Geostatistical approach for meteo-oceanographic variables evaluation at the Brazilian coast

MODIS chlorophyll-a concentration (chla), sea surface temperature (SST), and photosynthetically active radiation (PAR) were used to perform a geographically weighted regression (GWR) analysis within a 150-km buffer of the Brazilian coast. The correlation was between chla as the regressed variable and SST or PAR as the predictors. Both a GWR and a Bayesian GWR (BGWR) were used for evaluating the variables. Colored matrices were plotted for displaying beta values, significance, residuals, and t-statistics. Coefficients of determination (R2) were computed for all months. Also, the ratio of the GWR beta estimates and the 95% confidence interval BGWR estimates was computed. Results showed overall better R2 for SST than for PAR regression but also better beta estimates for PAR than for SST in relation to BGWR beta significance range. Northern regions of the Brazilian coast exhibited lower statistical significance. July had lowest GWR beta values and best significance, January highest beta values and worst significance, and April and October highly variable results.