM+ bands 1-5 and 7, NDVI, and band 4:5 ratio. Regression tree normalization reduces the interdate differences more consistently than linear regression or histogram matching. Normalizing ETM+ images with regression trees can produce more seamless imagery than linear normalization, histogram matching, or image-based atmospheric correction via dark object subtraction. More seamless imagery enhances visual interpretation and helps reveal the distributions of forest formations in these landscapes. Decision tree classification of cloud-filled Landsat imagery can accurately map land cover and detailed forest formations. Decision tree classification accuracy is not highly dependent on the method used to make the cloud-filled imagery, however, at least as long as (i) classification model training data reflect class spectral variability, and (ii) ancillary spatial data that relate to the distributions of classes are used in the classification. Cloud-filled imagery is also known as cloud-cleared imagery.

Image-based atmospheric correction of QuickBird imagery of Minnesota cropland

High spatial resolution QuickBird satellite data have provided new opportunities for remote sensing applications in agriculture. In this study, image-based algorithms for atmospheric correction were evaluated on QuickBird imagery for retrieving surface reflectance ( ρ λ ) of corn and potato canopies in Minnesota. The algorithms included the dark object subtraction technique (DOS), the cosine approximation model (COST), and the apparent reflectance model (AR). The comparison with ground-based measurements of canopy reflectance during a 3-year field campaign indicated that the AR model generally overestimated ρ λ in the visible bands, but underestimated ρ λ in the near infrared (NIR) band. The DOS–COST model was most effective for the visible bands and produced ρ λ with the root mean square errors (RMSE) of less than 0.01. However, retrieved ρ λ in the NIR band were more than 20% (mean relative difference or MRD) lower than ground measurements and the RMSE was as high as 0.16. The evaluation of the COST model showed that atmospheric transmittance ( T λ θ ) was substantially overestimated on humid days, particularly for the NIR band because of the undercorrection of water vapor absorption. Alternatively, a contour map was developed to interpolate appropriate T λ θ for the NIR band for clear days under average atmospheric aerosol conditions and as a function of precipitable water content and solar zenith angle or satellite view angle. With the interpolated T λ θ , the accuracy of NIR band ρ λ was significantly improved where the RMSE and MRD were 0.06 and 0.03%, respectively, and the overall accuracy of ρ λ was acceptable for agricultural applications.

Validation of satellite data for quality assurance in lake monitoring applications

The operational application of remote sensing technologies to lake water quality monitoring requires products derived from remote sensing to be quantitatively self-consistent and have a certified accuracy. Fundamental elements in this quality assurance framework are sensor radiometric calibration and atmospheric correction models, which are briefly discussed in the paper. In order to evaluate the accuracy of present operational techniques to retrieve basic parameters from satellite data, such as water-leaving radiance and reflectance, an experiment was organised in the frame of SAtellite remote sensing for Lake MONitoring (SALMON), a European Union co-funded research project. A series of ship-based radiometric and atmospheric measuring campaigns were conducted on Lake Iseo and Lake Garda (Italy) together with limnological sampling. Four Landsat-5 Thematic Mapper (TM) scenes were acquired during different seasons and simultaneous in situ measurements were made. After the radiometric calibration procedure, satellite digital images were processed by applying two entirely image-based atmospheric correction models. These models account for the effects of both additive scattering and multiplicative transmittance effects in the atmosphere on the at-satellite measured signal. The results achieved using these procedures were evaluated by comparing satellite-based estimates with in situ measurements of water reflectance. The root mean square difference between Landsat TM-derived reflectance values and ground measurements was close to 0.010 reflectance for each TM spectral band. Such image-based correction models, requiring no in situ field measurements during the satellite overpass, constitute a valid method of lake water monitoring.

Determination of chlorophyll concentration changes in Lake Garda using an image-based radiative transfer code for Landsat TM images

The distribution of phytoplankton chlorophyll concentration in Lake Garda (Italy) was estimated using Landsat Thematic Mapper (TM) data acquired at two different times, February 1992 and March 1993. To investigate the waterleaving radiance adequately, the contribution of the atmospheric path radiance reaching the sensor should be removed. In this work a completely image-based atmospheric correction method was applied by means of an inversion technique based on a simplified radiative transfer code (RTC). A semi-empirical approach of relating atmospherically corrected TM spectral reflectances to in situ measurements through regression analysis was used. Limnological parameters were measured near to the TM images dates; some of the in situ measurements were used to define algorithms relating chlorophyll concentration measurements to water surface reflectance and the others too were used to validate the results of the predictive model. The models developed, which performed better (r2 = 0.818) when concentrations were higher than > 3.0 mg m3, were used to map chlorophyll concentration throughout the lake. Spatial distribution maps of chlorophyll concentration and concentration changes were produced with contour intervals of 1 mg m3.