Premium
A neuro‐fuzzy modeling tool to estimate fluvial nutrient loads in watersheds under time‐varying human impact
Author(s) -
Marcé Rafael,
Comerma Marta,
García Juan Carlos,
Armengol Joan
Publication year - 2004
Publication title -
limnology and oceanography: methods
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.898
H-Index - 72
ISSN - 1541-5856
DOI - 10.4319/lom.2004.2.342
Subject(s) - adaptive neuro fuzzy inference system , fluvial , computer science , series (stratigraphy) , estimator , watershed , time series , environmental science , statistics , fuzzy logic , data mining , mathematics , machine learning , fuzzy control system , artificial intelligence , geology , paleontology , structural basin
Fluvial nutrient loads are usually calculated through a bilogarithmic regression relating flow and river nutrient concentration. This relationship, however, can be highly nonlinear, due to changes in watershed land uses over time. Also, retransformation of data can result in important biases, and available databases usually do not provide the statistical properties needed to apply parametric statistics or time‐series analysis methods. The validity and advantages over customary methods of an Adaptive Neuro‐Fuzzy Inference System (ANFIS) for estimating fluvial nutrient loads in watersheds under time‐varying human impact was tested. Fluvial nutrient loads time‐series were modeled in two watersheds of different size and human impact history. ANFIS and methods based on rating curves and ratio estimators were applied to compare results. The ANFIS approximation gave unbiased estimates of loads and showed advantages over the other methods: It allowed the implementation of a homogeneous, model‐free methodology throughout the data series, avoiding the presence of artifacts in the final load histories; it fitted the observed concentration time‐series better than the other procedures; it worked in real space without the need to logarithmically transform and retransform data; and it gave annual dispersion values, which could be interpreted as annual uncertainties. In addition, the parameters fitted during the ANFIS modeling could be ecologically interpreted, and were a valuable tool to describe features of modeled data and to understand historical changes in human impact on watersheds. MATLAB codes and instructions to implement the new method are provided.