A self-identification Neuro-Fuzzy inference framework for modeling rainfall-runoff in a Chilean watershed

Modeling the relationship between rainfall and runoff is an important issue in hydrology, but it is a complicated task because both the high levels of complexity in which both processes are embedded and the associated uncertainty, affect the forecasting. Neuro-fuzzy models have emerged as a useful approach, given the ability of neural networks to optimize parameters in a fuzzy system. In this work a Self-Identification Neuro-Fuzzy Inference Model (SINFIM) for modeling the relationship between rainfall and runoff on a Chilean watershed is proposed to reduce the uncertainty of selecting both the rainfall and runoff lags and the number of membership functions required in a fuzzy system. The data comes from the Diguillin river located in Nuble region and average daily runoff and average daily rainfall recorded from years 2000 to 2018, according to the Chilean directorate of water resources (DGA). In addition, we worked with the Colorado River basin, located in the Maule region, to validate the method developed. The experimental results showed a good adjustment using the last 3 years as validation set, further improvement was achieved using only the last year was used as validation test, obtaining 84% of R-2 and 0, 91 Kling Gupta Efficiency, higher than other forecasting models such as Adaptive Neuro-Fuzzy Inference System (ANFIS), Artificial neural networks (ANN), and Long Short-Term Memory (LSTM) approach. In addition, Nash-Sutcliffe efficiency and percent BIAS indicate the method is a promising model. On the other hand, even better results were obtained in the validation basin, whose adjustment was 94% and an efficiency of 97%. Therefore, the proposed model is a solid alternative to forecast the runoff in a given watershed, obtaining good performance measurements, managing to predict both the low and peak runoff values from rainfall events, avoiding the requirement to determine a priori the lags of time series and the number of fuzzy rules.