Spatial Prediction of Flood Potential in Ardabil Plain Using Random Forest Algorithm

Objective: Extracting floodplains in watersheds is the first management strategy in developing flood control and containment programs. For flood management in the Qarasu River watershed located in the Ardabil plain, the present study was conducted using the Random Forest (RF) technique.

Research Method: For this purpose, first, the training points of the random forest technique were determined, including 183 flood points and 163 non-flood observation points in the spatial distribution of the basin. Then, 17 effective parameters in the flooding of the watershed, including DEM map, the slope of the river bed, distance from the channel, distance from the waterway, Topographic Position Index (TPI), Modified Catchment Area (MCA), Terrain Ruggedness Index (TRI), Topographic Wetness Index (TWI), Stream Power Index (SPI), distance from the dam, the density of waterways, drainage density, the direction of waterway flow, land use, geology, soil erosion and geomorphological landforms selected and entered the model as independent variables. Using the random forest model, a flood hazard map was drawn for the Ardabil plain.

Results: The results showed that the model evaluation metrics including AUC, MSE, MAI, RMSE and R2 were obtained as 0.996, 0.026, 0.045, 0.161 and 0.90, respectively, which indicates the reliability of the model used. Among the independent parameters, distance from the Qarasu River, DEM, MCA, and TWI had the greatest impact on flooding in the Qarasu River watershed with a correlation coefficient of -0.753, -0.471, 0.382, and 0.366, respectively. The final map obtained from the model indicated that lands close to the Qarasu River, especially in the Ardabil Plain, have the highest risk of flooding.

Conclusion: In general, about 80,497 hectares of the Qarasu River watershed, which are mainly located in the central part of Ardabil, are at very high risk of flooding with heavy rainfall.