Statistical models for sediment/detritus and dissolved absorption coefficients in coastal waters of the northern Gulf of Mexico

摘要

We developed statistically-based, optical models to estimate tripton (sediment/detrital) and colored dissolved organic matter (CDOM) absorption coefficients (a(sd), a(g)) from physical hydrographic and atmospheric properties. The models were developed for northern Gulf of Mexico shelf waters using multi-year satellite and physical data. First, empirical algorithms for satellite-derived a(sd) and a(g) were developed, based on comparison with a large data set of cruise measurements from northern Gulf shelf waters; these algorithms were then applied to a time series of ocean color (SeaWiFS) satellite imagery for 2002-2005. Unique seasonal timing was observed in satellite-derived optical properties, with a(sd) peaking most often in fall/winter on the shelf, in contrast to summertime peaks observed in a(g). Next, the satellite-derived values were coupled with the physical data to form multiple regression models. A suite of physical forcing variables were tested for inclusion in the models: discharge from the Mississippi River and Mobile Bay, Alabama; gridded fields for winds, precipitation, solar radiation, sea surface temperature and height (SST, SSH); and modeled surface salinity and currents (Navy Coastal Ocean Model, NCOM). For satellite-derived ald and ag time series (2002-2004), correlation and stepwise regression analyses revealed the most important physical forcing variables. Over our region of interest, the best predictors of tripton absorption were wind speed, river discharge, and SST, whereas dissolved absorption was best predicted by east-west wind speed, river discharge, and river discharge lagged by 1 month. These results suggest the importance of vertical mixing (as a function of winds and thermal stratification) in controlling ald distribution patterns over large regions of the shelf, in comparison to advection as the most important control on ag. The multiple linear regression models for estimating a(sd) and a(g) were applied on a pixel-by-pixel basis and results were compared to monthly SeaWiFS composite imagery. The models performed well in resolving seasonal and interannual optical variability in model development years (2002-2004) (mean error of 32% for a(sd) and 29% for a(g)) and in predicting shelfwide optical patterns in a year independent of model development (2005; mean error of 41% for a,d and 46% for ag). The models provide insight into the dominant processes controlling optical distributions in this region, and they can be used to predict the optical fields from the physical properties at monthly timescales.