Deep Learning-Assisted Mapping of Wetland Dynamics in the Niger Delta Using Open-Access Multi-Sensor Remote Sensing Data

Wetlands play a crucial role in the global hydrological cycle and act as natural buffers that enhance resilience to extreme climatic events. However, they are increasingly being threatened due to human pressures. In this study, we assembled a deep learning (DL) framework to document recent changes in wetland dynamics in the Niger Delta using multi-temporal, multi-sensor satellite data. To achieve this, specific objectives were: (i) identifying optimal sensor combinations for wetland classification, (ii) evaluating DL model performance, (iii) performing land use/land cover (LULC) classification across two timeframes (2019/2020 and 2021/2022), and (iv) analyzing hydrological variability through satellite indicators and comparing with wetland changes. Several tests were undertaken to determine the best model for wetland mapping using the optimal multi-sensor image dataset as input in the DL models (Res-UNet and DeepLabV3). The results showed that the Res-UNet model, combined with the ResNet 152 backbone, outperformed the DeepLabV3 model (with identical backbones) in segmenting wetlands, achieving high F1 scores and overall accuracy (the F1 scores ranged from 0.83 to 0.91). Estimates of wetland transition across the landscape were obtained using the optimal DL model outputs as inputs for change detection analysis. The change detection analysis revealed that converting dense vegetation to wetlands was the main driver of wetland gain, while wetland loss primarily emanated from transitions to dense vegetation. The spatial extent of wetland loss was estimated to be 27,346 hectares, whereas gains totalled 98,453 hectares. Furthermore, the study established links between hydroclimatic indicators and changes in wetland, emphasizing the influence of climatic conditions and human activities on wetland expansion. Ultimately, our findings illustrate the potential of advanced remote sensing and deep learning for wetland monitoring and highlight the necessity of informed conservation strategies for fragile ecosystems.