Convolutional neural network-based deep learning for landslide susceptibility mapping in the Bakhtegan watershed

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Апрель 17, 2025

Landslides pose a significant threat to infrastructure, ecosystems, and human safety, necessitating accurate efficient susceptibility assessment methods. Traditional models often struggle capture the complex spatial dependencies interactions between geological environmental factors. To address this gap, study employs deep learning approach, utilizing convolutional neural network (CNN) for high-precision landslide mapping in Bakhtegan watershed, southwestern Iran. A comprehensive inventory was compiled using 235 documented locations, validated through remote sensing field surveys. An equal number of non-landslide locations were systematically selected ensure balanced model training. Fifteen key conditioning factors-including topographical, geological, hydrological, climatological variables-were incorporated into model. While traditional statistical methods fail extract hierarchies, CNN effectively processes multi-dimensional geospatial data, intricate patterns influencing slope instability. The outperformed other classification approaches, achieving an accuracy 95.76% precision 95.11%. Additionally, error metrics confirmed its reliability, with mean absolute (MAE) 0.11864, squared (MSE) 0.18796, root (RMSE) 0.18632. results indicate that northern northeastern regions watershed are highly susceptible landslides, highlighting areas where proactive mitigation strategies crucial. This demonstrates learning, particularly CNNs, offers powerful scalable solution assessment. findings provide valuable insights urban planners, engineers, policymakers implement effective risk reduction enhance resilience landslide-prone regions.

Язык: Английский

Time Series Analysis of Vegetation Recovery After the Taum Sauk Dam Failure

Remote Sensing, Год журнала: 2025, Номер 17(9), С. 1605 - 1605

Опубликована: Апрель 30, 2025

On 14 December 2005, there was a catastrophic flood after failure in the upper reservoir at Taum Sauk Plant southern Missouri. While has been extensive research on cause of dam’s and flood’s immediate impact, limited investigation how vegetation around resulting scour changed since this event. This study fills gap through time-series analysis using imagery sourced from GloVis Planet Explorer to quantify levels prior (2005) 2024. Vegetation level calculated Normalized Difference Index (NDVI), which measures greenness via light reflected by vegetation. inside were compared two 120 m buffer areas surrounding scour, immediately adjacent (0–120 m) 120–240 scour’s edge. Within NDVI showed dramatic loss flood, followed varying for several years, before steady increase proportion with starting 2014. The area edge similar pattern, but lower magnitudes change, likely reflects ragged created flood. farther consistent pattern high vegetation, broader landscape. ground truthing confirmed these patterns between 2006 2011, 2012, revealed much recovery small local within that not apparent though analysis. These recolonization nearby glades (i.e., natural habitats exposed bedrock) glade flora eastern collared lizard (Crotaphytus collaris collaris), an apex predator adapted living rocky, open bioindicator recovery. occurred steadily indicated original oak/hickory forest now minor component recovery, species dominated former forested area.

Язык: Английский

Advanced Modeling of Forest Fire Susceptibility and Sensitivity Analysis Using Hyperparameter-Tuned Deep Learning Techniques in the Rajouri District, Jammu and Kashmir

Advances in Space Research, Год журнала: 2025, Номер unknown

Опубликована: Май 1, 2025

Язык: Английский