Methodology to estimate lake turbidity using object-oriented classification of multispectral images.

This work implements an object-oriented multispectral image classification to quantify turbidity levels in the Grande Lagoon of San Pedro (Chile) (36 degrees 51' S, 73 degrees 06' W). The first step in this methodology is multiscale segmentation; then, to characterize the lagoon cover, different classes are defined according to the selection of training areas associated with data recorded in situ and texture descriptors. In the last stage, the accuracy of each test is evaluated using the Global Membership (P-G) and the Global Stability (E-G) indices proposed by this work and the results underwent a refinement process. The proposed methodology resulted in the creation of turbidity maps of the Grande Lagoon of San Pedro, Chile, where 86% of the lagoon surface is associated with a turbidity level between 1.0 and 1.7 NTU, indicating that the turbidity of this lagoon is low and homogeneous as compared to other lentic systems studied.