ORIGINAL RESEARCH
Comparative Prediction of Stream Water Total
Nitrogen from Land Cover Using Artificial Neural
Network and Multiple Linear Regression
Approaches
B. J. Amiri, K. Nakane
More details
Hide details
Division of Environmental Dynamics and Management, Graduate School of Biosphere Science,
Hiroshima University, 1-7-1 Kagamiyama, Higashi-Hiroshima 739-8521 Japan
Pol. J. Environ. Stud. 2009;18(2):151-160
KEYWORDS
ABSTRACT
Performance of two data-driven models that were developed using Artificial Neural Networks (ANNs)
and Multiple Linear Regression (MLR) approaches were investigated in prediction of Total Nitrogen (TN)
concentration in twenty-one river basins in Chugoku district of Japan. Comparison of TN concentration predictions,
which were carried out using an ANN-based model and MLR-based model indicated that prediction
of the former model (r2=0.94, p<0.01) was more accurate than that of the latter model (r2=0.85, p<0.01). Lack
of a sufficient data set that might be considered an obstacle for cross-validating models that are developed was
dealt with using a Monte Carlo-based sensitivity analysis of the developed models. This initiative could provide
reliable information for judging predictive capacity of the developed models stochastically. Result of sensitivity
analysis revealed that predictive capacity of the ANN-based model varied between 0-2 mg/L.
Moreover, prediction of the negative outputs was not observed. using the ANN-based model for TN concentration
in stream water.