Rainfall–runoff modelling in low- and high-flow sub-basins using data-driven and PSO-based hybrid approaches

Accurate daily streamflow estimation is important for rainfall-runoff analysis and water-resources assessment. In this study, various data-driven models were developed to estimate same-day streamflow using hydrometeorological data obtained from two gauging stations representing contrasting flow regimes within the Pawtuxet River basin: the low-flow Moswansicut Stream station and the high-flow Pawtuxet River station. Multiple Linear Regression (MLR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Particle Swarm Optimization–Adaptive Neuro-Fuzzy Inference System (PSO-ANFIS), and PSO-Random Tree (PSO-RT) models were evaluated for daily streamflow estimation under two predefined hydrometeorological input structures. Two modelling structures were considered to investigate the effect of antecedent precipitation on model performance. Model 1 used precipitation, air temperature, and lagged streamflow as input variables, whereas Model 2 additionally incorporated one-day-lagged precipitation. Model performance was evaluated using statistical indicators including R 2 , NSE, RMSE, MAE and MAPE. The results showed that machine learning and hybrid models generally outperformed the traditional regression approach, confirming their ability to capture nonlinear rainfall-runoff relationships. For the Moswansicut Stream station, RF2 achieved the highest R 2 and NSE values and the lowest RMSE (R 2 = 0.962, NSE = 0.958, and RMSE = 0.045 m 3 /s), whereas XGBoost2 obtained the best overall multi-metric test rank. For the Pawtuxet River Station, PSO-ANFIS1 (R 2 = 0.960, NSE = 0.960, RMSE = 2.654 m 3 /s, MAE = 1.517 m 3 /s and MAPE = 16.360%) achieved the best test performance across the five primary test metrics. However, the rank-based comparison indicated that no single model was universally superior for both stations. The comparison of input structures also showed that the contribution of antecedent rainfall information was station dependent. While lagged precipitation improved model performance for the Moswansicut Stream station, the original input structure was more effective for the Pawtuxet River station. This finding highlights the importance of selecting input variables according to catchment-specific hydrological behavior.

Authors

Institutions

Publication Details

Journal
Engineering Science and Technology an International Journal
Published
2026-11-18
DOI
https://doi.org/10.1016/j.jestch.2026.102595
Primary Topic
Hydrological Forecasting Using AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Rainfall–runoff modelling in low- and high-flow sub-basins using data-driven and PSO-based hybrid approaches

Bestami Taşar, Nurullah Karaca
2 links
Engineering Science and Technology an International Journal
Hydrological Forecasting Using AI
article

Rainfall–runoff modelling in low- and high-flow sub-basins using data-driven and PSO-based hybrid approaches

Bestami Taşar, Nurullah Karaca
article en

Abstract

Accurate daily streamflow estimation is important for rainfall-runoff analysis and water-resources assessment. In this study, various data-driven models were developed to estimate same-day streamflow using hydrometeorological data obtained from two gauging stations representing contrasting flow regimes within the Pawtuxet River basin: the low-flow Moswansicut Stream station and the high-flow Pawtuxet River station. Multiple Linear Regression (MLR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Particle Swarm Optimization–Adaptive Neuro-Fuzzy Inference System (PSO-ANFIS), and PSO-Random Tree (PSO-RT) models were evaluated for daily streamflow estimation under two predefined hydrometeorological input structures. Two modelling structures were considered to investigate the effect of antecedent precipitation on model performance. Model 1 used precipitation, air temperature, and lagged streamflow as input variables, whereas Model 2 additionally incorporated one-day-lagged precipitation. Model performance was evaluated using statistical indicators including R 2 , NSE, RMSE, MAE and MAPE. The results showed that machine learning and hybrid models generally outperformed the traditional regression approach, confirming their ability to capture nonlinear rainfall-runoff relationships. For the Moswansicut Stream station, RF2 achieved the highest R 2 and NSE values and the lowest RMSE (R 2 = 0.962, NSE = 0.958, and RMSE = 0.045 m 3 /s), whereas XGBoost2 obtained the best overall multi-metric test rank. For the Pawtuxet River Station, PSO-ANFIS1 (R 2 = 0.960, NSE = 0.960, RMSE = 2.654 m 3 /s, MAE = 1.517 m 3 /s and MAPE = 16.360%) achieved the best test performance across the five primary test metrics. However, the rank-based comparison indicated that no single model was universally superior for both stations. The comparison of input structures also showed that the contribution of antecedent rainfall information was station dependent. While lagged precipitation improved model performance for the Moswansicut Stream station, the original input structure was more effective for the Pawtuxet River station. This finding highlights the importance of selecting input variables according to catchment-specific hydrological behavior.

Engineering Science and Technology an International JournalVol. 84
İskenderun Technical University (TR)
Clean water and sanitation, Climate action
Hydrological Forecasting Using AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon