Remote Sensing

This technical note aims to present a method for developing a Digital Terrain Model (DTM) of the coastal zone based on topobathymetric data from remote sensors. This research was conducted in the waterbody adjacent to the Vistula Śmiała River mouth in Gdańsk, which is char-acterised by dynamic changes in its seabed topography. Bathymetric and topographic measurements were conducted using an Unmanned Aerial Vehicle (UAV) and two hydrographic methods (a Single-Beam Echo Sounder (SBES) and a manual survey using a Global Navigation Satellite System (GNSS) Real-Time Kinematic (RTK) receiver). The result of this research was the development of a topobathymetric chart based on data recorded by the above-mentioned sensors. It should be emphasised that bathymetric data for the shallow waterbody (less than 1 m deep) were obtained based on high-resolution photos taken by a UAV. They were processed using the "Depth Predic-tion" plug-in based on the Support Vector Regression (SVR) algorithm, which was implemented in the QGIS software as part of the INNOBAT project. This plug-in allowed us to generate a dense cloud of depth points for a shallow waterbody. Research has shown that the developed DTM of the coastal zone based on topobathymetric data from remote sensors is characterised by high accuracy of 0.248 m (p = 0.95) and high coverage of the seabed with measurements. Based on the research conducted, it should be concluded that the proposed method for developing a DTM of the coastal zone based on topobathymetric data from remote sensors allows the accuracy requirements provided in the International Hydrographic Organization (IHO) Special Order (depth error ≤ 0.25 m (p = 0.95)) to be met in shallow waterbodies.

This study conducted multi-sensor image classification by utilizing Google Earth Engine (GEE) and applying satellite imagery from Korean Multi-purpose Satellite 3 (KOMPSAT-3), KOMPSAT-5 SAR, Compact Advanced Satellite 500-1 (CAS500-1), Sentinel-1, and Sentinel-2 within GEE. KOMPSAT-3/5 and CAS500-1 images are not provided by GEE. The land-use and land-cover (LULC) classification was performed using the random forest (RF) algorithm provided by GEE. The study experimented with 10 cases of various combinations of input data, integrating Sentinel-1/-2 imagery and high-resolution imagery from external sources not provided by GEE and those normalized difference vegetation index (NDVI) data. The study area is Boryeong city, located on the west coast of Korea. The classified objects were set to six categories, reflecting the region’s characteristics. The accuracy of the classification results was evaluated using overall accuracy (OA), the kappa coefficient, and the F1 score of the classified objects. The experimental results show a continued improvement in accuracy as the number of applied satellite images increased. The classification result using CAS500-1, Sentinel-1/-2, KOMPSAT-3/5, NDVI from CAS500-1, and NDVI from KOMPSAT-3 achieved the highest accuracy. This study confirmed that the use of multi-sensor data could improve classification accuracy, and the high-resolution characteristics of images from external sources are expected to enable more detailed analysis within GEE.

LiDAR sensors have great potential for enabling crop recognition (e.g., plant height, canopy area, plant spacing, and intra-row spacing measurements) and the recognition of agricultural working environments (e.g., field boundaries, ridges, and obstacles) using agricultural field machinery. The objective of this study was to review the use of LiDAR sensors in the agricultural field for the recognition of crops and agricultural working environments. This study also highlights LiDAR sensor testing procedures, focusing on critical parameters, industry standards, and accuracy benchmarks; it evaluates the specifications of various commercially available LiDAR sensors with applications for plant feature characterization and highlights the importance of mounting LiDAR technology on agricultural machinery for effective recognition of crops and working environments. Different studies have shown promising results of crop feature characterization using an airborne LiDAR, such as coefficient of determination (R2) and root-mean-square error (RMSE) values of 0.97 and 0.05 m for wheat, 0.88 and 5.2 cm for sugar beet, and 0.50 and 12 cm for potato plant height estimation, respectively. A relative error of 11.83% was observed between sensor and manual measurements, with the highest distribution correlation at 0.675 and an average relative error of 5.14% during soybean canopy estimation using LiDAR. An object detection accuracy of 100% was found for plant identification using three LiDAR scanning methods: center of the cluster, lowest point, and stem–ground intersection. LiDAR was also shown to effectively detect ridges, field boundaries, and obstacles, which is necessary for precision agriculture and autonomous agricultural machinery navigation. Future directions for LiDAR applications in agriculture emphasize the need for continuous advancements in sensor technology, along with the integration of complementary systems and algorithms, such as machine learning, to improve performance and accuracy in agricultural field applications. A strategic framework for implementing LiDAR technology in agriculture includes recommendations for precise testing, solutions for current limitations, and guidance on integrating LiDAR with other technologies to enhance digital agriculture.

Winter catch crops are promoted in the European Union under the Common Agricultural Policy to improve soil health and reduce nitrate leaching from agricultural fields. Currently, Member States often monitor farmers’ adoption through on-site inspections for a limited subset of parcels. Because of its potential for region-wide coverage, this study investigates the potential of Sentinel-2 satellite time series to classify catch crops at the field level in Flanders (Belgium). The first objective was to classify catch crops and identify the optimal model and time-series input for this task. The second objective was to apply these findings in a real-world scenario, aiming to provide reliable early-season predictions in a separate target year, testing early-season performance and temporal transferability. The following three models were compared: Random Forest (RF), Time Series Forest (TSF), and a One-Dimensional Convolutional Neural Network (1D-CNN). The results showed that, with a limited field-based training dataset, RF produced the most robust results across different time-series inputs, achieving a median F1-score of >88% on the best dataset. Additionally, the early-season performance of the models was delayed in the target year, reaching the F1-score threshold of 85% at least one month later in the season compared to the training years, with large timing differences between the models.

Pan-sharpening aims to generate high-resolution (HR) multispectral (MS) images by fusing HR panchromatic (PAN) and low-resolution (LR) MS images covering the same area. However, due to the lack of real HR MS reference images, how to accurately evaluate the quality of a fused image without reference is challenging. On the one hand, most methods evaluate the quality of the fused image using the full-reference indices based on the simulated experimental data on the popular Wald’s protocol; however, this remains controversial to the full-resolution data fusion. On the other hand, existing limited no reference methods, most of which depend on manually crafted features, cannot fully capture the sensitive spatial/spectral distortions of the fused image. Therefore, this paper proposes a perceptual quality assessment method based on deep feature similarity measure. The proposed network includes spatial/spectral feature extraction and similarity measure (FESM) branch and overall evaluation network. The Siamese FESM branch extracts the spatial and spectral deep features and calculates the similarity of the corresponding pair of deep features to obtain the spatial and spectral feature parameters, and then, the overall evaluation network realizes the overall quality assessment. Moreover, we propose to quantify both the overall precision of all the training samples and the variations among different fusion methods in a batch, thereby enhancing the network’s accuracy and robustness. The proposed method was trained and tested on a large subjective evaluation dataset comprising 13,620 fused images. The experimental results suggested the effectiveness and the competitive performance.

Urban tree canopies are a vital component of green infrastructure, especially in the context of the accelerating urban heat island effect and global climate change. Quantifying urban canopy cover in relation to land use and land cover changes is therefore crucial. However, accurately evaluating visual changes remains a challenge. In this study, we introduced the Urban Cover View Factor (VF) and Potential Influence Intensity Grade (PIIG) for tree canopy (TC) mapping using airborne Light Detection and Ranging (LiDAR) remote-sensing three-dimensional point clouds (3DPCs) from the Incheon metropolitan area, South Korea. The results demonstrated that airborne LiDAR 3DPCs effectively segmented non-sky urban cover views. Furthermore, the PIIG map, derived from the TC VF map, showed a significant correlation between surface heat risks and energy consumption patterns. Areas with lower PIIG grades tended to have higher energy consumption and greater vulnerability to surface heat risks, while areas with higher PIIG grades exhibited the opposite trend. Nevertheless, further exploration of complex urban cover and the collection of sufficient ground-based evidence is crucial for practical PIIG application. Further remote sensing research should support the management of urban tree canopies and urban agriculture to promote sustainable urban greening in response to evolving environmental needs.

Cropland is a fundamental basis for agricultural development and a prerequisite for ensuring food security. The segmentation and extraction of croplands using remote sensing images are important measures and prerequisites for detecting and protecting farmland. This study addresses the challenges of diverse image sources, multi-scale representations of cropland, and the confusion of features between croplands and other land types in large-area remote sensing image information extraction. To this end, a multi-source self-annotated dataset was developed using satellite images from GaoFen-2, GaoFen-7, and WorldView, which was integrated with public datasets GID and LoveDA to create the CRMS dataset. A novel semantic segmentation network, the Global–Local Context Aggregation Network (GLCANet), was proposed. This method integrates the Bilateral Feature Encoder (BFE) of CNNs and Transformers with a global–local information mining module (GLM) to enhance global context extraction and improve cropland separability. It also employs a multi-scale progressive upsampling structure (MPUS) to refine the accuracy of diverse arable land representations from multi-source imagery. To tackle the issue of inconsistent features within the cropland class, a loss function based on hard sample mining and multi-scale features was constructed. The experimental results demonstrate that GLCANet improves OA and mIoU by 3.2% and 2.6%, respectively, compared to the existing advanced networks on the CRMS dataset. Additionally, the proposed method also demonstrated high precision and practicality in segmenting large-area croplands in Chongzhou City, Sichuan Province, China.

Cloud contamination is a critical source of errors in the data assimilation of hyperspectral infrared radiance (IR). Therefore, it is necessary to filter out cloudy observations. In this study, we review and summarize the principles and research progress of cloud detection methods for the hyperspectral IR in the past two decades. Based on the impact of IR data utilization on cloud detection results, cloud detection methods are categorized into five types, namely clear field-of-view (FOV) detection, clear channel detection, three-dimensional cloud detection, cloud-clearing and deep learning methods. Clear FOV methods and clear channel methods aim to identify the purely clear FOVs and spectral channels that are not affected by clouds, respectively. Cloud-clearing methods are used to reconstruct clear-column radiance for cloudy observations. Deep learning cloud detection methods can quickly learn the mapping relationship between infrared hyperspectral radiation characteristics and FOV cloud distribution from a large amount of infrared radiative information with known FOV cloud labels. In this paper, we discuss and provide an outlook on the key issues in current hyperspectral IR cloud detection. Specifically, we analyze and summarize the factors affecting cloud detection, such as surface background information, vertical cloud distribution, hyperspectral IR channel selection, improvements in cloud detection algorithms and model applicability. The results indicate the use of deep learning methods offer advantages in detection accuracy and algorithm efficiency of hyperspectral IR cloud detection.

Wave breaking is a fundamental process in ocean energy dissipation and plays a crucial role in the exchange between ocean and nearshore sediments. Foam, the primary visible feature of wave breaking areas, serves as a direct indicator of wave breaking processes. Monitoring the distribution of foam via remote sensing can reveal the spatiotemporal patterns of nearshore wave breaking. Existing studies on wave breaking processes primarily focus on individual wave events or short timescales, limiting their effectiveness for nearshore regions where hydrodynamic processes are often represented at tidal cycles. In this study, video imagery from a typical low-tide terrace (LTT) beach was segmented into four categories, including the wave breaking foam, using the DeepLabv3+ architecture, a convolutional neural networks (CNNs)-based model suitable for semantic segmentation in complex visual scenes. After training and testing on a manually labelled dataset, which was divided into training, validation, and testing sets based on different time periods, the overall classification accuracy of the model was 96.4%, with an accuracy of 96.2% for detecting wave breaking foam. Subsequently, a heatmap of the wave breaking foam distribution over a tidal cycle on the LTT beach was generated. During the tidal cycle, the foam distribution density exhibited both alongshore variability, and a pronounced bimodal structure in the cross-shore direction. Analysis of morphodynamical data collected in the field indicated that the bimodal structure is primarily driven by tidal variations. The wave breaking process is a key factor in shaping the profile morphology of LTT beaches. High-frequency video monitoring further showed the wave breaking patterns vary significantly with tidal levels, leading to diverse geomorphological features at various cross-shore locations.

In light detection and ranging (LiDAR) applications, correct intensities from echo data intuitively contribute to the characterization of target reflectivity. However, the power in raw echo waveforms may be clipped owing to the limited dynamic range of LiDAR sensors, which directly results in false intensity values generated by existing LiDAR systems working in scenarios involving highly reflective objects or short distances. To tackle the problem, an asymmetric Gaussian echo model is proposed in this paper so as to recover echo power–time curves faithfully to its optical physics. Considering the imbalance in temporal length and steepness between rising and falling edges, the echo model features a shared mean and two distinct standard deviations on both sides. The accuracy and effectiveness of the proposed model are demonstrated by correcting the power–time curve from a real LiDAR loaded with avalanche photodiode (APD) sensors and estimating the reflectivities of real targets. As when tested by targets with reflectivities from low to high placed at distances from near to far, the model achieves a maximum of 41.8-fold improvement in relative error for the same target with known reflectivity and a maximum of 36.0-fold improvement in the coefficient of variation for the same target along the whole range of 100 m. Providing accurate and stable characterization of reflectivity in different ranges, the model greatly boosts applications consisting of semantic segmentation and object recognition, such as autonomous driving and environmental monitoring.

Spatial infrared spot target (SIST) discrimination based on infrared radiation sequences (IRSs) can be considered a univariate trending time series classification task. However, due to the complexity of actual scenarios and the limited opportunities for acquiring IRSs, resulting in noise interference, extremely small-scale datasets with imbalanced distribution of classes and widely varying sequence lengths range from a few hundred to several thousand time steps. Current research is primarily based on idealized simulation datasets, resulting in a performance gap when applied to actual applications. To address these issues, firstly, we construct a simulation dataset tailored to the challenges of actual scenarios. Secondly, we design a practical data preprocessing method to achieve uniform sequence length, coarse alignment of shapelets and filtering while preserving key points. Thirdly, we propose a residual network Res-LK-SLR for IRS classification based on large kernels (LKs, providing long-term dependence) and shapelet-level representations (SLRs, where the hidden layer features are aligned with the learned high-level representations to obtain the optimal segmentation and generate shapelet-level representations). Additionally, we conduct extensive evaluations and validations on both the simulation dataset and 18 UCR time series classification datasets. The results demonstrate the effectiveness and generalization ability of our proposed Res-LK-SLR.

Understanding the mechanisms of glacial surging is crucial, as surges can lead to severe hazards and significantly impact a glacier’s mass balance. We used various remote sensing data to investigate the surge of Garmo Glacier in the western Pamir. Our findings indicate that the glacier surged between 27 April and 30 September 2022, with peak speeds reaching 8.3 ± 0.03 m d−1. During April 2020 and September 2022, the receiving zone thickened by 37.9 ± 0.55 m, while the reservoir zone decreased by 35.2 ± 0.55 m on average. The velocity decomposition suggests that this meltwater gradually warmed the glacier bed, accelerating the glacier during the pre-surge phase. During the surge, substantial drainage events coincided with sharp deceleration, ultimately halting the surge and suggesting hydrological control. Extreme climate events may not immediately trigger glacial surges; they can substantially impact glacial surging processes over an extended period.

A time–space shift method was recently developed for relocating the best ensemble member predicted tornado vortex to the radar-observed location, aiming to improve the model’s initial condition and subsequent prediction of tornadoes. To further improve tornado prediction, a variational method for analyzing vortex flows, referred to as VF-Var, is used in this paper to retrieve high-resolution vortex winds from the earliest radar volume scan of tornado and the retrieved vortex winds are then assimilated as “observations” after the vortex relocation. The previous three-step method is also adaptively modified to estimate the tornado vortex center location, denoted by xc ≡ (xc, yc) as a continuous function of height z and time t, from the earliest two consecutive radar volume scans of the tornado, so the estimated xc(z, t) can have the VF-Var required accuracy for retrieving high-resolution vortex winds and the retrieved vortex winds can be assimilated as “observations” with a minimized observation latency. This approach, combined with vortex relocation, is applied to the 20 May 2013 Oklahoma Newcastle–Moore tornado, and is shown to be very effective in further improving the tornado intensity prediction and the continuity of predicted tornado track. Although assimilating the retrieved high-resolution vortex winds after the vortex relocation does not greatly affect the overall trajectory of the predicted tornado track, it proves highly beneficial.

Introducing artificial intelligence techniques in agriculture offers new opportunities for improving crop management, such as in coffee plantations, which constitute a complex agroforestry environment. This paper presents a comparative study of three deep learning frameworks: Deep Forest, RT-DETR, and Yolov9, customized for coffee plant detection and trained from images with a high spatial resolution (cm/pix). Each frame had dimensions of 640 × 640 pixels acquired from passive RGB sensors onboard a UAS (Unmanned Aerial Systems) system. The image set was structured and consolidated from UAS-RGB imagery acquisition in six locations along the Central Valley, Costa Rica, through automated photogrammetric missions. It was evidenced that the RT-DETR and Yolov9 frameworks allowed adequate generalization and detection with mAP50 values higher than 90% and mAP5095 higher than 54%, in scenarios of application with data augmentation techniques. Deep Forest also achieved good metrics, but noticeably lower when compared to the other frameworks. RT-DETR and Yolov9 were able to generalize and detect coffee plants in unseen scenarios that include complex forest structures within tropical agroforestry Systems (AFS).

Remote sensing provides an efficient roadmap in geological analysis and interpretation. However, some challenges arise when remote sensing techniques are integrated with machine learning in geological surveys. Factors including irregular spatial distribution, sample imbalance, interclass resemblances, regolith, and geochemical similarities impede geological feature diagnosis, interpretation, and identification across varied remote sensing datasets. To address these limitations, a hybrid-attention-integrated long short-term memory (LSTM) network is employed to diagnose, interpret, and identify lithological feature representations in a remote sensing-based geological analysis using multisource data fusion. The experimental design integrates varied datasets including Sentinel-2A, Landsat-9, ASTER, ALOS PALSAR DEM, and Bouguer anomaly gravity data. The proposed model incorporates a hybrid attention mechanism (HAM) comprising channel and spatial attention submodules. HAM utilizes an adaptive technique that merges global-average-pooled features with max-pooled features, enhancing the model’s accuracy in identifying lithological units. Additionally, a channel separation operation is employed to allot refined channel features into clusters based on channel attention maps along the channel dimension. The comprehensive analysis of results from comparative extensive experiments demonstrates HAM-LSTM’s state-of-the-art performance, outperforming existing attention modules and attention-based models (ViT, SE-LSTM, and CBAM-LSTM). Comparing HAM-LSTM to baseline LSTM, the HAM module’s integrated configurations equip the proposed model to better diagnose and identify lithological units, thereby increasing the accuracy by 3.69%.

This study aims to address the challenges associated with rockfall assessment and monitoring, focusing on the coastal cliffs of “Navagio Shipwreck Beach” in Zakynthos. A complete time-series analysis was conducted using state-of-the-art methodologies including a 2020 survey using unmanned aerial systems (UASs) and two subsequent surveys, incorporating terrestrial laser scanning (TLS) and UAS survey techniques in 2023. Achieving high precision and accuracy in georeferencing involving direct georeferencing, the utilization of pseudo ground control points (pGCPs), and integrating post-processing kinematics (PPK) with global navigation satellite system (GNSS) permanent stations’ RINEX data is necessary for co-registering the multitemporal models effectively. For the change detection analysis, UAS surveys were utilized, employing the multiscale model-to-model cloud comparison (M3C2) algorithm, while TLS data were used in a validation methodology due to their very high-resolution model. The synergy of these advanced technologies and methodologies offers a comprehensive understanding of rockfall dynamics, aiding in effective assessment and monitoring strategies for coastal cliffs prone to rockfall risk.

The Tarim River Basin (TRB) faces significant ecological challenges due to global warming, making it essential to understand the changes in the climates of its sub-basins for effective management. With this aim, data from national meteorological stations, ERA5_Land, and climate indices from 1961 to 2021 were used to analyze the temperature and precipitation variations in the TRB and its sub-basins and to assess their climate sensitivity. Our results showed that (1) the annual mean temperature increased by 0.2 °C/10a and precipitation increased by 7.1 mm/10a between 1961 and 2021. Moreover, precipitation trends varied significantly among the sub-basins, with that in the Aksu River Basin increasing the most (12.9 mm/10a) and that in the Cherchen River Basin increasing the least (1.9 mm/10a). Moreover, ERA5_Land data accurately reproduced the spatiotemporal patterns of temperature (correlation 0.92) and precipitation (correlation 0.72) in the TRB. (2) Empirical Orthogonal Function analysis identified the northern sections of the Kaidu, Weigan, and Yerqiang river basins as centers of temperature sensitivity and the western part of the Kaidu and Cherchen River Basin as the center of precipitation sensitivity. (3) Global warming is closely correlated with sub-basin temperature (correlation above 0.5) but weakly correlated with precipitation (correlation 0.2~0.5). TRB temperatures were found to have a positive correlation with AMO, especially in the Hotan, Kashgar, and Aksu river basins, and a negative correlation with AO and NAO, particularly in the Keriya and Hotan river basins. Precipitation correlations between the climate indices were complex and varied across the different basins.

Citation: Mercogliano, F.; Barone, A.; D'Auria, L.; Castaldo, R.; Silvestri, M.; Bellucci Sessa, E.; Caputo, T.; Stroppiana, D.; Caliro, S.; Minopoli, C.; et al. Thermal Patterns at the Campi Flegrei Caldera Inferred from Satellite Data and Independent Component Analysis. Remote Sens. 2024, 16, 4615. Abstract: In volcanic regions, the analysis of Thermal InfraRed (TIR) satellite imagery for Land Surface Temperature (LST) retrieval is a valid technique to detect ground thermal anomalies. This allows us to achieve rapid characterization of the shallow thermal field, supporting ground surveillance networks in monitoring volcanic activity. However, surface temperature can be influenced by processes of different natures, which interact and mutually interfere, making it challenging to interpret the spatio-temporal variations in the LST parameter. In this paper, we use a workflow to detect the main thermal patterns in active volcanic areas by analyzing the Independent Component Analysis (ICA) results applied to satellite nighttime TIR imagery time series. We employed the proposed approach to study the surface temperature distribution at the Campi Flegrei caldera volcanic site (Southern Italy, Naples) during the 2013-2022 time interval. The results revealed the contribution of four main distinctive thermal patterns, which reflect the endogenous processes occurring at the Solfatara crater, the environmental processes affecting the Agnano plain, the unique microclimate of the Astroni crater, and the morphoclimatic aspects of the entire volcanic area.

A new method is proposed for identifying ground clutter in single-polarization radar data based on the gated recurrent unit (GRU) neural network. This method needs five independent input variables related to radar reflectivity structure, which are the reflectivity at current tilt, the reflectivity at the upper tilt, the reflectivity at 3.5 km, the echo top height, and the texture of reflectivity at current tilt, respectively. The performance of the new method is compared with that of the traditional method used in the Weather Surveillance Radar 1988-Doppler system in four cases with different scenarios. The results show that the GRU method is more effective than the traditional method in capturing ground clutter, particularly in situations where ground clutter exists at two adjacent elevation angles. Furthermore, in order to assess the new method more comprehensively, 709 radar scans from Nanchang radar in July 2019 and 708 scans from Jingdezhen radar in June 2019 were collected and processed by the two methods, and the frequency map of radar reflectivity exceeding 20 dBZ was analyzed. The results indicate that the GRU method has a stronger ability than the traditional method to identify and remove ground clutter. Meanwhile, the GRU method can also preserve meteorological echoes well.

River width is a crucial parameter that correlates and reflects the hydrological, geomorphological, and ecological characteristics of the channel. However, the width data with high spatial resolution is limited owing to the difficulties in extracting channel width under complex and variable riverine surroundings. To address this issue, we aimed to develop an automatic framework specifically for delineating river channels and measuring the bankfull widths at small spatial intervals along the channel. The DeepLabV3+ Convolutional Neural Network (CNN) model was employed to accurately delineate channel boundaries and a Voronoi Diagram approach was complemented as the river width algorithm (RWA) to calculate river bankfull widths. The CNN model was trained by images across four river types and performed well with all the evaluating metrics (mIoU, Accuracy, F1-score, and Recall) higher than 0.97, referring to the accuracy over 97% in prediction. The RWA outperformed other existing river width calculation methods by showing lower errors. The application of the framework in the Lillooet River, Canada, presented the capacity of this methodology to obtain detailed distributions of hydraulic and hydrological parameters, including flow resistance, flow energy, and sediment transport capacity, based on high-resolution channel widths. Our work highlights the significant potential of the newly developed framework in acquiring high-resolution channel width information and characterizing fluvial dynamics based on these widths along river channels, which contributes to facilitating cost-effective integrated river management.

Forest ecosystems play an essential role in ecological balance, supporting biodiversity and climate change mitigation. These ecosystems are crucial not only for ecological stability but also for the local economy. Performing a tree census at a country scale via traditional methods is resource-demanding, error-prone, and requires significant effort by a large number of experts. While emerging technologies such as satellite imagery and AI provide the means for achieving promising results in this task with less effort, considerable effort is still required by experts to annotate hundreds or thousands of images. This study introduces a novel methodology for a tree census classification system which leverages historical and partially labeled data, employs probabilistic data imputation and a weakly supervised training technique, and thus achieves state-of-the-art precision in classifying the dominant tree species of Cyprus. A crucial component of our methodology is a ResNet50 model which takes as input high spatial resolution satellite imagery in the visible band and near-infrared band, as well as topographical features. By applying a multimodal training approach, a classification accuracy of 90% among nine targeted tree species is achieved.

Hyperspectral remote sensing images offer a unique opportunity to quickly monitor water depth, but how to utilize the enriched spectral information and improve its spatial resolution remains a challenge. We proposed a water depth estimation framework to improve spatial resolution using deep learning and four inversion methods and verified the effectiveness of different super resolution and inversion methods in three waterbodies based on HJ-2 hyperspectral images. Results indicated that it was feasible to use HJ-2 hyperspectral images with a higher spatial resolution via super resolution methods to estimate water depth. Deep learning improves the spatial resolution of hyperspectral images from 48 m to 24 m and shows less information loss with peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and spectral angle mapper (SAM) values of approximately 37, 0.92, and 2.42, respectively. Among four inversion methods, the multilayer perceptron demonstrates superior performance for the water reservoir, achieving the mean absolute error (MAE) and the mean absolute percentage error (MAPE) of 1.292 m and 22.188%, respectively. For two rivers, the random forest model proves to be the best model, with an MAE of 0.750 m and an MAPE of 10.806%. The proposed method can be used for water depth estimation of different water bodies and can improve the spatial resolution of water depth mapping, providing refined technical support for water environment management and protection.

This study aimed to determine whether canopy and air temperature difference (ΔT) as an existing simple normalizing index can be used to detect an increase in canopy temperature induced by soil drought in urban parks, regardless of the unique energy balance and three-dimensional (3D) structure of urban trees. Specifically, we used a thermal infrared camera to measure the canopy temperature of Zelkova serrata trees and compared the temporal variation of ΔT to that of environmental factors, including solar radiation, wind speed, vapor pressure deficit, and soil water content. Normalization based on a 3D energy-balance model was also performed and used for comparison with ΔT. To represent the 3D structure, a terrestrial light detection and ranging-derived 3D tree model was used as the input spatial data. The temporal variation in ΔT was similar to that of the index derived using the energy-balance model, which considered the 3D structure of trees and 3D radiative transfer, with a correlation coefficient of 0.85. In conclusion, the thermal-image-based ΔT performed comparably to an index based on the 3D energy-balance model and detected the increase in canopy temperature because of the reduction in soil water content for Z. serrata trees in an urban environment.

Soil and plant analyzer development (SPAD) value and leaf nitrogen concentration (LNC) based on dry weight are important indicators affecting rice yield and quality. However, there are few reports on the use of machine learning algorithms based on hyperspectral monitoring to synchronously predict SPAD value and LNC of indica rice. Meixiangzhan No. 2, a high-quality indica rice, was grown at different nitrogen rates. A hyperspectral device with an integrated handheld leaf clip-on leaf spectrometer and an internal quartz-halogen light source was conducted to monitor the spectral reflectance of leaves at different growth stages. Linear regression (LR), random forest (RF), support vector regression (SVR), and gradient boosting regression tree (GBRT) were employed to construct models. Results indicated that the sensitive bands for SPAD value and LNC were displayed to be at 350–730 nm and 486–727 nm, respectively. Normalized difference spectral indices NDSI (R497, R654) and NDSI (R729, R730) had the strongest correlation with leaf SPAD value (R = 0.97) and LNC (R = −0.90). Models constructed via RF and GBRT were markedly superior to those built via LR and SVR. For prediction of leaf SPAD value and LNC, the model constructed with the RF algorithm based on whole growth periods of spectral reflectance performed the best, with R2 values of 0.99 and 0.98 and NRMSE values of 2.99% and 4.61%. The R2 values of 0.98 and 0.83 and the NRMSE values of 4.88% and 12.16% for the validation of leaf SPAD value and LNC were obtained, respectively. Results indicate that there are significant spectral differences associated with SPAD value and LNC. The model built with RF had the highest accuracy and stability. Findings can provide a scientific basis for non-destructive real-time monitoring of leaf color and precise fertilization management of indica rice.

Rivers are crucial pathways for transporting organic carbon from land to ocean, playing a vital role in the global carbon cycle. Dissolved organic carbon (DOC) and chromophoric dissolved organic matter (CDOM) are major components of dissolved organic matter and have significant impacts on maintaining the stability of river ecosystems and driving the global carbon cycle. In this study, the in situ samples of aCDOM(355) and DOC collected along the main stream of the Songhua River were matched with Sentinel-2 imagery. Multiple linear regression and five machine learning models were used to analyze the data. Among these models, XGBoost demonstrated a superior, highly stable performance on the validation set (R2 = 0.85, RMSE = 0.71 m−1). The multiple linear regression results revealed a strong correlation between CDOM and DOC (R2 = 0.73), indicating that CDOM can be used to indirectly estimate DOC concentrations. Significant seasonal variations in the CDOM distribution in the Songhua River were observed: aCDOM(355) in spring (6.23 m−1) was higher than that in summer (5.3 m−1) and autumn (4.74 m−1). The aCDOM(355) values in major urban areas along the Songhua River were generally higher than those in non-urban areas. Using the predicted DOC values and annual flow data at the sites, the annual DOC flux in Harbin was calculated to be approximately 0.2275 Tg C/Yr. Additionally, the spatial variation in annual CDOM was influenced by both natural changes in the watershed and human activities. These findings are pivotal for a deeper understanding of the role of river systems in the global carbon cycle.