Search

Mangrove ecosystems provide numerous ecological services and serve as vital habitats for a wide range of flora and fauna. Thus, accurate mapping and monitoring of relevant land covers in mangrove ecosystems are crucial for effective conservation and management efforts. In this study, we proposed a novel approach for mangrove ecosystem mapping using a Hybrid Selective Kernel-based Convolutional Neural Network (HSK-CNN) framework and multi-temporal Sentinel-2 imagery. A time series of the Normalized Difference Vegetation Index (NDVI) products derived from Sentinel-2 imagery was produced to capture the temporal behavior of land cover types in the dynamic ecosystem of the study area. The proposed algorithm integrated Selective Kernel-based feature extraction techniques to facilitate the effective learning and classification of multiple land cover types within the dynamic mangrove ecosystems. The model demonstrated a high Overall Accuracy (OA) of 94% in classifying eight land cover classes, including mangrove, tidal zone, water, mudflat, urban, and vegetation. The HSK-CNN demonstrated superior performance compared to other algorithms, including random forest (OA = 85%), XGBoost (OA = 87%), Three-Dimensional (3D)-DenseNet (OA = 90%), Two-Dimensional (2D)-CNN (OA = 91%), Multi-Layer Perceptron (MLP)-Mixer (OA = 92%), and Swin Transformer (OA = 93%). Additionally, it was observed that the structure of the network, such as the types of convolutional layers and patch sizes, affected the classification accuracy using the proposed model and, thus, the optimum scenarios and values of these parameters should be determined to obtain the highest possible classification accuracy. Overall, it was observed that the produced map could offer valuable insights into the distribution of different land cover types in the mangrove ecosystem, facilitating informed decision-making for conservation and sustainable management efforts. [ABSTRACT FROM AUTHOR]

Burn severity is fundamental to post‐fire impact assessment and emergency response. Vegetation Burn Severity (VBS) can be derived from satellite observations. However, Soil Burn Severity (SBS) assessment—critical for mitigating hydrologic and geologic hazards—requires costly and laborious field recalibration of VBS maps. Here, we develop a physics‐informed Machine Learning model capable of accurately estimating SBS while revealing the intricate relationships between soil and vegetation burn severities. Our SBS classification model uses VBS, as well as climatological, meteorological, ecological, geological, and topographical wildfire covariates. This model demonstrated an overall accuracy of 89% for out‐of‐sample test data. The model exhibited scalability with additional data, and was able to extract universal functional relationships between vegetation and soil burn severities across the western US. VBS had the largest control on SBS, followed by weather (e.g., wind, fire danger, temperature), climate (e.g., annual precipitation), topography (e.g., elevation), and soil characteristics (e.g., soil organic carbon content). The relative control of processes on SBS changes across regions. Our model revealed nuanced relationships between VBS and SBS; for example, a similar VBS with lower wind speeds—that is, higher fire residence time—translates to a higher SBS. This transferrable model develops reliable and timely SBS maps using satellite and publicly accessible data, providing science‐based insights for managers and diverse stakeholders. Plain Language Summary: Post‐fire impact assessment and hazard mitigation heavily relies on burn severity metrics. Vegetation burn severity (VBS)—most relevant for ecological impacts—can be remotely sensed, but soil burn severity (SBS)—most relevant for hydrological impacts—requires laborious field recalibration of VBS maps. Lack of near real‐time SBS information is currently a data gap. Climate change‐driven weather whiplash can narrow the time interval between large wildfires and the ensuing precipitation events, requiring tools for rapid and accurate assessment of SBS without the need for laborious field recalibration of satellite‐derived metrics. Here we developed a physic‐informed Machine Learning model that can accurately develop SBS maps using readily available data without the need for resource and time intensive field assessments of burn severity. Additionally, SBS maps are developed for only a small fraction (less than 0.1%) of all wildfires in the US. Our model can develop SBS maps for all satellite‐observed wildfires in the western US, and has the potential to accomplish this task globally with further training. Finally, our model exhibited scalability with additional data, and was able to extract universal functional relationships between vegetation and soil burn severities across the western US. Key Points: Soil burn severity (SBS) assessment is key for hydrological hazard mitigationVegetation burn severity (VBS) can be remotely sensed, SBS is measured in the fieldPhysics‐informed Machine Learning models to translate VBS to SBS are developed [ABSTRACT FROM AUTHOR]

Floods are one of the most destructive natural disasters, causing financial and human losses every year. As a result, reliable Flood Susceptibility Mapping (FSM) is required for effective flood management and reducing its harmful effects. In this study, a new machine learning model based on the Cascade Forest Model (CFM) was developed for FSM. Satellite imagery, historical reports, and field data were used to determine flood-inundated areas. The database included 21 flood-conditioning factors obtained from different sources. The performance of the proposed CFM was evaluated over two study areas, and the results were compared with those of other six machine learning methods, including Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Deep Neural Network (DNN), Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost). The result showed CFM produced the highest accuracy compared to other models over both study areas. The Overall Accuracy (AC), Kappa Coefficient (KC), and Area Under the Receiver Operating Characteristic Curve (AUC) of the proposed model were more than 95%, 0.8, 0.95, respectively. Most of these models recognized the southwestern part of the Karun basin, northern and northwestern regions of the Gorganrud basin as susceptible areas. [ABSTRACT FROM AUTHOR]

• Landsat-9 offers enhanced opportunities for more accurate burned area mapping (BAM). • We develop a novel convolutional shift transformer (CST) model for BAM with Landsat-9. • CST model outperforms residual convolutional neural network and vision transformer. Satellite imagery, specifically Landsat, have been widely used for mapping and monitoring wildfire burned areas. The new Landsat-9 satellite – with higher radiometric resolution compared to its predecessors, and improved temporal resolution when combined with Landsat-8 (∼8 days) – enables a wide range of applications, particularly burned area mapping (BAM). We propose a novel deep learning BAM model that leverages the strengths of the convolutional layers for deep feature generation from Landsat-9 imagery and shift-transformer block for burned area classification. The performance of the model is evaluated in five large fire case studies across the globe. BAM results are also compared with two state-of-the-art models, namely residual convolutional neural network and vision transformer. The proposed convolutional shift-transformer (CST) outperforms other models with an F1-score of greater than 96% across the case studies. Furthermore, CST only requires a single post-fire image that reduces computational costs compared to traditional models that use bi-temporal images. [ABSTRACT FROM AUTHOR]

The Earth has always been under the influence of population growth and human activities. This process causes the changes in land use. Thus, for optimal management of the use of resources, it is necessary to be aware of these changes. Satellite remote sensing has several advantages for monitoring land use/cover resources, especially for large geographic areas. Change detection and attribution of cultivation area over time present additional challenges for correctly analyzing remote sensing imagery. In this regards, for better identifying change in multi temporal images we use hyperspectral images. Hyperspectral images due to high spectral resolution created special placed in many of field. Nevertheless, selecting suitable and adequate features/bands from this data is crucial for any analysis and especially for the change detection algorithms. This research aims to automatically feature selection for detect land use changes are introduced. In this study, the optimal band images using hyperspectral sensor using Hyperion hyperspectral images by using genetic algorithms and Ratio bands, we select the optimal band. In addition, the results reveal the superiority of the implemented method to extract change map with overall accuracy by a margin of nearly 79% using multi temporal hyperspectral imagery. [ABSTRACT FROM AUTHOR]