The effect of hybrid models integration sequence: taking Fushun area as an example

Landslide susceptibility zoning (LSZ) is a crucial step in reducing landslide related losses, and its accuracy remains a long-standing research focus. Hybrid models are widely used, but little is known about coupling sequences. Here, we show that the accuracy of hybrid models can be improved solely through model integration in areas with weak computing resources.In this paper, results indicate that the Weight of evidence (WoE) - Multi-Layer Perceptron (MLP) - Random Forest (RF) (WRM) performs the best (AUC = 0.841, Precision = 80.00%, accuracy = 71.96%, recall = 58.95%, F1 = 67.88%). Compared with WoE - MLP (WM), the accuracy has been improved by 1.95%. The accuracy of the WoE - RF (WR) has increased from 70.59% to 79.63% by the MLP. In addition, the proportion of landslides in areas classified as high and very high by WRM increased by 4.21% compared to WM, while the proportion in these areas only increased by 1.51%. WMR boosted landslide share by 3.15 % while expanding area by merely 1.51 % than WR. The increase in the proportion of higher areas is much smaller than the increase in the proportion of landslides. This result can provide support for local disaster prevention and models accuracy improvement.