Estimation Of Colored Dissolved Organic Matter(CDOM) For Complex Estuarine And Coastal Waters

Colored dissolved organic matter(CDOM)is acted as the optically active constituent of dissolved organic matter(DOM)in waters.CDOM has significant impact on cycling carbon and climate change,and also affect drinking water safety,aquatic ecosystem development,and carbon balance.Many studies have payed more and more attention to CDOM.Remote sensing technology is a key method to monitor the CDOM abundance.Traditional ocean color satellite sensors,including CZCS(Coastal Zone Color Scanner),SeaWiFS(Sea-Viewing Wide Field-of-View Sensor),MODIS(Moderate Resolution Imaging Spectroradiometer),MERIS(Medium Resolution Imaging Spectrometer),and VIIRS(Visible Infrared Imaging Radiometer Suite),generally are equiped with coarse spatial resolution,which make it difficult to observe small-scale area for estuarine and coastal waters.This research mainly focused on the issues that CDOM estimation models development and application from the two remote sensing data,covering Landsat-8 with 30-m spatial resolution and Sentinei-2 with 10-60m spatial resolution,for complex case II waters.In the present study,study site is the Saginaw River estuarine regions.Three field cruises were made on May 10,2012,October 18,2012,and May 7,2013,and during those dates,ten,eighteen and thirteen samples(41 samples in total)were measured,respectively.The northern Gulf of Mexico was selected for assessment of model generalization,and it is typical estuarine and coastal region.Its water quality is influenced by the material inputs from Mississippi River and Atchafalaya River.For the topic about CDOM models development from Landsat,there are three major issues of building an empirical model from current studies:(1)band-ratio setting is not unified,(2)there are a variety of empirical function types used for establishing model,(3)and the data level of band-ratio model input is not unified,including RTOA and Rrs.For these three issues,we used original model ranking method to calibrate and validate algorithms.The results showed that the best band ratio was B3/B4(B3:green band,525-600nm,B4:red band,630-680nm).The band-ratio model input at Rrs data level performed better than Rt level.Power and exponential models gave better accuracies than linear and logarithmic models,as well as presented stronger robustness.The B3/B4 empirical algorithms via fitting with power and exponential functions gave the best performance with R2 = 0.819,RMSE = 0.889 m-1 and R2 = 0.829,RMSE = 0.863 m-1,respectively.An independent data group was further empoyed to validate power and exponential models accuracy and satisfied results were achieved with the RMSE of 0.642 m-1 and 0.504 m-1,respectively.The best model for CDOM estimation from Landsat-8 was aCDOM(440)=40.75e-2 463x,x = Rrs(B3)/Rrs(B4).CDOM abundance was driven by temperature and discharge in Saginaw River via observeing CDOM estimations from Landsat-8 images between April 2013 to November 2016.Saginaw River presents relative higher CDOM concentration between April to June and lower value between July to September.In addition,spectral bands of Landsat-4,Landsat-5 and Landsat-7 were simulated for CDOM models development and validation,it showed that Landsat series satellite had consistent capacity for CDOM retrieval from the perspective of spectra simulationFor the topic about CDOM retrieval models development for Sentinel-2,the relationship between different wavelength and CDOM present great variation.Thus the correlation for these two variables was explored and we found that B3 band showed the significant correlation with CDOM,followed by B1 and B2 band.Relatively,B4,B5,B6,and B7 band weakly correlated with CDOM.Adjacent bands generally appeared to significant relationship.It is thus the band with great atmospheric correction error can be replaced via near-by band.Based on LOOCV(leave one out cross validation)method,the best CDOM estimation model for Sentinel-2 was explored.The band-ratio setting was followed by this rules that B1 B2 and B3 highly correlated with CDOM was selected as numerator and B4,B5,B6 and B7 weakly correlated with CDOM was selected as denominator.The best model was aCDOM(440)= 22.283e-1.724x,x＝Rrs(B3)/Rrs(B5).Due to unavailability of concurrent in situ observation,Rrs of Sentinel-2 was assessed based on MODIS Rrs product.During the Landsat-8 image preprocessing,the Hydrolight(water optics radiative transfer model)was employed to remove the effects of water suface reflectance,and thereafter Sentinel-2 image preprocessing used the same scheme.The results indicated that Rrs(560)had the highest consistency and Rrs(443)worst.CDOM inversion from Sentinel-2 image kept consistent with the in situ measured CDOM on the same month from different year.In addition,models applicability was evaluated for other complex waters:case study of the northern Gulf of Mexico.The newest atmospheric correction algorithm,ACOLITE(Atmospheric Correction for OLI ’lite’),was employed to Landsat-8 image.Meanwhile,Rrs of the first four bands of Landsat-8 was cross compared based on MODIS Rrs product.Validation results showed perfect consistency.Combined with previous study about CDOM retrieval from MODIS for the same study site,it showed that CDOM estimations from Landsat-8 were higher than MODIS-based model.Because the Sentinel-2 satellite is still in the status of instrument testing,we cannot obtain good quality image with rigorous 5-days revisit time.In the future,it is very promising and meaningful that fusion of MODIS image with high temporal resolution and Landsat/Sentinel-2 image with high spatial resolution to study high spatio-temporal resolution dynamics of CDOM.Further,we can recognize the environment effects on CDOM distribution variation with defined-scale.