Assessing the effect of future landslide on ecosystem services in Aqabat Al-Sulbat region, Saudi Arabia

Ecosystem conservation requires monitoring natural resources, such as vegetation, freshwater, and wetland systems, as well as their adaptations to natural and anthropogenic stressors. As a result, the present study intends to assess the effect of future landslide on different ecosystem services in Aqabat Al-Sulbat regions. In the present study, landslide susceptibility maps (LSM) at different return periods as a proxy of future landslide have been generated with integration of ensemble machine learning algorithms. Stacking framework based on bagging, dagging, and artificial neural network (ANN) along with random forest has been proposed to generate LSMs. Also, bagging, dagging, and ANN have been applied individually. Receiver operational characteristics (ROC curve) were used to validate the LSM model against reality. The best LSM predicting model determined by the ROC curve was used to create future LSM maps by combining rainfall at 2–100 year return periods (using Gumble extreme value distribution). The potential loss of ecosystem services value in the Aqabat Al-Sulbat region was estimated. All models anticipate 6–16 km2 and 27–41 km2 of the study region as very high and high land susceptibility (LS) zones, respectively. According to the area under the curve (AUC) of the ROC curve, stacking-bagging outperformed all other models (AUC: 0.91). While the region encompassed by very high LS zones would steadily rise from 2 to 100 year return periods (6.1–20.44 km2). As a result, ecosystem services (ES) would be destroyed as well. The results revealed that the ES generating region with zero value will steadily expand with increasing return periods (39–42 km2) owing to the expansion of high LS zones. The research delivers reliable spatially explicit results in order to influence policy decisions on forest management investment, landslide management, and development activity management as well as replication possibilities in other data sparse places.