Avaliação de metodologias de interpolação espacial aplicadas a dados hidrográficos costeiros quase-sinóticos

The objective of this work relies on the evaluation of interpolation techniques applied to 28 sampling sites of radial distribution, located at the Itajaí-açu river plume. The data used was salinity at 0.5 meter depth collected by CTD. The methods tested were: (1) inverse distance to a power, (2) kriging, (3) minimum curvature, (4) modified shepard’s method, (5) natural neighbor, (6) nearest neighbor, (7) polynomial regression, (8) radial basis function, (9) triangulation with linear interpolation, (10) moving average and (11) local polynomial, and the software used was Surfer version 8 (Golden Software Inc.). For each method, were created isoline maps of distinct grid resolutions (lines x columns): 5x4, 10x7, 20x14, 30x21, 40x28, 50x34, 100x69 and 1000x682, using data from the high river discharge, when the salinity gradient is maximum and better for comparisons. Were checked the residuals from interpolations, and estimated their means, variances and visual evaluation of the maps. Results showed that grid resolution must be coherent with sample spatial distribution, and it was observed that the higher the grid resolutions, the lower the means and variances, however it ended up generating interpolation problems. The best method was local polynomial, with very smooth curves represented perfectly the river plume. The gain on resolution tends to honor data, but it ends up generating bull’s eyes, and to represent isolines from geophysical fluids, it has to be taken account that the processes are very dynamic, and there are always noise in data, so to get good representations it’s not necessary to honor data closely, but instead, it’s useful to get only the patterns of the flow