ABSTRACT The integration of multimodal and multi-temporal remote sensing data has become an important approach for improving land cover classification in agricultural irrigation districts. However, how different feature organization strategies influence classification performance remains insufficiently explored. This study systematically investigates the effects of multimodal feature combinations, temporal organization strategies, and temporal accumulation periods on land cover classification. An ensemble learning-based classification framework was developed to quantitatively evaluate these factors using multi-temporal Sentinel-1 SAR, Sentinel-2 imagery, and derived spectral indices. Two representative irrigation districts in China, the Hongqi Canal Irrigation District and the Changshou Irrigation District, were selected as study areas. Experimental results demonstrate that the organization strategy of multimodal and multi-temporal features significantly affects classification accuracy. Specifically, optical MSI features generally contributed more to classification performance than SAR features, while multimodal integration provided complementary information that improved robustness and spatial consistency. Temporal ordering of multimodal features influenced model performance, and longer temporal accumulation periods enhanced classification accuracy by better capturing crop phenological dynamics, although diminishing returns were observed beyond a certain period. Overall, this study provides a systematic evaluation of multimodal and multi-temporal feature organization strategies and offers practical guidelines for optimizing feature construction in irrigation district land cover classification.

ABSTRACT The Tibetan Plateau, known as the ‘Asian Water Tower’, faces significant water security challenges due to its high sensitivity to climate change. Advancing Earth observation for sustainable water monitoring is thus essential for building climate resilience in this region. This study proposes a two-stage transfer learning strategy using the SegFormer model to overcome domain shift and data scarcity – key barriers in developing robust AI for climate-sensitive applications. After pre-training on a diverse source domain, our model was fine-tuned for the arid Zhada Tulin area. Experimental results show a substantial performance boost: the Intersection over Union (IoU) for water body segmentation surged from 25.50% (direct transfer) to 64.84%. This AI-driven accuracy is crucial for disaster risk reduction, particularly in monitoring flash flood-prone systems. Beyond quantitative improvements, our analysis demonstrates that domain-specific transfer learning enhances the model’s detection capability. Specifically, it achieves high recall rates ( > 80 % ) while maintaining robust precision ( > 73 % ) – a balance critical for risk monitoring applications. The resulting high-precision maps provide valuable support for understanding hydrological patterns and informing water management strategies in this climatically sensitive region. Our work thus demonstrates an effective technical solution for monitoring arid plateau regions and contributes to advancing AI-powered Earth observation for disaster preparedness in critical transboundary river headwaters.

ABSTRACT The identification of land cover types using multispectral remote sensing data remains one of the most practical and widely adopted approaches. In this study, we evaluated the potential of Pixxel Tech Demo-1 (TD-1) hyperspectral imagery in comparison with Landsat-9 multispectral data for land cover classification. Specifically, this study aimed to investigate which land cover classes benefit most from the enhanced spectral resolution of hyperspectral data, leading to improved classification accuracy. To this end, CORINE 2018 data were used as a reference, focusing on seven land cover types: water, forest, rangeland, agriculture, artificial surfaces, bareland, and flooded vegetation. Both Landsat-9 and Pixxel TD-1 images were acquired at the surface reflectance level, eliminating the need for additional atmospheric correction. For classification, we employed Extreme Gradient Boosting (XGBoost), an ensemble learning algorithm known for its high accuracy and robust performance in handling complex remote sensing data. To address the high dimensionality of Pixxel TD-1 hyperspectral imagery, Principal Component Analysis (PCA) was applied, reducing the feature set to 30 components. The results consistently demonstrated the superior performance of hyperspectral data: Pixxel TD-1 achieved a significantly higher overall accuracy (91.8%) compared to Landsat-9 (87.1%). Notably, hyperspectral imagery proved especially effective in distinguishing complex land cover types such as flooded vegetation, bareland, and rangeland, as indicated by their higher F1 scores. These findings underscore the strong potential of Pixxel’s TD-1 hyperspectral sensors for advanced land cover classification.

ABSTRACT Unoccupied Aerial Systems (UAS) have emerged as powerful tools for ultra-shallow (<2 m) bathymetric mapping using Structure-from-Motion (SfM) or spectrally derived bathymetry (SDB) based methods. However, the impact of equipment selection on the accuracy of bathymetric maps created through these methods is not well characterized. This study investigates how different UAS navigation and imaging sensors impact bathymetric accuracy and whether this differs for SfM- and SDB-based methods. We assessed five common UAS across two coral reef habitats. real-time kinematics or post-processing kinematics based positioning was critical for accurate SfM-based bathymetry (RMSE = 0.14 ± 0.02 m). However, shallow water adaptations of Stumpf’s band ratio trained on as few as 10 ground truth samples consistently outperformed SfM, regardless of the UAS configuration (RMSE = 0.11 ± 0.003 m). Spectrally trained support vector machines and single-layer neural networks demonstrated the strongest performance (RMSE = 0.09 ± 0.005 m) but required 200 ground truth measurements to reach such accuracies. Multispectral sensors improved the accuracy of spectrally derived bathymetric maps, but broadband RGB cameras also performed well (RMSE difference ≤0.06 m). SfM-derived bathymetry could be combined with spectral models to improve bathymetry reconstruction in texture-poor sand habitats. However, the overall accuracy of such combined models remained constrained by the accuracy of the original SfM reconstruction. Ultimately, Stumpf’s band ratio using the red image band as the denominator provided the most practical and accurate method for UAS-based ultra-shallow bathymetry retrieval when ground truth data was sparse, while spectral machine learning models performed the best when ground truth data were abundant.

ABSTRACT Simple elevation, defined here as the unprocessed vertical height of the land surface above a reference datum without the application of derivative computations such as slope or curvature, is frequently cited as the most or among the most important predictor variables in wetland classification. However, importance is often measured using biased metrics and there have been numerous cases in which excluding simple elevation has had little to no impact on model performance. Additionally, although wetlands can form at any elevation, they are often more abundant and/or obvious at lower elevations. This can lead to biased training conditions that result in overfitting. To advance wetland mapping, it is essential to accurately quantify the importance of predictor variables and evaluate their impact on model generalization. The objectives of this study were therefore to re-evaluate the importance and highlight the potential risks associated with incorporating simple elevation into wetland classification models. Seasonal composites of C-band Sentinel-1 Synthetic Aperture Radar are classified in combination with three different simple elevation datasets to: (i) demonstrate a scenario in which, regardless of its quality, simple elevation offers limited predictive power to the model, and (ii) simulate biased training conditions to show how its inclusion can lead to overfitting. Results show that the correlation between wetland presence and elevation is spatially variable, limiting the ability of models to generalize across domains when elevation is included.

ABSTRACT The accelerating impacts of climate change and human activities on marine and coastal ecosystems demand advanced monitoring strategies that are both accurate and sustainable. Hyperspectral imaging (HSI) offers unparalleled capability to characterize water constituents, benthic habitats, and coastal vegetation, yet its application in aquatic environments is challenged by spectral redundancy, spatial heterogeneity, and severe domain shift. To address these limitations, we propose a KAN-based Masked Autoencoder (KAN-MAE) framework that integrates artificial intelligence and machine learning innovations into ocean remote sensing. The model employs decoupled spectral and spatial encoders with adaptive spectral masking, cross-attention fusion, and KAN-enhanced Transformer blocks for robust representation learning. During pre-training, Brownian distance covariance (BDC) regularization preserves nonlinear inter-band dependencies, while fine-tuning leverages a lightweight KAN classifier optimized with cross-entropy and Gaussian-kernel triplet loss to enhance discriminability under scarce labels. Experiments on Gaofen-5 coastal imagery and benchmark HSI datasets demonstrate improved classification accuracy, resilience to noise, and strong generalization across domains. By advancing AI-driven self-supervised learning for ocean remote sensing, this work supports operational monitoring of coastal ecosystems, sustainable fisheries, and blue carbon management, aligning with the UN Sustainable Development Goals on climate action (SDG 13) and life below water (SDG 14).

ABSTRACT This study integrated remote-sensing data (RSD) from the Arabian Sea area and coastal community perspectives in Kerala to demonstrate the complementarity of transdisciplinary research in generating knowledge on blue carbon ecosystems (BCEs) and advancing sustainable coastal resource governance solutions. Community perspectives demonstrated that although Kerala is endowed with BCEs, notably mangroves, since the 1980s, they have been declining due to human-environmental drivers. Although local communities predominantly attribute the losses to human activities, e.g. dredging and aquaculture, RSD analysis revealed sustained environmental vulnerability drivers. Elevated sea surface temperature (SST), sea surface salinity (SSS), and sea surface chlorophyll (SSC) levels have been observed since 2010, ranging from 1°C, 0.2 PSU, and 0.1 mg/m3, respectively. Correlating with community perspectives, elevated SSS have affected mangrove abundance and species, including in post-summer and monsoon periods. Hotter summer temperatures above 30°C progressively occurring in April, May, and June, and extending to October and December, increase livelihood and mangrove losses, yet collaborative governance options are limited. Overall, the satellite-derived SST, SSS, and SSC findings correlated with coastal community perspectives, in reinforcing physically consistent interannual variability, confirming their reliability and suitability for subsequent oceanographic research and ecological modelling applications. Through transdisciplinary (TD) research approaches, local observations on threatened BCEs and livelihood change could be integrated and strengthened with marine satellite and scientific data to advance sustainable resource governance options. In this study, six policy and research recommendations that could advance marine TD research are given.

ABSTRACT Skin colour can be a deceptive or confusing indicator of fruit ripeness because low storage temperatures can result in fruits with less skin darkening, and late-season skin darkening can also happen before harvest. In addition to harming the fruit, pressing it to determine its firmness also poses major hygienic issues. The demand for fresh and high-quality food goods, as well as the expanding market for tropical fruits, is driving the creation of new ‘intelligent packaging’ that takes into account the quality characteristics and stages of fruit ripening. In this work, a non-invasive technique for predicting the health condition of fruits is elucidated. A Sierpinski Fractal Antenna sensor is designed with 3 GHz bandwidth for exciting the target. Random forest algorithm has been used to optimize the dimensions of the antenna to exhibit resonance at 5.3 GHz. For microwave imaging at multiple locations, the sensor is moved automatically by a robotic car for capturing it’s reflection characteristics after being scattered from the target. A fresh mango and rotten mango have been used as targets in front of the antenna for detecting it’s health condition. Synthetic aperture radar (SAR) based microwave imaging system has been leveraged for detecting the regions of spoilage in the mango.

ABSTRACT Accurate crop structure information is fundamental for precision agriculture, particularly in regions with highly fragmented farmland. This study focuses on Jiangsu Province, China, and proposes a multi-source remote sensing approach to address the classification challenges posed by complex agricultural landscapes. Optical Sentinel-2, synthetic aperture radar (SAR) Sentinel-1, and high-resolution Gaofen-2 (GF-2) imagery were integrated to exploit their spatial and temporal complementarities. A modified multi-scale attention U-Net (MSA-UNet) was developed to extract cropland parcels from GF-2 imagery, which served as the basic classification units. Time-series spectral and polarimetric features derived from Sentinel-1/2 data were used to train a convolutional neural network – bidirectional long short-term memory (CNN – BiLSTM) model for parcel-level summer crop classification. The proposed MSA-UNet achieved an intersection over union (IoU) of 73.89% and an F1-score of 84.66%, outperforming the baseline U-Net by 5.87 and 4.07% points, respectively. Incorporating both optical and SAR features improved the precision to 91.28%, representing gains of 2.37 and 13.87% points compared to models using single-source inputs. The CNN – BiLSTM architecture effectively captured both intra-seasonal dynamics and long-term dependencies in the time series. These results demonstrate that the proposed method significantly improves the accuracy of crop mapping in fragmented agricultural regions and offers a practical solution for large-scale agricultural monitoring and decision support.

ABSTRACT This research started with the objective to document and analyse the changes between spring and summer on the dyke Quenin, an earthen structure subjected to erosion by the sea waves. As test flights, two surveys were done and some conclusions were drawn from the corresponding structure from motion (SfM) pipeline outputs, orthomosaic and digital surface model (DSM). Initially, we conducted a study to assess the achievable accuracy of topographic models with the available equipment, correction networks and flight parameters. Next, we evaluated the impact of various ground control point (GCP) and checkpoint (CP) distributions on the accuracy of the DSM to identify the optimal configuration for future monitoring missions. Finally, we were able to detect changes in topography, vegetation and shallow bathymetry using only RGB images. In terms of topography, we identified areas affected by overtopping and erosion. Vegetation changes were observed using visible band vegetation indices (VVI), although these were influenced by the site lighting conditions during the missions. Shallow bathymetry, covering depths from 0 to −5 m, was derived using well-established methods based on water colour. The main contribution of this article is to demonstrate the potential of red, green and blue (RGB) imagery acquired by unmanned aerial vehicle (UAV) for monitoring changes in topography, bathymetry, armour rocks and vegetation and to define flight parameters for future monitoring missions on the dyke.