Multi-spectral Remote Estimating Of Concentration Of Marine Oil Emulsions Using Lab-derived HSV Color Model

Marine oil spills cause great damage to economy and ecological environment with different pollutions formed on the sea surface in the process of weathering,such as black oils,oil emulsions and oil slicks.Oil emulsions,including oil-in-water(OW)and water-in,oil(WO),are the most hazardous among them.Remote sensing is an important means of marine oil spill monitoring,and various sensors have been used,such as microwave radar,laser fluorescence,thermal infrared and optical sensors.Emulsions have a strong scattering effect in near infrared(NIR)and short wave infrared(SWIR)wavelengths,which made them could be detected by optical sensors.Furthermore,the‘-C-H' absorption characteristics on hydrocarbons' spectra enable hyperspectral sensors be used in emulsions quantifications.However,there are still challenges in quantifications of emulsions with multispectral imageries,while the effects of spectral resolution,mixing pixels and sunglint are still unclear.Although the '-C-H' spectral absorption characteristics can't be reflected in multispectral imageries,the reflectance difference in NIR and SWIR of emulsions might be helpful in oil quantifications with multispectral sensors.In this paper,based on the principle of absorption and scattering,the absorption coefficients were measured for analyzing the spectral characteristics of OW and WO;and the effects of mixing pixels on oil spills estimation in multispectral imagery were analyzed on Landsat,while the sunglint reflectance was modeled and used in oil slicks roughness estimation with MODIS imagery.The main research contents are listed as follows:(1)The absorption coefficients of different emulsions(26 emulsions with oil concentration ranged from 0.03%to 95%)were measured to analyze the spectral response of different emulsions in conjunction with reflectance spectra.(2)The Landsat spectra of different emulsions were simulated with spectral resampling method to find out the optimum bands for oil estimation on multispectral imageries.Then,these spectra were mixed to analyze the effects of emulsions' type,oil concentration and coverage area on the reflectance.(3)Simulation and application of sunglint on MODIS imagery.On the base of Cox-Munk model,sunglint reflectance of background water and oil slicks were modeled on a MODIS imagery,and an oil slick roughness estimation method based on critical angle was proposed.(4)A HSV model based color segmentation method was developed for estimation of oil emulsions concentrations with multi-spectral imageries.The method was validated with Airborne Visible Infrared Imaging Spectrometer(AVIRIS),and the influence of sunglint were analyzed and eliminated.Conclusions could be given as follows:(1)Mixing pixels reflectance was related to emulsions types,oil concentration and emulsions coverage.When oil concentration was determined,reflectance linearly increased with oil content.When emulsions coverage was ascertained,reflectance of Class I(with no WO)arould NIR increased with oil content in a nagtive exponential shape,and the reflectance of Class II(containing WO)in SWIR linearly increased oil content.In concentration estimation,it is necessary to identify the major impactor of reflectance in advance.(2)Sunglint is helpful in oil slicks detection,but it also makes classification and estimation of emulsions difficult.Based on the critical angle constraints and Cox-Munk model,the surface roughness and refractive index of oil slicks could be derived from MODIS imageries.However,it could not be directly used in Landsat imageries as a result of scale effect.In this study,a baseline fitting method of sunglint reflectance of clear water was used to indirectly eliminate a certain amount of sunglint effect on oils,which further promoted the emulsions recognition and estimation in Landsat images.(3)When SWIR,NIR and VIS were chosen as the bands of red,green and blue respectively,WO,OW and clear water had distinguished colors,which made HSV-based segmentation method could be used to determine the emulsions types and model the oil content distribution.This would also be helpful to improve the estimation accuracy of oil content according to the different statistical relationships between reflectance and oil content in different mixing types.