The monitoring of vegetation dates back to the 1970s, but ever since, it has also been a very crucial task for providing a good quality and quantity of food supplies. In the following article, the authors present their technological scheme and results from processing a prediction yield model for agricultural fields in Bulgaria. The team used open public data in a vector format presenting all crop fields subsided by the state and machine learning techniques for satellite image classification for the territory of the whole country. The authors decided to use optical data from Sentinel-2. Multitemporal data were used in order to train a prediction model to help farmers predict their crop yield. The authors also focus on the struggles using big data for the whole country and any ambiguities throughout the computation process. The basic technological scheme is as follows: data preparation (vector, meta, and imagery data), defining a proper coordinate system for the whole country, machine learning application using the defined region of interest and a segmentation network, and classification using 25 predefined classes. The results of this paper aim to present a reliable technological method for both farmers and the state to monitor the current state of certain crops and to predict with high accuracy any future yield.

This study investigated the utility of Sentinel-2 spectral data for estimating the leaf area index (LAI), and the leaf and canopy chlorophyll content of maize at different growth stages. Vegetation indices based on the visible–near infrared and red-edge regions of the spectrum were computed from Sentinel-2 imagery acquired within one or two days of field data collection. Results showed that the green chlorophyll index (CI_green) and red-edge chlorophyll index (CI_red-edge)_, using the red-edge variant centred at 705 nm, consistently showed a stronger relationship with maize LAI with r² values of 0.65 and 0.63 during the early stages of growth, respectively, and r² values of 0.79 and 0.81 during the tassel stage, respectively. Regarding canopy chlorophyll content, the results indicated the spectral advantage of the Sentinel-2 sensor with the presence of two red-edge bands for the continuous monitoring of maize chlorophyll content. Overall, the study results indicated that maize biophysical variables can be monitored at the satellite level and with acceptable accuracy when using Sentinel-2 data.

Estimating crop biophysical variables is essential to farmers for assessing and monitoring crop growth at different stages. A comprehensive comparison and integration of retrieval methods is needed for accurately estimating crop biophysical variables such as chlorophyll over a heterogenous sweet potato canopy. In this paper, we explored the capabilities of PROSAIL radiative transfer models (RTMs) applied to 5 cm resolution drone multispectral imagery to retrieve the leaf chlorophyll content (LCC) of over 20 sweet potato cultivars at the peak growth stage. Various vegetation indices spanning broadband, leaf pigment, and narrowband indices were tested on numerous PROSAIL simulation databases in order to optimize their retrieval performance. The results show that the most accurate retrievals of LCC from drone data over 20 sweet potato cultivars wereachieved by integrating larger (11000) PROSAIL simulations with broadband indices, particularly the enhanced vegetation index (EVI) with an R² of 0.85, an RMSE of 5.93 µg/cm², and a RRMSE of 9.98%. This performance was followed by that of narrowband indices, particularly the modified normalized vegetation index (mNDVI), with an R² of 0.84, an RMSE of 5.95 µg/cm², and an RRMSE of 9.91%. Furthermore, a polynomial fitting type model best captured the variability of the LCC compared to the linear model. An attempt to integrate lesser PROSAIL simulations (i.e., of 1500, 5000, and 9000 reflectance samples) with the indices showed deteriorating LCC retrieval performance. These findings suggest that ultra-high-resolution drone imagery may be appropriate for the accurate retrieval and monitoring of LCC over a heterogenous canopy comprising numerous crop cultivars.

Accurate and timely air quality monitoring relies on data integration methods that bridge the spatial coverage of satellite imagery and the temporal granularity of low-cost sensors (LCSs). LCSs provide high-frequency measurements but are prone to noise, while satellite data offer extensive coverage but suffer from coarse temporal resolution and retrieval uncertainties. To address these limitations, we present a novel probabilistic data fusion framework rooted in generic Bayesian filtering. Our approach employs Kalman Filters (KFs) for dynamic state estimation and uncertainty quantification. We enrich the KF state estimation with covariates generated by Land Use Regression (LUR) to incorporate local spatial context.

We fuse nitrogen dioxide (NO2) data from low-cost sensor networks with satellite-derived aerosol optical depth (AOD) measurements from Sentinel-5P, using ground reference data for calibration and validation. Evaluated in the Dublin City area, the preliminary results demonstrate a significant reduction in bias and improved accuracy and precision of air quality estimates.

This framework addresses critical challenges in multi-source data integration, including a lack of a consistent model, resolution mismatches, noise propagation, and bias correction. By bridging global and local scales, it provides actionable insights for air quality management and environmental policy. Our case study in urban environments highlights our framework’s potential for scalable applications in public health and environmental monitoring elsewhere.

The exploitation of earth observation data plays a crucial role in analyzing and tackling some of the most pressing global issues and societal challenges as defined in the United Nations' Sustainable Development Goals, especially in fields like urban planning and monitoring for sustainable cities or climate action and natural disaster response. A recently developed multi-modal automatic change-detection pipeline allows for the hybridization of 2D and 3D satellite data in order to combine uncertainty-aware semantic segmentation maps with detected changes in altitude and to obtain fine-grained information about urbanization or affected buildings in the aftermath of natural disasters like floods, wildfires, and earthquakes. The 2D part is based on modern deep neural networks, which often suffer from under-/overconfident predictions and lack a reliable representation of uncertainty. This work compares different common uncertainty-aware deep neural networks in the context of remote sensing imagery, like approximate Bayesian neural networks, neural network ensembles, and test-time data augmentations. It includes a quantitative evaluation of different widely used uncertainty estimation approaches for modern deep neural networks in the specific use-case of leveraging earth observation data for urban monitoring and natural disaster response. The evaluation examines well-known standard metrics to assess the quality of model calibration, like the Brier Score and the Expected Calibration Error in the context of multi-temporal semantic segmentation and change detection. Although calibration metrics can give insightful information about the reliability of model predictions with the implicit consideration of associated uncertainty estimates, evaluating uncertainty itself remains a challenging task, due to a lack of ground-truth data. Different approaches for a quantitative evaluation of uncertainty estimates have been proposed in the literature, like the “Patch Accuracy vs Patch Uncertainty” metric, which is thus included in the evaluation in order to complete the quantitative comparison of different uncertainty estimation methods for EO change detection.

At the mouth of the Mataquito River, a coastal bar has formed due to the high availability of sediments from the ocean–basin interaction. The evolution of this bar has been marked by disruptive events such as the earthquake followed by the tsunami of February 27th, 2010, frontal systems, and strong ENSO Niño Niña phenomena. The bar erodes and builds up naturally, and in the last decade has been affected by increased storm surges and sediment scarcity due to the mega drought in central Chile since 2010, as well as intensive forestry use in the upper sections of the basin, which reduce the flow of sediment to the ocean and put this landform at risk. The objective of this work is to quantify the multidecadal evolution and spatio-temporal changes of the bar from 1985 to 2024 using Landsat satellite images (5, 7, 8 and 9 and Sentinel 2), ERA5 wave data, and in situ data. The analysis includes the effect of waves and sediment characteristics, thus taking into account the inherent complexity of the coastal environment. The dynamics of the mouth are reconstructed, showing the point at which the bar was turned to the left by the 8.8Mw 2010 earthquake, during which the bar disappeared, showing the force of the tsunami versus the inertia of the river and favouring the growth of an elongated channel with variable width towards the north. In this bar, fluvial avulsion is evident, which promotes coastal erosion since sediments are temporarily retained in the floodplain. The capacity of the existing riverbed to transport all the water and sediments it receives is noteworthy. On the other hand, the subsidence caused by the 2010 earthquake caused ebb and deposition, changing the morphology of the bar.

Fire risk maps are fundamental tools for the management and prevention of forest fires, allowing the identification of the most vulnerable areas and thus facilitating the planning and optimization of firefighting resources. To obtain risk maps, it is necessary to consider propagation risks (slope, orientations, NDVI and NDWI index from Sentinel-2 and backscatter coefficient from Sentinel-1) and ignition risks, where human factors come into play, such as proximity to power lines, picnic areas, and agricultural plots. The combination of these variables allows for a precise analysis of fire risk and the generation of a detailed map of the study area using machine learning.

Quantifying carbon loss after a fire is crucial to assess the environmental impact. In this work, forest mass stocks have been evaluated using LiDAR data and biomass volume-based models to estimate the carbon loss in the Zamora's large fires in 2022 (Spain). The main species affected were forest stands of Quercus ilex, Quercus pyrenaica, Pinus sylvestris and Pinus pinaster. The carbon estimate has been calculated from the severity of the fire, making a final map of carbon stock losses. In this area, a total of approximately 0,5 megatons of carbon were lost, 106.304,10 tons of biomass of Quercus pyrenaica, 340.694,25 tons of Pinus sylvestris and 457.611,79 tons of Pinus pinaster.

The main objective of this study is to develop a clear and accurate methodology to create detailed and updated forest fire risk maps and forest fire carbon loss estimation maps, using remote sensing data in one of the most devastating wildfires that Spain has suffered in recent years. The application of these methodologies and the resulting maps contribute to the design of forest restoration strategies and climate change mitigation policies, promoting a more sustainable management of natural resources.

The coastal towns of Licantén, Maule Region, Chile, were affected by a magnitude 8.8 Mw earthquake on February 27, 2010, followed by a tsunami that caused significant damage and changes in the shoreline position, marking a clear before-and-after point due to this major seismic event. Since 2015, coastal storms have increased in frequency and intensity, exposing these towns to flood risks. Using satellite imagery from Landsat 5 TM, Landsat 7 ETM+, Landsat 8 OLI, Landsat 9 OLI, Sentinel-2A/B, the CoastSat algorithm, ERA5 data, in situ data, and high-resolution Maxar images complemented by UAV flights, this work aims to determine urban expansion rates in areas exposed to tsunami flood risks defined by the Hydrographic and Oceanographic Service of the Navy (SHOA); calculate coastal erosion rates; and analyze extreme wave events in the study area. The results show that urban expansion has increased in areas affected by the 2010 tsunami, such as Iloca (36.88%), La Pesca (33.34%), and Pichibudi (20.78%). A 39-year reconstruction of the shoreline position (1985–2024) was carried out, serving as a basis to determine erosion rates and quantify shoreline dynamics. This dynamic is influenced by local water surface elevation, due to a combination of local tide variations, coastal storms, and wave-induced configurations. Erosion rates in Iloca and Pichibudi showed significant differences after the event, with rates varying from +0.94 m/year to -1.39 m/year in Iloca, +0.39 m/year to +0.47 m/year (erosion located in the distal zone) in La Pesca, and +0.42 m/year to -3.48 m/year in Pichibudi. These results highlight the lack of regulations and urban planning in areas exposed to extreme events, showing neglect in urban development and the absence of governmental action to build resilient coastal communities in the face of climate change.

Hyperspectral (HS) sensors are widely used for geological surveys and mineral classification. In areas with minimal vegetation and exposed minerals, it is ideal for mineral maps to remain consistent regardless of imaging time or sensor used. However, practical mineral maps often vary due to factors like atmospheric correction, sensor calibration, mixed pixels, and noise. This study introduces an approach combining the classification methodology of the knowledge-based expert system USGS Tetracorder with a data-driven approach, aiming to minimize discrepancies between mineral maps derived from independent datasets. To reduce the influence of noise and mixed pixels in HS data, Minimum Noise Fraction (MNF) transformations were applied, followed by a Pixel Purity Index (PPI) analysis. High-entropy pixels identified through PPI analysis were found to exhibit distinctive mineral-specific features compared to other pixels. These high-purity pixels were processed using the conventional Tetracorder method, serving as ground truth. For pixels deemed pure by PPI analysis, Tetracorder-derived mineral labels were used as the target variable, while the MNF-transformed bands were employed as feature variables. A Random Forest classifier was trained on these data and subsequently applied to map minerals in the remaining pixels. This method utilized the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) as the HS sensor and was validated over overlapping regions in the Cuprite area of Nevada, USA, using datasets captured at different times during the years 2011 to 2013. The results indicate that the proposed method provides a comparable level of accuracy to the standard Tetracorder implementation while significantly improving robustness and reducing errors compared to conventional processing. Future work will focus on validating this method in other regions and with different sensors to evaluate its generalizability.

Debris flows and debris floods are the most significant geological and geomorphological hazards in the foothills and mountain areas of Santiago. In recent decades, the most devastating events have been triggered by intense precipitation for short periods and thermal anomalies, affecting the population and causing fatalities and significant damage to infrastructure and natural ecosystems. Based on satellite and cartographic data, the analysis of conditioning factors and the use of Geographic Information Systems (GISs), the susceptibility to debris flows and debris floods in the Arrayán and Gualtatas creek basins (Santiago, Metropolitan Region of Chile) was determined. The susceptibility index (SI) was calculated using qualitative and quantitative methodologies, based on the sum of the weighted scores of 14 conditioning factors grouped into three categories (geology, geomorphology, and soil conditions). The level of importance of each factor was determined using the Analytical Hierarchical Process (AHP). The debris flow and debris flood generation susceptibility index map obtained for the study area indicates that 60.78% of the area has low to very low susceptibility, 23.64% has moderate susceptibility, and 15.58% has high to very high susceptibility. The latter is located at the headwaters of the basins, valley bottoms, steep slopes, and the main watercourses and streams and has records of recent debris flow events in urban areas. Freely available satellite data and GIS tools contribute to land use planning to determine possible areas of debris flow and debris flood generation that may affect the population of the northeastern sector of Santiago.