Linking hydrological security and landscape insecurity in the moribund deltaic wetland of India using tree-based hybrid ensemble method in python
2021
Pal, Swades | Paul, Satyajit
The main goal of the present study is to develop hydrological security model (HSM) and landscape insecurity model (LIM) of the wetlands in moribund deltaic floodplain using a tree-based hybrid ensemble method. The study employs four tree-based novel hybrid approaches such as Random Forest (RF), Extremely randomized forest (ETC), gradient boosting (GBM), and eXtreme gradient boosting (XGB) for modelling hydrological security and landscape insecurity. Six hydrological parameters such as water presence frequency (WPF), water depth, Hydro-duration, variability of water depth using standard deviation, distance from rivers, and regression slope of wetland depth have been employed for hydrological security modelling, and nine landscape parameters such as aggregation index, patch cohesion index, edge density, mean radius of gyration arithmetic, largest patch Index, mean perimeter-area ratio, percentage of landscape, splitting index, total edge have been employed for landscape insecurity modelling. The performance of each model is evaluated by estimating precision, recall, F1-score, Matthew's correlation coefficient (MCC), and the area under the receiver operating characteristic (ROC) curve (AUC). The outcomes revealed that GBM and XGB pose the highest accuracy level (AUC more than 0.95 for HSM and 0.85 for LIM), followed by RF, ETC models. Models' outcome shows that about 50% of wetland area belongs to the low hydrological secure zone. From phase I to phase III this area increased by more than 18%. The area under high hydrological secure zones reduces by about 55%. Landscape insecurity in this region raised by 41% from phase I to phase III. Linking HSM and LIM shows that reduction of hydrological security is responsible for enhancing landscape insecurity in this region.
显示更多 [+] 显示较少 [-]