Effect of Southern Oscillation Index and spatially distributed climate data on improving the accuracy of Artificial Neural Network, Adaptive Neuro‐Fuzzy Inference System and K‐Nearest Neighbour streamflow forecasting models

Abstract

AbstractStreamflow forecast models are essential ingredients for water resources management. Due to nonlinear nature of streamflow generation, Artificial Neural Networks (ANNs), Adaptive Neuro‐Fuzzy Inference System (ANFIS) and K‐Nearest Neighbour (K‐NN) have turned into popular forecast models. The main objective of this paper was to study the effect of various non‐commonly used data, including spatially distributed climatic data, as well as Southern Oscillation Index (SOI), on the improvement of one‐ to three‐month ahead flow forecasts in the Karun basin, Iran. Temperature maps were produced via elevation–temperature relationships, while inverse distance‐weighted interpolation was applied to generate precipitation maps. In addition, three different input architectures were constructed for ANNs and ANFIS models using cross‐correlation and principal component analysis techniques. Next, the most accurate models with input point data were identified based on the root mean squared error and mean absolute error. Finally, the effect of using spatial climatic data and SOI on the performance of the most accurate models was investigated. Results showed that distributed precipitation data improved the performance of ANN and ANFIS models, while K‐NN model accuracy was only marginally affected. Adding SOI as an additional input improved the forecast performance slightly in all lead times.