|
Superiority of the maximum entropy (MaxEnt) model over random forest for flood susceptibility mapping with limited data in semi-arid regions (Case Study: Alashtar Watershed, Lorestan Province, Iran)
|
Ali Haghi Zadeh *    , Negar Arjmand , Reza Fathi Ganji |
| Department of Watershed Management Engineering, Faculty of Natural Resources, Lorestan University, Khorramabad, Lorestan Province, Iran. |
|
|
Abstract: (391 Views) |
| This study aimed to conduct flood susceptibility zoning using artificial intelligence models in the Alashtar watershed, located in Lorestan Province, Iran. The study area, covering 797.64 km², is part of the Karkheh sub-basin and experiences a semi-arid, cold climate with an average annual precipitation of 570 mm.In this research, ten influencing factors namely, slope, aspect, precipitation, distance from roads, distance from rivers, soil type, land use, geological formation, Topographic Wetness Index (TWI), and drainage densitywere employed as input data. From a total of 58 observation points (comprising 33 flood points and 25 non-flood points), 70% were allocated for model training and the remaining 30% for validation. The performance of two models, Random Forest (RF) and Maximum Entropy (MaxEnt), was evaluated using the Receiver Operating Characteristic (ROC) curve and the Area Under the Curve (AUC) index.The results indicated that the MaxEnt model, with AUC values of 0.98 in the training phase and 0.93 in the validation phase, demonstrated superior performance in predicting flood susceptibility. Analysis of the influencing factors revealed that most floods occurred under the following conditions: precipitation classes of 600–700 mm, distances of 0–100 m from roads and rivers, slope classes of 0–5% and 5–15%, northwest-facing aspects, rainfed agricultural land use, and old alluvial formations. According to the susceptibility map, 83.23% of the area fell into the very low susceptibility class, 9.22% into the low class, 4.66% into the moderate class, 2.12% into the high class, and only 0.74% into the very high class. Furthermore, a direct relationship was observed between the TWI and flood occurrence. Finally, the flood susceptibility zonation map was prepared using the superior MaxEnt model. |
Article number: 5 |
| Keywords: Flood susceptibility, Geographic Information System, Machine learning, Mapping, Semi-arid regions. |
|
|
|
Type of Study: Research |
Subject:
Special
Received: 2026/07/2 | Revised: 2026/08/30 | Accepted: 2026/07/22 | ePublished ahead of print: 2026/07/31 | ePublished: 2026/07/31
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
