| چکیده انگلیسی مقاله |
Runoff modelling and forecasting is one of the main issues in flood control and water resources management. The most important runoff triggering/controlling variables are climatic, physiographic and land use/cover variables. Landscape metrics which can be defined as quantitative spatial characteristics of patches, classes, or entire landscape, play an important role in rainfall-runoff process. In this study the effect of different landscape metrics and climatic and physiographic variables on runoff generation were investigated for 42 sub-basins of the Urmia Lake basin. To this end, land use/cover maps were produced for 2000 and 2019. Several important landscape metrics were calculated in class and landscape levels. Important climatic variables such as rainfall and temperature were also considered as well as important physiographic variables i.e. elevation, area and slope. Three different models namely partial least square (PLS) regression, random forest (RF) and group method of data handling (GMDH) were used for stream flow modelling in two different paradigms i.e. global and local modelling. Particle swarm optimization (PSO) and genetic algorithm (GA) were used to cluster sub-basins into two homogenous groups. The most important variables were selected by principal component analysis (PCA) for GMDH and RF. Results showed that streamflow modelling in homogenous clusters (local modelling) can significantly enhance the performance of modelling methods. It was also shown that GMDH outperformed PLS and RF. |
