Search Results (35)

Land cover mapping is essential for territorial management due to its links with ecological, hydrological, climatic, and socioeconomic processes. Traditional methods use discrete classes per pixel, but this study proposes estimating cover fractions with Sentinel-2 imagery (20 m) and AI. We employed the French Land cover from Aerospace ImageRy (FLAIR) dataset (810 km² in France, 19 classes), with labels co-registered with Sentinel-2 to derive precise fractional proportions per pixel. From these references, we generated training sets combining spectral bands, derived indices, and auxiliary data (climatic and temporal variables). Various machine learning models—including XGBoost three deep neural network (DNN) architectures with different depths, and convolutional neural networks (CNNs)—were trained and evaluated to identify the optimal configuration for fractional cover estimation. Model validation on the test set employed RMSE, MAE, and R² metrics at both pixel level (20 m Sentinel-2) and scene level (100 m FLAIR). The training set integrating spectral bands, vegetation indices, and auxiliary variables yielded the best MAE and RMSE results. Among all models, DNN2 achieved the highest performance, with a pixel-level RMSE of 13.83 and MAE of 5.42, and a scene-level RMSE of 4.94 and MAE of 2.36. This fractional approach paves the way for advanced remote sensing applications, including continuous cover-change monitoring, carbon footprint estimation, and sustainability-oriented territorial planning. Full article

Accurate sea clutter modeling is crucial for clutter suppression in edge radar processing. On resource-constrained edge radar platforms, spatiotemporal statistics, together with device-level computation and memory limits, hinder the learning of representative clutter features. This study presents a transformer-based generative adversarial model for sea clutter modeling. The core design of this work uses axial attention to factorize self-attention along pulse and range, preserving long-range dependencies under a reduced attention cost. It also introduces a two-dimensional variable-length spatiotemporal window that retains temporal and spatial coherence across observation lengths. Extensive experiments are conducted to verify the efficacy of the proposed method with quantitative criteria, including a cosine similarity score, spectral-parameter error, and amplitude–distribution distances. Compared with CNN-based GAN, the proposed model achieves a high consistency with real clutter in marginal amplitude distributions, spectral characteristics, and spatiotemporal correlation patterns, while incurring a lower cost than standard multi-head self-attention. The experimental results show that the proposed method achieves improvements of 9.22% and 7.8% over the traditional AR and WaveGAN methods in terms of the similarity metric, respectively. Full article

Land degradation in drylands represents a critical environmental challenge, with persistent bare soil serving as a key indicator of ecosystem vulnerability, including in the Caatinga biome. This study maps and analyzes the spatial and temporal dynamics of persistent bare soils over three decades using multi-temporal remote sensing data. We applied Spectral Mixture Analysis (SMA), temporal metrics, and machine learning classifiers within Google Earth Engine to process long-term Landsat datasets and to derive the Normalized Difference Fraction Index Adjusted (NDFIa). The results indicate a widespread increase in bare soil, with over 63% of mapped hexagons showing expansion, particularly in the São Francisco Basin. Peaks in soil exposure coincided with severe drought events, highlighting the link between climate variability and land degradation. Moreover, abandoned agricultural lands and pasturelands emerged as the dominant contributors to persistent bare soils. These findings reinforce the need for targeted policies to mitigate land degradation and to promote sustainable land management in semi-arid ecosystems. This research provides a robust framework for long-term environmental monitoring in drylands by integrating satellite data with advanced analytical techniques. These advancements support more effective land management and conservation strategies in semi-arid ecosystems. Full article

Accurate crop yield modeling (CYM) is inherently challenging due to the complex, nonlinear, and temporally dynamic interactions of biotic and abiotic factors. Crop traits, which historically capture the cumulative effect of these factors, exhibit functional relationships critical for optimizing productivity. This underscores the necessity of multi-trait-based CYM approaches. Crop growth models enable trait dynamics with reflectance data and spectral indices as proxies for crop health and traits, respectively, to have real-time, spatially explicit monitoring. The Agricultural Production Systems sIMulator was calibrated to simulate multiple traits across the growth season based on geo-tagged wheat field ground information. Reflectance and spectral indices were processed for the geo-tagged fields across temporal observations to enable real-time, spatially explicit monitoring. Based on these parameters, this study addresses a critical gap in existing CYM frameworks by proposing a machine learning-based model that synergized multiple crop traits with reflectance and spectral indices to generate site-specific yield estimates. The performance evaluation revealed that the Long Short-Term Memory (LSTM) model achieved superior accuracy for the integrated parameters (RMSE = 250.68 kg/ha, MAE = 193.76 kg/ha, and R² = 0.84), followed by traits alone. The Random Forest model followed the LSTM model, with an RMSE = 293.56 kg/ha, MAE = 230.68 kg/ha, and R² = 0.78 for integrated parameters, and an RMSE = 291.73 kg/ha, MAE = 223.17 kg/ha, and R² = 0.78 for crop traits. The superior prediction demonstrated the dominant role of multiple crop traits with satellite-derived reflectance metrics to develop robust CYM frameworks capable of capturing intra- and inter-field yield variability. Full article

Multi-source remote sensing fusion and machine learning are effective tools for forest monitoring. This study aimed to analyze various fusion techniques, their application with machine learning algorithms, and their assessment in estimating forest type and aboveground biomass (AGB). A keyword search across Web of Science, Science Direct, and Google Scholar yielded 920 articles. After rigorous screening, 72 relevant articles were analyzed. Results showed a growing trend in optical and radar fusion, with notable use of hyperspectral images, LiDAR, and field measurements in fusion-based forest monitoring. Machine learning algorithms, particularly Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), leverage features from fused sources, with proper variable selection enhancing accuracy. Standard evaluation metrics include Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Overall Accuracy (OA), User’s Accuracy (UA), Producer’s Accuracy (PA), confusion matrix, and Kappa coefficient. This review provides a comprehensive overview of prevalent techniques, data sources, and evaluation metrics by synthesizing current research and highlighting data fusion’s potential to improve forest monitoring accuracy. The study underscores the importance of spectral, topographic, textural, and environmental variables, sensor frequency, and key research gaps for standardized evaluation protocols and exploration of multi-temporal fusion for dynamic forest change monitoring. Full article

Although significant progress has been made in the remote sensing extraction of mangroves, research at the species level remains relatively limited. Kandelia obovata is a dominant mangrove species and is frequently used in ecological restoration projects in China. However, owing to the fragmented distribution of K. obovata within mixed mangrove communities and the significant spectral and textural similarities among mangrove species, accurately extracting large-scale K. obovata-based remote sensing data remains a challenging task. In this study, we conducted extensive field surveys and developed a comprehensive sampling database covering K. obovata and other mangrove species across mangrove-distributing areas in China. We identified the optimal bands for extracting K. obovata by utilizing time-series remote sensing data from Sentinel-1 and Sentinel-2, along with the Google Earth Engine (GEE), and proposed a method for extracting K. obovata communities. The main conclusions are as follows: (1) The spectral-temporal variability characteristics of the blue and red-edge bands play a crucial role in the identification of K. obovata communities. The 90th percentile metric of the blue wavelength band ranks first in importance, while the 75th percentile metric of the blue wavelength band ranks second; (2) This method of remote sensing extraction using spectral-temporal variability metrics with time-series optical and radar remote sensing data offers significant advantages in identifying the K. obovata species, achieving a producer’s accuracy of up to 94.6%; (3) In 2018, the total area of pure K. obovata communities in China was 4825.97 ha; (4) In the southern provinces of China, Guangdong Province has the largest K. obovata community area, while Macau has the smallest. This research contributes to the understanding of mangrove ecosystems and provides a methodological framework for monitoring K. obovata and other coastal vegetation using advanced remote sensing technologies. Full article

Increased wildfire activity is the most significant natural disturbance affecting forest ecosystems as it has a strong impact on their natural recovery. This study aimed to investigate how burn severity (BS) levels and climate factors, including land surface temperature (LST) and precipitation variability (Pr), affect forest recovery in the Middle Volga region of the Russian Federation. It provides a comprehensive analysis of post-fire forest recovery using Landsat time-series data from 2000 to 2023. The analysis utilized the LandTrendr algorithm in the Google Earth Engine (GEE) cloud computing platform to examine Normalized Burn Ratio (NBR) spectral metrics and to quantify the forest recovery at low, moderate, and high burn severity (BS) levels. To evaluate the spatio-temporal trends of the recovery, the Mann–Kendall statistical test and Theil–Sen’s slope estimator were utilized. The results suggest that post-fire spectral recovery is significantly influenced by the degree of the BS in affected areas. The higher the class of BS, the faster and more extensive the reforestation of the area occurs. About 91% (40,446 ha) of the first 5-year forest recovery after the wildfire belonged to the BS classes of moderate and high severity. A regression model indicated that land surface temperature (LST) plays a more critical role in post-fire recovery compared to precipitation variability (Pr), accounting for approximately 65% of the variance in recovery outcomes. Full article

Aboveground biomass (AGB) is a vital indicator for studying carbon sinks in forest ecosystems. Semiarid forests harbor substantial carbon storage but received little attention due to the high spatial–temporal heterogeneity that complicates the modeling of AGB in this environment. This study assessed the performance of different data sources (annual monthly time-series radar was Sentinel-1 [S1]; annual monthly time series optical was Sentinel-2 [S2]; and single-temporal airborne light detection and ranging [LiDAR]) and seven prediction approaches to map AGB in the semiarid forests on the border between Gansu and Qinghai Provinces in China. Five experiments were conducted using different data configurations from synthetic aperture radar backscatter, multispectral reflectance, LiDAR point cloud, and their derivatives (polarimetric combination indices, texture information, vegetation indices, biophysical features, and tree height- and canopy-related indices). The results showed that S2 acquired better prediction (coefficient of determination [R²]: 0.62–0.75; root mean square error [RMSE]: 30.08–38.83 Mg/ha) than S1 (R²: 0.24–0.45; RMSE: 47.36–56.51 Mg/ha). However, their integration further improved the results (R²: 0.65–0.78; RMSE: 28.68–35.92 Mg/ha). The addition of single-temporal LiDAR highlighted its structural importance in semiarid forests. The best mapping accuracy was achieved by XGBoost, with the metrics from the S2 and S1 time series and the LiDAR-based canopy height information being combined (R²: 0.87; RMSE: 21.63 Mg/ha; relative RMSE: 14.45%). Images obtained during the dry season were effective for AGB prediction. Tree-based models generally outperformed other models in semiarid forests. Sequential variable importance analysis indicated that the most important S1 metric to estimate AGB was the polarimetric combination indices sum, and the S2 metrics were associated with red-edge spectral regions. Meanwhile, the most important LiDAR metrics were related to height percentiles. Our methodology advocates for an economical, extensive, and precise AGB retrieval tailored for semiarid forests. Full article

We propose a novel mono-temporal framework with a physical basis and ecological consistency to retrieve fire severity at very high spatial resolution. First, we sampled the Composite Burn Index (CBI) in 108 field plots that were subsequently surveyed through unmanned aerial vehicle (UAV) flights. Then, we mimicked the field methodology for CBI assessment in the remote sensing framework. CBI strata were identified through individual tree segmentation and geographic object-based image analysis (GEOBIA). In each stratum, wildfire ecological effects were estimated through the following methods: (i) the vertical structural complexity of vegetation legacies was computed from 3D-point clouds, as a proxy for biomass consumption; and (ii) the vegetation biophysical variables were retrieved from multispectral data by the inversion of the PROSAIL radiative transfer model, with a direct physical link with the vegetation legacies remaining after canopy scorch and torch. The CBI scores predicted from UAV ecologically related metrics at the strata level featured high fit with respect to the field-measured CBI scores (R² > 0.81 and RMSE < 0.26). Conversely, the conventional retrieval of fire effects using a battery of UAV structural and spectral predictors (point height distribution metrics and spectral indices) computed at the plot level provided a much worse performance (R² = 0.677 and RMSE = 0.349). Full article

Biophysical variables play a crucial role in understanding phenological stages and crop dynamics, optimizing ultimate agricultural practices, and achieving sustainable crop yields. This study examined the effectiveness of the Sentinel-2 Biophysical Processor (S2BP) in accurately estimating crop dynamics descriptors, including fractional vegetation cover (FVC), leaf area index (LAI), leaf chlorophyll a and b (LC_ab), and canopy water content (CWC). The evaluation was conducted using estimation quality indicators (EQIs) and comprehensive ground throughout the entire growing season at the field scale. To identify soil and vegetation pixels, the spectral unmixing technique was employed. According to the EQIs, the best retrievals were obtained for FVC in around 99.9% of the 23,976 pixels that were analyzed during the growth season. For LAI, LC_ab, and CWC, over 60% of the examined pixels had inputs that were out-of-range. Furthermore, in over 35% of the pixels, the output values for LC_ab and CWC were out-of-range. The FVC, LAI, and LC_ab estimates agreed well with ground measurements (R² = 0.62–0.85), whereas a discrepancy was observed for CWC estimates when compared with ground measurements (R² = 0.51). Furthermore, the uncertainties of FVC, LAI, LC_ab, and CWC estimates were 0.09, 0.81 m²/m², 60.85 µg/cm², and 0.02 g/cm² through comparisons to ground FVC, LAI, C_ab, and CWC measurements, respectively. Considering EQIs and uncertainty metrics, the order of the estimation accuracy of the four variables was FVC > LAI > LC_ab > CWC. Our analysis revealed that temporal variations of FVC, LAI, and LC_ab were primarily driven by field-scale events like sowing date, growing period, and harvesting time, highlighting their sensitivity to agricultural practices. The robustness of S2BP results could be enhanced by implementing a pixel identification algorithm, like embedding spectral unmixing. Overall, this study provides detailed, pixel-by-pixel insights into the performance of S2BP in estimating FVC, LAI, LC_ab, and CWC, which are crucial for monitoring crop dynamics in precision agriculture. Full article

The vertical distribution of light and its spectral composition are critical factors influencing numerous physical, chemical, and biological processes within the oceanic water column. In this study, we present vertically resolved models of downwelling irradiance (ED) at three different wavelengths and photosynthetically available radiation (PAR) on a global scale. These models rely on the SOCA (Satellite Ocean Color merged with Argo data to infer bio-optical properties to depth) methodology, which is based on an artificial neural network (ANN). The new light models are trained with light profiles (ED/PAR) acquired from BioGeoChemical-Argo (BGC-Argo) floats. The model inputs consist of surface ocean color radiometry data (i.e., $R_{rs}$, PAR, and $k_d\left(490\right)$) derived by satellite and extracted from the GlobColour database, temperature and salinity profiles originating from BGC-Argo, as well as temporal components (day of the year and local time in cyclic transformation). The model outputs correspond to ED profiles at the three wavelengths of the BGC-Argo measurements (i.e., 380, 412, and 490 nm) and PAR profiles. We assessed the retrieval of light profiles by these light models using three different datasets: BGC-Argo profiles that were not used for the training (i.e., 20% of the initial database); data from four independent BGC-Argo floats that were used neither for the training nor for the 20% validation dataset; and the SeaBASS database (in situ data collected from various oceanic cruises). The light models show satisfactory predictions when thus compared with real measurements. From the 20% validation database, the light models retrieve light variables with high accuracies (root mean squared error (RMSE)) of 76.42 μmol quanta m⁻² s⁻¹ for PAR and 0.04, 0.08, and 0.09 W m⁻² nm⁻¹ for ED380, ED412, and ED490, respectively. This corresponds to a median absolute percent error (MAPE) that ranges from 37% for ED490 and PAR to 39% for ED380 and ED412. The estimated accuracy metrics across these three validation datasets are consistent and demonstrate the robustness and suitability of these light models for diverse global ocean applications. Full article

Forests, commonly viewed as the Earth’s lungs, play a crucial role in mitigating greenhouse gas emissions, regulating the globe, and maintaining ecological equilibrium. The assessment of aboveground biomass (AGB) serves as a pivotal indicator for evaluating forest quality. By integrating remote sensing images with a small number of ground-measured samples to map, forest AGBs can significantly reduce time and labor costs. Current research mainly focuses on improving the accuracy of mapping forest AGBs, such as integrating multiple-sensors remote sensing data and models. However, due to uncertainties associated with remote sensing images and complexities inherent in forest structures, the accuracy of mapping forest AGBs is constrained by both the quantity and distribution of ground samples available. The development of transfer learning methods can fully utilize ground-based measurement data and enable the application of samples across regions and time. To evaluate the potential of transfer learning methods in mapping forest AGBs, this study conducted a spatial–temporal transfer of spectral variables (SVs) and prediction models (PMs) using a direct-push transfer method, and a new evaluation metric, relative change of R-squared (RCRS), was proposed to assess the transferability of SVs and PMs. The results showed that the transferability of SVs and PMs in the spatial target domain is obviously greater than that in the temporal target domain. Compared to the temporal target domain, the RCRS for transfer SVs in the spatial target domain was lower by 20.89 (oak) and 20.88 (Chinese fir) and for transfer PMs by 24.16 (oak) and 24.79 (Chinese fir). Tree species is also one of the main factors affecting the spatial and temporal transfer of SVs, and it is challenging to transfer SVs between different tree species. The results also show that nonparametric models have better generalization performance, and their transferability is much greater than that of parametric models. Full article

The images of the Sentinel-2 constellation can help the verification process of farmers’ declarations, providing, among other things, accurate spatial explicit maps of the agricultural land cover. The aim of the study is to design, develop, and evaluate two deep learning (DL) architectures tailored for agricultural land cover and crop type mapping. The focus is on a detailed class scheme encompassing fifteen distinct classes, utilizing Sentinel-2 imagery acquired on a monthly basis throughout the year. The study’s geographical scope covers a diverse rural area in North Greece, situated within southeast Europe. These architectures are a Temporal Convolutional Neural Network (CNN) and a combination of a Recurrent and a 2D Convolutional Neural Network (R-CNN), and their accuracy is compared to the well-established Random Forest (RF) machine learning algorithm. The comparative approach is not restricted to simply presenting the results given by classification metrics, but it also assesses the uncertainty of the classification results using an entropy measure and the spatial distribution of the classification errors. Furthermore, the issue of sampling strategy for the extraction of the training set is highlighted, targeting the efficient handling of both the imbalance of the dataset and the spectral variability of instances among classes. The two developed deep learning architectures performed equally well, presenting an overall accuracy of 90.13% (Temporal CNN) and 90.18% (R-CNN), higher than the 86.31% overall accuracy of the RF approach. Finally, the Temporal CNN method presented a lower entropy value (6.63%), compared both to R-CNN (7.76%) and RF (28.94%) methods, indicating that both DL approaches should be considered for developing operational EO processing workflows. Full article

The Gulf of California (GC) is the only interior sea in the Eastern Pacific Ocean and is the most important fishing area in the northwestern region of the Mexican Pacific. This study focuses on the oceanographic variability of the GC, including its southern portion, which is an area with a high flow of energy and exchange of properties with the Pacific Ocean (PO), in order to determine its role in physical–biological cycles and climate change. The purpose of this work is to analyze the sea surface temperature (SST) and chlorophyll a concentration (Chl-a) during the period from 1998–2022 as indicators of long-term physical and biological processes, oceanographic variability, and primary production in the GC. In total, 513 subareas in the GC were analyzed, and a cluster analysis was applied to identify similar areas in terms of SST and Chl-a via the K-means method and using the silhouette coefficient (>0.5) as a metric to validate the clusters obtained. The trends of the time series of both variables were analyzed, and a fast Fourier analysis was performed to evaluate cycles in the series. A descriptive analysis of the SST and Chl-a series showed that the SST decreased from south to north. Six bioregions were identified using a combined of both SST and Chl-a data. The spectral analysis of the SST showed that the main frequencies in the six bioregions were annual and interannual (3–7 years), and the frequencies of their variations were associated with basin-level weather events, such as El Niño and La Niña. The SST in the GC showed a heating trend at an annual rate of ~0.036 °C (~0.73 °C in 20 years) and a decrease in Chl-a at an annual rate of ~0.012 mg/m³ (~0.25 mg/m³ in 20 years), with potential consequences for communities and ecosystems. Additionally, cycles of 10–13 and 15–20 years were identified, and the 10–13-year cycle explained almost 40–50% of the signal power in some regions. Moreover, mesoscale features (eddies and filaments) were identified along the GC, and they were mainly associated with the clusters of the SST. All these spatial and temporal variabilities induce conditions that generate different habitats and could explain the high biodiversity of the GC. If the warming trend of the SST and the decreasing trend of the Chl-a continue in the long term, concerns could be raised, as they can have important effects on the dynamics of this important marine ecosystem, including habitat loss for numerous native species, declines in the catches of the main fishery resources, and, consequently, support for the arrival of harmful invasive species. Full article

This paper presents empirical models developed through stepwise multiple linear regression to estimate the live fuel moisture content (LFMC) in a Mediterranean area. The models are based on LFMC data measured in 50 field plots, considering four groups with similar bioclimatic characteristics and vegetation types (trees and shrubs). We also applied a species-specific LFMC model for Rosmarinus officinalis in plots with this dominant species. Spectral indices extracted from Sentinel-2 images and their averages over the study time period in each plot with a spatial resolution of 10 m were used as predictors, together with interpolated meteorological, topographic, and seasonal variables. The models achieved adjusted R² values ranging between 52.1% and 74.4%. Spatial and temporal variations of LFMC in shrub areas were represented on a map. The results highlight the feasibility of developing satellite-derived LFMC operational empirical models in areas with various vegetation types and taking into account bioclimatic zones. The adjustment of data through GAM (generalized additive models) is also addressed in this study. The different error metrics obtained reflect that these models provided a better fit (most adjusted R² values ranged between 65% and 74.1%) than the linear models, due to GAMs being more versatile and suitable for addressing complex problems such as LFMC behavior. Full article