Timely detection of flooding is paramount for saving lives as well as evaluating levels of damage. Floods generally occur under specific weather conditions, such as excessive precipitation, which makes the presence of clouds very likely. For this reason, radar-based sensors are most suitable for near-real-time flood mapping. The public dataset Sen1Floods11 recently released by the Cloud to Street is one example of ongoing beneficial initiatives to employ deep learning for flood detection with synthetic aperture radar. The present study used this dataset to improve flood detection using well-known segmentation architectures, such as SegNet and UNet, as networks. In addition, this study provided a deeper understanding of which set of polarized band combination is more suitable for distinguishing permanent water, as well as flooded areas from the SAR image. The overall performance of the models with various kinds of labels and a combination of bands to detect all surface water areas were also assessed. Finally, the trained models were tested on a completely different location at Kerala, India, during the 2018 flood for verifying their performance in the real-world situation of a flood event outside of the given test set in the dataset. The results prove that trained models can be used as off-the-shelf models to achieve an intersection over union (IoU) as high as 0.88 in comparison with optical images. The omission and commission error were less than 6%. However, the most important result is that the processing time for the whole satellite image was less than 1 min. This will help significantly for providing analysis and near-real-time flood mapping services to first responder organizations during flooding disasters.

Spatiotemporal fusion (STF) techniques play important roles in Earth observation analysis as they can produce images with both high spatial and temporal resolution. However, existing STF models often fuse images from various satellites, not satisfying the demand for precise crop monitoring. In contrast, unmanned aerial vehicle (UAV) images can deliver detailed data, and deep learning (DL)-based STF models have a high potential for automatically extracting abstract features. To this end, this study proposed a novel end-to-end DL-based STF model named UAV-Net, which can produce centimeter-scale UAV images. UAV-Net has an encoder-decoder architecture with Modified ResNet (MResNet), Feature Pyramid Network (FPN), and decoder modules. The encoder uses MResNet modules to extract input features, while the FPN module performs a multiscale fusion of these features before reconstructing UAV images using transposed convolution in the decoder module. Through the comparison and ablation experiments, this study evaluated the efficacies of the MResNet modules with 18, 34, and 50 layers, along with FPN module of UAV-Net. The experimental results on real-world datasets demonstrated that UAV-Net adequately produce UAV images both visually and quantitatively. Furthermore, Comparison with state-of-the-art STF models highlights the innovative and effective of UAV-Net for producing centimeter-scale images. The predicted centimeter-scale images have the potential to be useful for various environmental monitoring applications.

Mongolia, situated in central Asia and bordered by Russia to the north and China to the south, experiences a semi-arid climate across most of its territory. Grasslands are pivotal in Mongolia’s agricultural sustainability and food security, facing rapid changes in the last two decades that underscore the ongoing need for innovative approaches to assess vegetation conditions. This study aims to evaluate grassland biomass measurement and prediction through the analysis of high-resolution satellite data. By conducting a time series assessment of grazing-induced changes in vegetation dynamics at the long-term monitoring sites of the Botanic Garden and Research Institute, Mongolian Academy of Sciences, we seek to refine our understanding. The investigation covers biomass estimation across various Mongolian grassland landscapes, encompassing desert, steppe, and mountain regions. Spanning the grassland growing season from May 2020 to October 2023, the research leveraged diverse ground data types, including surface reflectance measurements, geographic coordinates for satellite data correction, and aboveground dry biomass. These components were instrumental in developing a biomass estimation model reliant on establishing correlations between the satellite-derived Normalized Difference Vegetation Index and biomass. The predicted biomass facilitated the time series map analysis and dynamic analysis. The PlanetScope surface reflectance correlates strongly at 0.97 with field measurements, indicating robust relations. Biomass and the Normalized Difference Vegetation Index show correlations of 0.82 for dry grassland, 0.80 for mountain grassland, and 0.65 for desert grassland, with a combined correlation coefficient of 0.62, revealing distinct characteristics across these grasslands. Time series dynamic analysis reveals rising biomass differences between grazed and ungrazed areas, suggesting potential grassland degradation. Variations in the slope coefficient of biomass differences among grassland types indicate differing degradation patterns, emphasizing the need for effective grazing management practices to sustain and conserve Mongolian grasslands. This highlights the potential of remote sensing in monitoring and managing grassland ecosystems.

Precise co-registration of multi-sensor, multispectral and multi-temporal satellite imagery is crucial for effective remote sensing applications. This paper presents an advanced method for satellite image co-registration using mirror arrays as artificial point sources, developed at the Center for Research and Application of Satellite Remote Sensing of Yamaguchi University (YUCARS), Japan. We combined highreflectivity acrylic mirror arrays with Fourier-based phase correlation techniques (AROSICS) for sub-pixel shift estimation to achieve accurate inter-sensor and band-to-band registration. The study focused on co-registering GRUS-1 (Axelspace Corporation) satellite imagery with Sentinel-2 data and improving band-to-band co-registration within GRUS-1 imagery, as well as finding the optimal combination of global (single tie point) and local (tie point grid) co-registration methods. A comparison of commonly used resampling techniques – nearest neighbor, bilinear interpolation, and cubic convolution – was also performed to assess the tradeoffs between image quality and positional accuracy. Results show notable improvements in geolocation accuracy, with RMSE values of 3.59, 4.05, and 4.14 meters in the sub-pixel range for three different GRUS-1 bands when compared to the Global Reference Image (GRI). The mirror array proved particularly effective in combination with the local method (global to local), and demonstrated superior results in high displacement cases on its own (global only), offering a reliable alternative to traditional co-registration methods with less loss of information due to resampling. Furthermore, this study builds upon previous research demonstrating the potential of mirror arrays for both geometric and radiometric calibration, highlighting the versatility of this approach in enhancing overall satellite image quality and consistency.

Understanding physical processes in nature, including the occurrence of slow-onset natural disasters such as droughts and landslides, requires knowledge of the change in soil moisture between two points in time. The study was conducted on a relatively bare soil, and the change in soil moisture was examined with an index called Normalized radar Backscatter soil Moisture Index (NBMI) using Sentinel-1 satellite data. Along with soil moisture measured with a probe on the ground, a study of correlation with satellite imagery was conducted using a Multiple Linear Regression (MLR) model. Furthermore, the Dubois model was used to predict soil moisture. Results have shown that NBMI on a logarithmic scale provides a good representation of soil moisture change with R²~86%. The MLR model showed a positive correlation of soil moisture with the co-polarized backscatter coefficient, but an opposite correlation with the surface roughness and angle of incidence. The results of the Dubois model showed poor correlation of 44.37% and higher RMSE error of 17.1, demonstrating the need for detailed and accurate measurement of surface roughness as a prerequisite for simulating the model. Of the three approaches, index-based measurement has been shown to be the most rapid for understanding soil moisture change and has the potential to be used for understanding some mechanisms of natural disasters under similar soil conditions.

Monitoring grassland biomass accurately and frequently is critical for ecological management, climate change assessment, and sustainable resource use. However, the use of single-satellite data faces challenges due to trade-offs between spatial resolution and temporal frequency, especially for large areas. High-resolution imagery, such as PlanetScope, provides detailed spatial data but presents significant challenges in data management and processing over large regions. Conversely, low-resolution sensors such as JPSS-VIIRS offer daily global coverage with low memory data but lack the spatial detail required for precise biomass estimation, making it difficult to retrieve or validate model parameters due to the mismatch with small ground reference data polygons. To overcome these limitations, this study introduces a robust methodology for accurate frequent biomass estimation based on JPSS-VIIRS data through spectral harmonization, adapting a high-resolution biomass estimation model originally developed from PlanetScope imagery. The core innovation is an optimized Spectral Band Adjustment Factor (SBAF) approach tailored specifically to grassland spectral characteristics. This method significantly enhances spectral alignment, reducing red-band reflectance discrepancies from 6.2% to 4.8% in grassy areas and from 6.9% to 4.0% in bare areas. NDVI discrepancies also improved substantially. Applied across Mongolia, the harmonized VIIRS data estimated a five-year average biomass of 71.4 g/m², clearly reflecting environmental variability. Specifically, the P375 dataset showed average biomass estimates of 54.8 g/m² for desert grasslands (10.5% higher than PlanetScope), 122.6 g/m² for dry grasslands (9.6% higher), and 134 g/m² for mountain grasslands (1.9% lower). The uncertainty analysis showed strong overall agreement with PlanetScope-derived biomass, with an RMSE of 11.6 g/m², a mean percentage difference of 10.74%, and an R² of 0.92. While mountain grasslands exhibited the lowest RMSE, a relatively lower R² indicated limited variability. Higher uncertainty in desert and dry grasslands highlighted the impact of ecological heterogeneity on biomass estimation accuracy. These detailed comparisons demonstrate the effectiveness and accuracy of the proposed methodology in bridging spatial and temporal gaps, providing a valuable tool for large-scale weekly grassland biomass monitoring with applicability beyond the Mongolian context.