Gully erosion susceptibility mapping using four machine learning methods in Luzinzi watershed, eastern Democratic Republic of Congo

Soil erosion by gullying causes severe soil degradation, which in turn leads to socio-economic and environmental damages tropical subtropical regions. To mitigate these negative effects guarantee sustainable management of natural resources, gullies must be prevented. Gully strategies start devising adequate assessment tools identification driving factors control measures. achieve this, machine learning methods are essential assist the implement site-specific This study aimed at assessing effectiveness four (Random Forest (RF), Maximum Entropy (MaxEnt), Artificial Neural Network (ANN), Boosted Regression Tree (BRT)) identify gully's factors, predict gully susceptibility Luzinzi watershed, Walungu territory, eastern Democratic Republic Congo (DRC). In this study, were first identified through multiple field surveys then digitized using a very high-resolution image (CNI/airbus) from Google Earth. Overall, 270 identified, 70% (189) randomly selected train topographical, hydrological, hypothesized gully-related conditioning factors. The remaining 30% (81 gullies) used for testing studied threshold-independent area under receiver operating characteristic (AUROC) true skill statistic (TSS) as performance results showed that RF MaxEnt algorithms outperformed other methods; model with AUROC = 0.82 (82%) MaxEtent (0.804: 80.4%) had higher prediction accuracies than BRT: 0.69 (69%) ANN: 0.55 (55%). TSS indicated best predicting watershed. On hand, such Digital Elevation Model (DEM), Normalized Difference Wetness Index (NDWI), Vegetation (NDVI), slope, distance roads, rivers, Stream Power (SPI) played key roles occurrence. Given significance gullies' occurrence, shown policy-makers adopt consider lower risk occurrence related consequences watershed scale DRC. • Four applied mapping scale. most suitable among models used. Elevation, NDWI, NDVI, ED roads influence on occurrences.