Earth’s forests play an important role in the fight against climate change and are in turn negatively affected by it. Effective monitoring of different tree species is essential to understanding and improving the health and biodiversity of forests. In this work, we address the challenge of tree species identification by performing tree crown semantic segmentation using an aerial image dataset spanning over a year. We compare models trained on single images versus those trained on time series to assess the impact of tree phenology on segmentation performance. We also introduce a simple convolutional block for extracting spatio-temporal features from image time series, enabling the use of popular pretrained backbones and methods. We leverage the hierarchical structure of tree species taxonomy by incorporating a custom loss function that refines predictions at three levels: species, genus, and higher-level taxa. Our best model achieves a mean Intersection over Union (mIoU) of 55.97%, outperforming single-image approaches particularly for deciduous trees where phenological changes are most noticeable. Our findings highlight the benefit of exploiting the time series modality via our Processor module. Furthermore, leveraging taxonomic information through our hierarchical loss function often, and in key cases significantly, improves semantic segmentation performance.

This study explores the potential of applying machine learning (ML) methods to identify and predict areas at risk of food insufficiency using a parsimonious set of publicly available data sources. We combine household survey data that captures monthly reported food insufficiency with remotely sensed measures of factors influencing crop production and maize price observations at the census enumeration area (EA) in Malawi. We consider three machine-learning models of different levels of complexity suitable for tabular data (TabNet, random forests, and LASSO) and classical logistic regression and examine their performance against the historical occurrence of food insufficiency. We find that the models achieve similar accuracy levels with differential performance in terms of precision and recall. The Shapley additive explanation decomposition applied to the models reveals that price information is the leading contributor to model fits. A possible explanation for the accuracy of simple predictors is the high spatiotemporal path dependency in our dataset, as the same areas of the country are repeatedly affected by food crises. Recurrent events suggest that immediate and longer-term responses to food crises, rather than predicting them, may be the bigger challenge, particularly in low-resource settings. Nonetheless, ML methods could be useful in filling important data gaps in food crises prediction, if followed by measures to strengthen food systems affected by climate change. Hence, we discuss the tradeoffs in training these models and their use by policymakers and practitioners.

The addition and refreezing of liquid water to Greenland’s accumulation area are increasingly important processes for assessing the ice sheet’s present and future mass balance, but uncertain initial conditions, complex infiltration physics and limited field data pose challenges. Satellite-based L-band radiometry offers a promising new tool for observing liquid water in the firn layer, although further validation is needed. This paper compares time series of liquid water amount (LWA) from three percolation zone sites generated by a localized point-model, a regional climate model, in situ measurement, and L-band radiometric retrievals. LWA integrates the interplay of liquid water generation and refreezing, which often occur simultaneously and repeatedly within firn layers on diurnal, episodic, and seasonal scales offering insights into methods for measuring and modeling meltwater processes. The four LWA records showed average discrepancies of up to 62% nRMSE, reflecting shortcomings inherent to each method. Better agreement between series occurred after excluding the regional climate model record, lowering nRMSE to 8–13%. The agreement between L-band radiometry and other LWA records inspires confidence in this observational tool for understanding firn meltwater processes and serving as a validation target for simulations of water processes in Greenland’s melting firn layer.

In this paper, frontal variations and surface area changes for each of the years 2017–2023 are assessed for 277 Swedish glaciers, of which the majority is contained within the Randolph Glacier Inventory 7.0. Mapping of all Swedish glaciers became possible by combining Sentinel-2 imagery, semi-automated mapping procedures and the open-source Margin Change Quantification Tool (MaQiT). In addition, manual mapping was performed at a subset of 22 glaciers historically associated with the Swedish Front Variation Program. At four of those, mapping accuracy was assessed by contrasting Sentinel-2 mapped fronts to fronts mapped in situ using Global Navigation Satellite System (GNSS), a total station and an uncrewed aerial vehicle. Results show widespread retreat of all Swedish glaciers, with cumulative frontal variation amounting on average to −55.6 m during 2017–2023 or −9.3 m a−1. Swedish glaciers had a total area of ∼237 km2 in 2017 and of 210 km2 in 2023. The reduction by ∼27 km2 corresponds to a loss of 11% with respect to the areal extent in the year 2017 but varies across regions. It is also almost as large as the combined area loss of Swedish glaciers in the preceding 15 years (∼31 km2, 2002–2017).

Sea surface salinity and temperature are essential climate variables in monitoring and modeling ocean health. Multispectral ocean color satellites allow the estimation of these properties at a resolution of 10 to 300 m, which is required to correctly represent their spatial variability in coastal waters. This paper investigates the effect of pre-applying an unsupervised classification in the performance of both temperature and salinity inversion. Two methodologies were explored: clustering based solely on spectral radiances, and clustering applied directly to satellite images. The former improved model generalization by identifying similar water clusters across different locations, reducing location dependency. It also demonstrated results correlating cluster type with salinity and temperature distributions thereby enhancing regression model performance and improving a global ocean color sea surface temperature regression model RMSE error by 10%. The latter approach, applying clustering directly to satellite images, incorporated spatial information into the models and enabled the identification of front boundaries and gradient information, improving global sea surface temperature models RMSE by 20% and sea surface salinity models by 30%, compared to the initial ocean color model. Beyond improving algorithm performance, optical water classification can be used to monitor and interpret changes to water optics, including algal blooms, sediment disturbance or other climate change or antropogenic disturbances. For example, the clusters have been used to show the impact of a category 4 hurricane landfall on the Mississippi estuarine region.

Globally, glaciers are changing in response to climate warming, with those that terminate in water often undergoing the most rapid change. In Alaska and northwest Canada, proglacial lakes have grown in number and size but their influence on glacier mass loss is unclear. We characterized the rates of retreat and mass loss through frontal ablation of 55 lake-terminating glaciers (>14 000 km2) in the region using annual Landsat imagery from 1984 to 2021. We find a median retreat rate of 60 m a−1 (interquartile range = 35–89 m a−1) over 1984–2018 and a median loss of 0.04 Gt a−1 (0.01–0.15 Gt a−1) mass through frontal ablation over 2009–18. Summed over 2009–18, our study glaciers lost 6.1 Gt a−1 to frontal ablation. Analysis of bed profiles suggest that glaciers terminating in larger lakes and deeper water lose more mass to frontal ablation, and that the glaciers will remain lake-terminating for an average of 74 years (38–177 a). This work suggests that as more proglacial lakes form and as lakes become larger, enhanced frontal ablation could cause higher mass losses, which should be considered when projecting the future of lake-terminating glaciers.

We present a practically simple methodology for tracking glacier surge onset and evolution using interferometric synthetic aperture radar (InSAR) coherence. Detecting surges early and monitoring their build-up is interesting for a multitude of scientific and safety-related aspects. We show that InSAR coherence maps allow the detection of surge-related instability on Svalbard many years before being detectable by, for instance, feature tracking or crevasse detection. Furthermore, we present derived data for two types of surges; down- and up-glacier propagating, with interestingly consistent surge propagation and post-surge relaxation rates. The method works well on Svalbard glaciers, and the data and core principle suggest a global applicability.

Photovoltaic (PV) energy grows rapidly and is crucial for the decarbonization of electric systems. However, centralized registries recording the technical characteristics of rooftop PV systems are often missing, making it difficult to monitor this growth accurately. The lack of monitoring could threaten the integration of PV energy into the grid. To avoid this situation, remote sensing of rooftop PV systems using deep learning has emerged as a promising solution. However, existing techniques are not reliable enough to be used by public authorities or transmission system operators (TSOs) to construct up-to-date statistics on the rooftop PV fleet. The lack of reliability comes from deep learning models being sensitive to distribution shifts. This work comprehensively evaluates distribution shifts’ effects on the classification accuracy of deep learning models trained to detect rooftop PV panels on overhead imagery. We construct a benchmark to isolate the sources of distribution shifts and introduce a novel methodology that leverages explainable artificial intelligence (XAI) and decomposition of the input image and model’s decision regarding scales to understand how distribution shifts affect deep learning models. Finally, based on our analysis, we introduce a data augmentation technique designed to improve the robustness of deep learning classifiers under varying acquisition conditions. Our proposed approach outperforms competing methods and can close the gap with more demanding unsupervised domain adaptation methods. We discuss practical recommendations for mapping PV systems using overhead imagery and deep learning models.

The accumulation area ratio (AAR) of a glacier reflects its current state of equilibrium, or disequilibrium, with climate and its vulnerability to future climate change. Here, we present an inventory of glacier-specific annual accumulation areas and equilibrium line altitudes (ELAs) for over 3000 glaciers in Alaska and northwest Canada (88% of the regional glacier area) from 2018 to 2022 derived from Sentinel-2 imagery. We find that the 5 year average AAR of the entire study area is 0.41, with an inter-annual range of 0.25–0.49. More than 1000 glaciers, representing 8% of the investigated glacier area, were found to have effectively no accumulation area. Summer temperature and winter precipitation from ERA5-Land explained nearly 50% of the inter-annual ELA variability across the entire study region (${R}^2=0.47$). An analysis of future climate scenarios (SSP2-4.5) projects that ELAs will rise by ∼170 m on average by the end of the 21st century. Such changes would result in a loss of 25% of the modern accumulation area, leaving a total of 1900 glaciers (22% of the investigated area) with no accumulation area. These results highlight the current state of glacier disequilibrium with modern climate, as well as glacier vulnerability to projected future warming.

Water hyacinth is a highly invasive aquatic species in the southern United States that requires intensive management through frequent herbicide applications. Quantifying management success in large-scale operations is challenging with traditional survey methods that rely on boat-based teams and can be time-consuming and labor-intensive. In contrast, an unmanned aerial system (UAS) allows a single operator to survey a waterbody more efficiently and rapidly, enhancing both coverage and data collection. Therefore, the objective of this research was to develop remote sensing techniques to assess herbicide efficacy for water hyacinth control in an outdoor mesocosm study. Experiments were conducted in spring and summer 2023 to compare and correlate data from visual evaluations of herbicide efficacy against nine vegetation indices (VIs) derived from UAS-based red-green-blue imagery. Penoxsulam, carfentrazone, diquat, 2,4-D, florpyrauxifen-benzyl, and glyphosate were applied at two rates, and experimental units were evaluated for 6 wk. The carotenoid reflectance index (CRI) had the highest Spearman’s correlation coefficient with visually evaluated efficacy for 2,4-D, diquat, and florpyrauxifen benzyl (> −0.77). The visible atmospherically resistance index (VARI) had the highest correlation with carfentrazone and penoxsulam treatments (> −0.70), and the excess greenness minus redness index had the highest correlation for glyphosate treatments (> −0.83). CRI had the highest correlation coefficient with the most herbicide treatments, and it was the only VI tested that did not include the red band. These VIs were satisfactory predictors of mid-range visually evaluated herbicide efficacy values but were poorly correlated with extremely low and high values, corresponding to nontreated and necrotic plants. Future research should focus on applying findings to real-world (nonexperimental) field conditions and testing imagery with spectral bands beyond the visible range.

Recent observations have shown a fast decrease in thickness and area of Pyrenean glaciers in some cases leading to a stagnation of ice flow. However, their transition to a new paraglacial stage is not well understood. Through the combination of uncrewed aerial vehicles imagery, airborne LiDAR, ground-penetrating radar and ground temperature observations, we characterized the recent evolution of Infiernos Glacier. In 2021, this glacier had small sectors thicker than 25 m, but most of area did not exceed 10 m. The thickness losses from 2011 to 2023 reached 9 m in average, of which 5 m occurring during the period 2020–23. This trend demonstrates the significant ice melt under current climatic conditions. In the last years, the glacier has also shown a remarkable increase of debris cover extent. In these areas, the ice loss was reduced by half when compared to the thickness decrease in the entire glacier. Sub-freezing ground temperatures evidence the highly probable presence of permafrost or buried ice in the surroundings of the glacier. The clear signs of ice stagnation and the magnitude of area and thickness decrease support the main hypothesis of this work: After 2023, the Infiernos Glacier can no longer be considered a glacier and has become an ice patch.

The severe ice losses observed for European glaciers in recent years have increased the interest in monitoring short-term glacier changes. Here, we present a method for constraining modelled glacier mass balance at the sub-seasonal scale and apply it to ten selected glaciers in the Swiss Alps over the period 2015–23. The method relies on observations of the snow-covered area fraction (SCAF) retrieved from Sentinel-2 imagery and long-term mean glacier mass balances. The additional information provided by the SCAF observations is shown to improve winter mass balance estimates by 22% on average over the study sites and by up to 70% in individual cases. Our approach exhibits good performance, with a mean absolute deviation (MAD) to the observed seasonal mass balances of 0.28 m w.e. and an MAD to the observed SCAFs of 6%. The results highlight the importance of accurately constraining winter accumulation when aiming to reproduce the evolution of glacier mass balance over the melt season and to better separate accumulation and ablation components. Since our method relies on remotely sensed observations and avoids the need for in situ measurements, we conclude that it holds potential for regional-scale glacier monitoring.

Forests play a crucial role in the Earth’s system processes and provide a suite of social and economic ecosystem services, but are significantly impacted by human activities, leading to a pronounced disruption of the equilibrium within ecosystems. Advancing forest monitoring worldwide offers advantages in mitigating human impacts and enhancing our comprehension of forest composition, alongside the effects of climate change. While statistical modeling has traditionally found applications in forest biology, recent strides in machine learning and computer vision have reached important milestones using remote sensing data, such as tree species identification, tree crown segmentation, and forest biomass assessments. For this, the significance of open-access data remains essential in enhancing such data-driven algorithms and methodologies. Here, we provide a comprehensive and extensive overview of 86 open-access forest datasets across spatial scales, encompassing inventories, ground-based, aerial-based, satellite-based recordings, and country or world maps. These datasets are grouped in OpenForest, a dynamic catalog open to contributions that strives to reference all available open-access forest datasets. Moreover, in the context of these datasets, we aim to inspire research in machine learning applied to forest biology by establishing connections between contemporary topics, perspectives, and challenges inherent in both domains. We hope to encourage collaborations among scientists, fostering the sharing and exploration of diverse datasets through the application of machine learning methods for large-scale forest monitoring. OpenForest is available at the following url: https://github.com/RolnickLab/OpenForest.

The Eridu region in southern Mesopotamia was occupied from the sixth until the early first millennium BC, and its archaeological landscape remains well preserved. The present study has identified and mapped a vast, intensive, well-developed network of artificial irrigation canals in this region.

As mid-southern U.S. rice producers continue to adopt furrow-irrigated rice production practices, supplementary management efforts will be vital in combating Palmer amaranth due to the extended germination period provided by the lack of a continual flood. Previous research has revealed the ability of cover crops to suppress Palmer amaranth emergence in corn, cotton, and soybean production systems; however, research on cover crop weed control efficacy in rice production is scarce. Therefore, trials were initiated in Arkansas in 2022 and 2023 to evaluate the effect of cover crops across five site-years on rice emergence, groundcover, grain yield, and total Palmer amaranth emergence. The cover crops evaluated were cereal rye, winter wheat, Austrian winterpea, and hairy vetch. Cover crop biomass accumulation varied by site-year, ranging from 430 to 3,440 kg ha−1, with cereal rye generally being the most consistent producer of high-quantity biomass across site-years. Rice growth and development were generally unaffected by cover crop establishment; however, all cover crops reduced rice emergence by up to 30% in one site-year. Rice groundcover was reduced by 13% from cereal rye in one site-year 2 wk before heading but cover crops did not affect rough rice grain yield in any of the site-years. Palmer amaranth emergence was reduced by 19% and 35% with cereal rye relative to the absence of a cover crop when rice was planted in April in Marianna, and May in Fayetteville, respectively. In most trials, Palmer amaranth emergence was not reduced by a cereal cover crop. In most instances, legume cover crops resulted in less Palmer amaranth emergence than without a cover crop. Based on these results, legume cover crops appear to provide some suppression of Palmer amaranth emergence in furrow-irrigated rice while having a minimal effect on rice establishment and yield.

Passive microwave measurements of Arctic sea ice have been conducted over the last 50 years from space and during airborne, ship- and ground-based measurement campaigns. The different radiometric signatures of distinct surface types have led to satellite retrievals of, e.g., sea-ice concentration. In contrast, ground-based upward-viewing radiometers measure radiation emitted from the atmosphere and are used to retrieve atmospheric variables. Here, we present results from a ship-based radiometer setup with a mirror construction, which allows us to switch between atmospheric and surface measurements flexibly. This way, in summer 2022, surface observations in the Arctic marginal sea-ice zone could be performed from the research vessel Polarstern by two radiometers covering the frequency range from 22 to 243 GHz. At low frequencies, the brightness temperatures show clear signatures of different surface conditions. We estimate emissivities at 53∘ zenith angle from infrared-based skin temperatures. Predominantly vertically polarized 22–31 GHz emissivities are between 0.51 and 0.55 for open ocean and around 0.95 for sea ice. Predominantly horizontally polarized 243 GHz ocean emissivities are around 0.78 and ice surfaces exhibit a large variability from 0.67 to 0.82. Our results can improve the characterization of surface emissions in satellite retrieval algorithms.

Extreme weather events caused by climate change, such as drought and heavy rainfall, will further increase in Central Europe in the near future. Resilient crop production requires in-depth knowledge of soil moisture (SM), its spatial and temporal variability and the dynamics of agriculturally used land. In the current study, different SM estimation methods, including measurement and simulation-based methods, were evaluated over a 17-ha experimental arable crop field with respect to their abilities to capture the spatial and temporal SM dynamics of within-field areas and their related uncertainty and spatial representativeness. The high-spatial resolution in-situ topsoil moisture measurements (50 m grid) were compared with the estimated SM from satellite-based remote sensing (S1ASCAT) and the simulated SM from three different crop water balance models (Agricultural Risk Information System [ARIS], AquaCrop and DSSAT). The evaluation revealed that the spatial variability in the experimental field obtained from the reference could not be captured by the alternative methods investigated because of the limitations of the grid size-related soil map information. Nevertheless, the analysis revealed a very good temporal correlation of SM dynamics with the field area average across all approaches, with AquaCrop and ARIS at a soil depth of 0–10 cm and S1ASCAT soil–water index 05 achieving a R2 and a Kling–Gupta efficiency >0.80. These results indicate the added value of complementary methods for estimating SM to reduce spatial and temporal uncertainties in the estimated topsoil water content.

Currently, methods for mapping agricultural crops have been predominantly developed for a number of the most important and popular crops. These methods are often based on remote sensing data, scarce information about the location and boundaries of fields of a particular crop, and involve analyzing phenological changes throughout the growing season by utilizing vegetation indices, e.g., the normalized difference vegetation index. However, this approach encounters challenges when attempting to distinguish fields with different crops growing in the same area or crops that share similar phenology. This complicates the reliable identification of the target crops based solely on vegetation index patterns. This research paper aims to investigate the potential of advanced techniques for crop mapping using satellite data and qualitative information. These advanced approaches involve interpreting features in satellite images in conjunction with cartographic, statistical, and climate data. The study focuses on data collection and mapping of three specific crops: lavender, almond, and barley, and relies on various sources of information for crop detection, including satellite image characteristics, regional statistical data detailing crop areas, and phenological information, such as flowering dates and the end of the growing season in specific regions. As an example, the study attempts to visually identify lavender fields in Bulgaria and almond orchards in the USA. We test several state-of-the-art methods for semantic segmentation (U-Net, UNet++, ResUnet). The best result was achieved by a ResUnet model (96.4%). Furthermore, the paper explores how vegetation indices can be leveraged to enhance the precision of crop identification, showcasing their advanced capabilities for this task.

Comprehensive housing stock information is crucial for informing the development of climate resilience strategies aiming to reduce the adverse impacts of extreme climate hazards in high-risk regions like the Caribbean. In this study, we propose an end-to-end workflow for rapidly generating critical baseline exposure data using very high-resolution drone imagery and deep learning techniques. Specifically, our work leverages the segment anything model (SAM) and convolutional neural networks (CNNs) to automate the generation of building footprints and roof classification maps. We evaluate the cross-country generalizability of the CNN models to determine how well models trained in one geographical context can be adapted to another. Finally, we discuss our initiatives for training and upskilling government staff, community mappers, and disaster responders in the use of geospatial technologies. Our work emphasizes the importance of local capacity building in the adoption of AI and Earth Observation for climate resilience in the Caribbean.