Multivariable statistical regression models of the areal extent of hypoxia over theTexas–Louisiana continental shelf

Abstract

Observations of the areal extent of seasonal hypoxia over the Texas–Louisiana continentalshelf from 1985 to 2010 are correlated with a variety of physical and biogeochemical forcingmechanisms. Significant correlation is found between hypoxic area and both nitrogen load (r2 = 0.24) and east–westwind speed (r2 = 0.16). There is also a significant increasing trend in the areal extent ofhypoxia in time; a linearly increasing trend over the entire record (r2 = 0.17), a step increase in area for the years 1994 and beyond (r2 = 0.21), and a step increasefor 1993 and beyond (r2 = 0.29) were all found to be significantly correlated with area. The year 1988, often included inother studies, was found to be a statistical outlier, in that the statistical regressionproperties are strongly modified when this year is included. The exclusion of any otheryear does not have as great an effect as excluding 1988 from the record. Theyear 1989 is also excluded, as this year had no full shelf survey, for a total of 24years of data for the record. Multivariable regression models using all possiblecombinations of the forcing variables considered were calculated. The best performingmodels included east–west wind, either a linear trend in time or step in time(1994 and beyond), and either nitrogen load or river discharge combined withnitrogen concentration. The range of adjusted correlation coefficients ranged fromr2 = 0.47 to 0.67. The best model (east–west wind, a step increase in time 1994 andbeyond, river discharge, and nitrogen concentration) has a standard error of 3008 km2.