The field of wildfire forecasting and detection is rapidly advancing with the development of innovative deep learning architectures and datasets. Recent research has focused on integrating local and global factors to improve predictability on subseasonal to seasonal timescales. Teleconnection-aware models have shown significant promise in capturing large-scale Earth-system context, leading to improved forecasting performance. Additionally, attention-enhanced deep neural networks have been proposed for methane plume detection, demonstrating the potential for accurate and efficient detection of greenhouse gas emissions. Other notable developments include the creation of multi-modal spatio-temporal benchmark datasets for fine-grained wildfire spread forecasting and the application of transformer-based models for observation-driven correction of numerical weather prediction. Noteworthy papers include: TeleViT1.0, which improves AUPRC performance over existing models for wildfire forecasting, and AttMetNet, which achieves state-of-the-art performance in methane plume detection with a lower false positive rate. FireSentry, a novel benchmark dataset, and PyroFocus, a two-stage pipeline for real-time wildfire detection, are also noteworthy for their contributions to fine-grained wildfire forecasting and dynamic disaster simulation.