Land-use land-cover (LULC) change is one of the most important anthropogenic threats to biodiversity and ecosystems integrity. As a result, the systematic generation of annual regional, national, and global LULC map products derived from the classification of satellite imagery data have become critical inputs for multiple scientific disciplines. The importance of quantifying pixel-level uncertainty to improve the robustness of downstream analyses has long been acknowledged but this practice is still not widely adopted in the generation of these LULC products. The lack of uncertainty quantification is likely due to the fact that most approaches that have been put forward for this task are too computationally intensive for large-scale analysis (e.g., bootstrapping). In this article, we describe how conformal statistics can be used to quantify pixel-level uncertainty in a way that is not computationally intensive, is statistically rigorous despite relying on few assumptions, and can be used together with any classification algorithm that produces class probabilities. Our simulation results show how the size of the predictive sets created by conformal statistics can be used as an indicator of classification uncertainty at the pixel level. Our analysis based on data from the Brazilian Amazon reveals that both forest and water have high certainty whereas pasture and the “natural (other)” category have substantial uncertainty. This information can guide additional ground-truth data collection and the resulting raster combining the LULC classification with the uncertainty results can be used to communicate in a transparent way to downstream users which classified pixels have high or low uncertainty. Given the importance of systematic LULC maps and uncertainty quantification, we believe that this approach will find wide use in the remote sensing community.

Oil palm is the most efficient oil-producing crop but its extension leads to increased deforestation in Southeast Asia. Oil palm height enables the quantitative estimation for carbon stock or palm oil yield. Nevertheless, there are still no accurate characterization of oil palm height providing information for the tradeoff between forest damage and carbon stock of oil palm in Southeast Asia. The new generation of spaceborne LiDAR provides large-extent canopy height samples, offering an opportunity for mapping the oil palm height at the regional scale yet with challenge to extrapolate the footprint heights to spatially coherent maps. Here, we proposed a new method by combining Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) footprint data with stand age that is closely related to tree height. We first developed a semi-automatic filtering algorithm to filter the low-quality ICESat-2 data, and used a change detection algorithm to optimize the planting year map of oil palm. Then, an empirical age-height model was derived by linking ICESat-2 footprint canopy height with oil palm age that was used to estimate spatially and temporally oil palm height for the whole Peninsular Malaysia. A validation with independent ICESat-2 footprint data suggests a high agreement for the height estimates in 2020 from the age-height model (R2 = 0.63; RMSE = 1.64 m) with a bias within ±3 m for >90% of the height estimates. Using the age-height model and planting year map of oil palm, we produced the first comprehensive wall-to-wall maps of long-term yearly oil palm height at a spatial resolution of 30 m in Peninsular Malaysia during 2001 through 2020. Our results suggest that the mean height of oil palm in all and regionally-disturbed areas have increased by 10.82 m and 9.29 m respectively during the last two decades in Peninsular Malaysia. Our results indicate that combining stand age and ICESat-2 footprint data has great potential in spatially-explicit mapping regional oil palm height that contributes to a better quantification for regional plantation carbon stock.

The rapid growth of the city of Sydney, Australia over the last decades, has led to significant development of residential and transportation infrastructure. Land subsidence associated with the urban development can lead to serious issues which should be thoroughly understood and carefully managed. To address this challenge, an enhanced polarisation time-series InSAR (Pol-TS-InSAR) processing framework was developed, using the dual polarisation (DP) Sentinel-1 data to integrate information from different polarimetric channels with different weighting during the TS-InSAR deformation analysis. Ninety DP Sentinel-1 images acquired between 2019 and 2022 are analysed using Pol-TS-InSAR to map the land subsidence in Sydney, with the assistance of the GPS measurements. Improvement of measurement points density from Pol-TS-InSAR is observed compared to the single polarimetric TS-InSAR counterpart for all land use types (ranging between 68% and 208%). The comparison between the Pol-TS-InSAR measurements and GPS measurements shows an absolute mean difference and RMS difference of 0.75 mm/yr and 0.95 mm/yr, respectively, in vertical direction. The results of the ground subsidence analysis revealed that the main subsidence factors in Sydney are related to groundwater extraction, mining activities, underground tunnel construction and landfill. The latter two factors were less well-known prior to this study. In additional to these factors, land subsidence related to high-rise building construction has also been observed, even though the impact seems to be less significant than other factors.

The water table and its dynamics are one of the key variables that control peatland greenhouse gas exchange. Here, we tested the applicability of the Optical TRApezoid Model (OPTRAM) to monitor the temporal fluctuations in water table over intact, restored (previously forestry-drained), and drained (under agriculture) northern peatlands in Finland, Estonia, Sweden, Canada, and the USA. More specifically, we studied the potential and limitations of OPTRAM using water table data from 2018 through 2021, across 53 northern peatland sites, i.e., covering the largest geographical extent used in OPTRAM studies so far. For this, we calculated OPTRAM based on Sentinel-2 data with the Google Earth Engine cloud platform. First, we found that the choice of vegetation index utilised in OPTRAM does not significantly affect OPTRAM performance in peatlands. Second, we revealed that the tree cover density is a major factor controlling the sensitivity of OPTRAM to water table dynamics in peatlands. Tree cover density greater than 50% led to a clear decrease in OPTRAM performance. Finally, we demonstrated that the relationship between water table and OPTRAM often disappears when WT deepens (ranging between 0 to −100 cm, depending on the site location). We identified that the water table where OPTRAM ceases to be sensitive to variations is highly site-specific. Overall, our results support the application of OPTRAM to monitor water table dynamics in intact and restored northern peatlands with low tree cover density (below 50%) when the water table varies from shallow to moderately deep. Our study makes significant steps towards the broader implementation of optical remote sensing data for monitoring peatlands subsurface moisture conditions over the northern region.

Thermal discharge from nuclear power plants poses a threat to the received natural water bodies, but the long-term extent and intensity of their surface thermal plumes remain unclear. In this study, we proposed a method to determine the background area for each drainage outlet and delineate the mixed surface thermal plumes based on 7,172 Landsat thermal infrared images. We further used a deep convolutional neural network integrated with prior location knowledge to extract core surface thermal plumes for 74 drainage outlets of 66 nuclear power plants worldwide. Our final model achieved a mean Intersection over Union (mIoU) of 0.8998 and an F1 score of 0.8886. We found that the mean maximal water surface temperature (WST) increment of the studied plants globally was 4.80 K. The Tianwan plant in China experienced the highest WST increase (8.51 K), followed by the Gravelines plant in France and the Ohi plant in Japan (7.91 K and 7.71 K, respectively). The Bruce plant in Canada had the largest thermal-polluted surface area (7.22 km2). We also provided the dataset, Global Coastal Nuclear power plant Thermal Plume (GCNT-Plume), to describe the long-term occurrence of water surface thermal plumes. Three influencing factors of the water surface thermal plume were further analyzed in this study, including total capacity, drainage type, and location type, which were associated with operating power, drainage method, and geographical features, respectively. Total capacity was more statistically related to the maximum of WST increment under shallow drainage condition. The mean WST increment of shallow drainage was 1.22 K higher than that of deep drainage. Surface plumes larger than 4 km2 frequently occurred in the Great Lakes, while small surface thermal plumes (< 1 km2) were primarily found in estuaries. The proposed method provides an important framework for future operational water surface thermal plume detection using remotely sensed observations and deep learning.

Crop phenology has been widely detected from multiple historical satellite observations. Conversely, Near-Real-Time (NRT) monitoring of crop progress from timely available remote sensing data is barely investigated because of the lack of high-frequency cloud-free satellite observations and future potential crop development. To address the challenge, this study proposes a novel algorithm for operational NRT monitoring of crop progress at the field scale. This algorithm first fuses the high spatial resolution (30 m) Harmonized Landsat and Sentinel-2 (HLS) data and the high temporal frequent (10 min) Advanced Baseline Imager (ABI) observations to generate cloud-free time series of HLS-ABI EVI2 (two-band Enhanced Vegetation Index) with a Spatiotemporal Shape-Matching Model (SSMM). It then predicts future potential EVI2 values at a given pixel using a reference EVI2 time series obtained from the neighboring pixels in the preceding year. Integrating the currently available HLS-ABI observations and the predicted future EVI2 values to generate annual EVI2 time series, the algorithm finally detects six crop phenometrics including greenup onset, mid-greenup phase, maturity onset, senescence onset, mid-senescence phase, and dormancy onset. The NRT monitoring, which are separated as near-real-time prediction (phenological event detected after the occurrence), real-time prediction (phenological event detected around the occurrence), and short-term prediction (phenological event detected before the occurrence), are continuously updated and improved with new HLS and ABI observations at a weekly basis throughout the growing season. We evaluate the NRT monitoring against standard phenology products, PhenoCam observations, as well as the weekly Crop Progress Reports (CPRs) released from the National Agricultural Statistics Service (NASS) of the United States Department of Agriculture (USDA) in 2020 across Iowa. The evaluation demonstrates the robustness of the developed algorithm in NRT monitoring of crop phenology. Although the uncertainties are relatively large for short-term prediction compared with standard detections, the real-time prediction shows that the Mean Absolute Difference (MAD) is  0.85, P < 0.001) for various phenological stages of corn and soybean. These results prove that the algorithm could be implemented for NRT monitoring of various crop phenometrics from field, state, to national scales.

Poverty is the leading cause of social instability around the world. Reducing poverty has become a crucial objective for global sustainable development. Timely and consistently monitoring poverty status plays a vital role in evaluating the efficiency of poverty reduction policies. Nighttime light images from satellites allow for frequent monitoring of poverty status by recording lighting intensity. The coarse spatial resolution of the recent, widely used, publicly available nighttime light images hinders research on smaller unit scales, such as county-level. The SDGSAT-1 satellite, with a glimmer sensor and a spatial resolution of 10-m, was launched in 2021 to meet the sustainable development goals. This study proposes a model to sharpen the existing nighttime light images with a coarse spatial resolution to expand the available period of nighttime light images with a 10-m spatial resolution in discriminating illuminating ground objects. The expanded nighttime light images with discriminated illuminating ground objects are then utilized to calculate the time-series county-level poverty index for three typical Chinese urban agglomerations with a statistically significant correlation coefficient of 0.675 (p 90% of the counties experienced a reduction in poverty in Beijing–Tianjin–Hebei region due to time-series economic development. Statistical analysis of the poverty index of impoverished counties demonstrates a remarkable reduction of poverty in the study areas and the significance of the persistent poverty reduction policy implementation. This study's proposed pipeline compensates for the lack of nighttime light images with a higher spatial resolution prior to 2021 and provides a reliable poverty analysis foundation for sustainable development and policymaking.

Satellite altimetry is currently one of the most widely used techniques for monitoring lake water levels. Raw altimetry measurements are generally delivered in an ellipsoid height system, while lake heights are finally required in a local level height system to ensure the lake surface being strictly flat under the physical constraints of both gravity and non-gravity factors. Therefore, there is always a need for height system transformation of altimetry data. Many of the previous studies have used orthometric heights in an approximate manner in the altimetric data processing. The geoid errors, together with other error sources, can introduce significant biases into water level estimations. In this paper, we propose a spatio-temporal regression method to calculate orthometric height to level height (O2L) differences, which can be used to yield accurate lake levels from satellite altimetry data. In the experiments conducted in this study, we chose ICESat-2 data to train the O2L bias correction model, and evaluated the algorithm performance on two large lakes in central Asian (Issyk-Kul and Baikal lakes) and one lake in North America (Lake Michigan). The experimental results indicate that the spatio-temporal modeling method performs better than the traditional spatial modeling method in terms of regression accuracy, and the water levels derived from the new method are much closer to in-situ gauging measurements. Influenced by O2L bias, there is often a jitter phenomenon in the time series of lake levels, and the results of different satellites are, accordingly, not mutually consistent. A visual analysis reveals that the spatio-temporal modeling method can greatly reduce the high-frequency bias.

Three volcanic plumes were detected during the Tajogaite volcano eruptive activity (Canary Islands, Spain, September–December 2021) over the Iberian Peninsula. The spatiotemporal evolution of these events is characterised by combining passive satellite remote sensing and ground-based lidar and sun-photometer systems. The inversion algorithm GRASP is used with a suite of ground-based remote sensing instruments such as lidar/ceilometer and sun-photometer from eight sites at different locations throughout the Iberian Peninsula. Satellite observations showed that the volcanic ash plumes remained nearby the Canary Islands covering a mean area of 120 ± 202 km2 during the whole period of eruptive activity and that sulphur dioxide plumes reached the Iberian Peninsula. Remote sensing observations showed that the three events were mainly composed of sulphates, which were transported from the volcano into the free troposphere. The high backscatter-related Ångström exponents for wavelengths 532–1064 nm (1.17 ± 0.20 to 1.40 ± 0.24) and low particle depolarization ratios (0.08 ± 0.02 to 0.09 ± 0.02), measured by the multi-wavelength Raman lidar, hinted at the presence of spherical small particles. The layer aerosol optical depth at 532 nm (AODL532) obtained from lidar measurements contributed between 49% and 82% to the AERONET total column AOD at 532 nm in event II (11–13 October). According to the GRASP retrievals, the layer aerosol optical depth at 440 nm (AODL440) was higher in all sites during event II with values between 0.097 (Badajoz) and 0.233 (Guadiana-UGR) and lower in event III (19–21 October) varying between 0.003 (Granada) and 0.026 (Évora). Compared with the GRASP retrievals of total column AOD at 440 nm, the AODL440 had contributions between 21% and 52% during event II. In the event I (25–28 September), the mean volume concentrations (VC) varied between 5 ± 4 μm3cm−3 (El-Arenosillo/Huelva) and 17 ± 10 μm3cm−3 (Guadiana-UGR), while in event II this variation was from 11 ± 7 μm3cm−3 (Badajoz) to 27 ± 10 μm3cm−3 (Guadiana-UGR). Due to the impact of volcanic events on atmospheric and economic fields, such as radiative forcing and airspace security, a proper characterization is required. This work undertakes it using advanced instrumentation and methods.

Contemporary research on Greenland surface mass balance (SMB) is largely focused on the characteristics of decadal or longer trends, periodic oscillations, and acceleration. However, the specific components of the SMB such as snowfall (SF), rainfall (RF) and runoff (RU), and their corresponding temporal and spatial variability remain poorly understood. Here, we explore the respective contributions of SF, RF, and RU to the seasonal and transient crustal deformations of Greenland during the past two decades using GPS network and satellite gravimetry (GRACE) datasets, and regional climate model output. Our study unraveled that the largest annual vertical displacement caused by precipitations is in southeastern Greenland, reaching 7.27 mm. The largest surface displacement caused by RU is in western Greenland, reaching 19.82 mm. Ice mass gain/loss in Greenland shows a clear correlation between latitude and temperature, with greater variations in the south compared to the north. The transient deformation signals in Greenland mainly manifested in terms of abrupt subsidence in 2010, followed by uplift in 2014. The 2014 uplift can mainly be attributable to the combined effect of SF, RF, and RU. The largest transient signal occurs in the southeast subregions, with peak-to-peak amplitude exceeding 10 mm. Transient crustal deformation is mainly caused by precipitation in southeastern Greenland, while the contribution of RU dominates most of the time and in most subregions. We find that even though RF is increasing due to an increasingly warmer climate, its effect on SMB is still negligible, when compared with SF and RU. In some subregions and some periods, SF could become the primary contributor to transient SMB variations in Greenland.

Providing forest biomass estimates with desired accuracy and precision for small areas is a key challenge to incorporating forest carbon offsets into commodity trading programs. Enrolled forest carbon projects and verification entities typically rely on probabilistically sampled field data and design-based (DB) estimators to estimate carbon storage and characterize uncertainty. However, this methodology requires a large amount of field data to achieve sufficient precision and collection of these data can be prohibitively expensive. This has spurred interest in developing regional-scale maps of forest biomass that incorporate remote sensing data as an alternative to collecting expensive plot data. These maps are often generated using machine learning (ML) algorithms that combine remote sensing products and field measurements. While these maps can produce estimates across large geographic regions at fine spatial resolutions, the estimates are prone to bias and do not have associated uncertainty estimates. Here, we assess one such map developed by the National Aeronautics and Space Administration's Carbon Monitoring System. We consider model-assisted (MA) and geostatistical model-based (GMB) estimators to address map bias and uncertainty quantification. The MA and GMB estimators use a sample of field observations as the response, and the ML-produced map as an auxiliary variable to achieve statistically defensible predictions. We compare MA and GMB estimator performance to DB and direct (DR) estimators. This assessment considers both counties and a small areal extent experimental forest, all within Oregon USA. Results suggest the MA and GMB estimators perform similar to the DB estimator at the state level and in counties containing many field plots. But in counties with moderate to small field sample sizes, the GMB and MA estimators are more precise than the DB estimator. As within-county sample sizes get smaller, the GMB estimator tends to outperform MA. Results also show the DR estimator's state-level estimates are substantially larger than the DB, MA and GMB estimates, indicating that that the DR estimator may be biased. When assessing the GMB estimator for the experimental forest, we find the GMB estimator has sufficient precision for stand-level carbon accounting even when no field observations are available within the stand. Plot-level GMB uncertainty interval coverage probabilities were estimated and showed adequate coverage. This suggests that the GMB estimator is producing statistically rigorous uncertainty estimates.

Surface fuel information is an essential input for models of fire behaviour and fire effects. However, spatially explicit, continuous information on surface fuel loads and fuelbed depth is scarce because the collection of field data is laborious, while suitable methods for deriving estimates from remote sensing data are still at an early stage of development. Fine-scale surface fuel mapping using both passive and active remote sensing has not yet been carried out in Central European forest types, and it remains unexplored how prediction uncertainties of different fuel components affect modelled fire behaviour. This study combines very detailed airborne lidar and multispectral satellite data to extract metrics describing forest structure and composition in two forested areas in southwestern Germany. These metrics were used to predict field-sampled surface fuel components using random forest regression. Accuracies of continuous fuel load predictions were compared to accuracies that could be achieved if only forest type-specific average fuels were assigned. Results revealed that models based on remotely sensed metrics explain part of the variance in litter and fine dead woody fuels (R2=0.27-0.41), but not in coarser dead woody fuels. Estimates for herb and shrub fuels were fairly accurate (R2=0.55-0.64) but limited for the more fire-relevant fine fraction of shrub fuels (R2=0.39). Fuelbed depth was moderately well predicted based on remote sensing data (R2=0.44). Lidar-derived metrics were particularly useful for predicting understory fuels and fuelbed depth. Litter and fine woody fuel predictions were linked to canopy characteristics captured with both lidar and multispectral data and similarly accurate estimates could be obtained using average values based on forest type. We used the fine-scale surface fuel maps derived from remote sensing to predict potential surface fire behaviour in the study area and analysed the sensitivity of modelled fire behaviour to errors in the predicted loads of different surface fuel components: fire behaviour was most sensitive to errors in litter and especially shrub fuel loads, hence estimates of these components need to be improved. Overall, this study showed that statistical relationships between remotely sensed metrics describing forest composition and structure and surface fuels have some potential for estimating fuel loads in Central European forest types and should be further developed to provide starting points for realistic fire behaviour models.

As a critical source of food and one of the most economically significant crops in the world, soybean plays an important role in achieving food security. Large area accurate mapping of soybean has long been a vital, but challenging issue in remote sensing, relying heavily on large-volume and representative training samples, whose collection is time-consuming and inefficient, especially for large areas (e.g., national scale). Thus, methods are needed that can map soybean automatically and accurately from single-date remotely sensed imagery. In this research, a novel Greenness and Water Content Composite Index (GWCCI) was proposed to map soybean from just a single Sentinel-2 multispectral image in an end-to-end manner without employing training samples. By capitalizing on the product of the NDVI (related to greenness) and the short-wave infrared (SWIR) band (related to canopy water content), the GWCCI provides the required information with which to discriminate between soybean and other land cover types. The effectiveness of the proposed GWCCI was investigated in seven typical soybean planting regions within four major soybean-producing countries across the world (i.e., China, the United States, Brazil and Argentina), with diverse climates, cropping systems and agricultural landscapes. In the experiments, an optimal threshold of 0.17 was estimated and adopted by the GWCCI in the first study site (S1) in 2021, and then generalised to the other study sites over multiple years for soybean mapping. The GWCCI method achieved a consistently higher accuracy in 2021 compared to two conventional comparative classifiers (support vector machine (SVM) and random forest (RF)), with an average overall accuracy (OA) of 88.30% and a Kappa coefficient (k) of 0.77; significantly greater than those of RF (OA: 80.92%, k: 0.62) and SVM (OA: 80.29%, k: 0.60). Furthermore, the OA of the extended years was highly consistent with that of 2021 for study sites S2 to S7, demonstrating the great generalisation capability and robustness of the proposed approach over multiple years. The proposed GWCCI method is straightforward, reliable and robust, and represents an important step forward for mapping soybean, one of the most significant crops grown globally.

There is a lack of understanding of the complex spatiotemporal patterns of heat exposure during heat waves, and the impact of urbanization intensity and urban morphology on heat exposure in urban thermal environments. To address these issues, this study used mean radiant temperature (Tmrt) as an index to indicate human exposure to extreme heat, and generated hourly heat exposure maps at a 1-m spatial resolution in Summer 2020 for heat wave and non-heat wave days across three diversely urbanized and climatically different U.S. cities (Riverside, CA; Des Moines, IA; and Boston, MA) using the SOlar LongWave Environmental Irradiance Geometry (SOLWEIG) model and multi-source remote sensing and GIS data. Based on these high-frequency and microscale maps, we found that heat exposure in urban canyons of downtown areas was high due to relatively low building's height to street's width (H/W) ratio, which resulted in a limited shading effect in the studied cities. Heat exposure during heat waves was enhanced by 6 °C to 10 °C compared to non-heat wave conditions, with the main differences occurring in the early afternoon between 12 pm and 2 pm. We found that hot cities (Riverside, 63 °C) had higher heat exposure than warm cities (Des Moines and Boston, 53 °C and 51 °C) during heat waves. Heat exposure in urban core areas was approximately 5C higher than that in rural areas during heat waves. Additionally, we found that sky view factor was the most important urban morphology factor influencing heat exposure, with a relative importance of over 67% in these cities, but the role of impervious surface and trees varied among these cities. Impervious surface area (ISA) contributed more to heat exposure than trees in dry and hot regions (Riverside), but not in humid and warm cities (Des Moines and Boston). This study is the first to generate hourly heat exposure maps at a 1-m resolution for heat wave and non-heat wave days, and to investigate spatiotemporal patterns and the impacts of urbanization intensity and urban morphology on heat exposure in multiple cities. The findings of this study can be useful in developing urban policies to improve urban thermal environments in diverse urban settings, and our transferable framework can potentially be applied to other cities for heat exposure studies.

Vegetation indices (VIs) have been used extensively for qualitative and quantitative remote sensing monitoring of vegetation vigor and growth dynamics. However, the saturation phenomenon of VIs (i.e., insignificant change at moderate to high vegetation densities) poses a known limitation to their ability to characterize surface vegetation over the dense canopy. Although the mechanisms underlying saturation are relatively straightforward and several VIs have been proposed to mitigate the saturation effect, the assessment of the saturation effect of VIs remains insufficient. Notably, no unified metric has been proposed to quantify the VI saturation phenomenon, limiting VI selection in practical applications. In this study, we proposed two indicators to describe the saturation phenomenon and utilized a well-validated three-dimensional (3D) canopy radiative transfer (RT) model large-scale remote sensing data and image simulation framework (LESS) to simulate the bidirectional reflectance factor (BRF) of six forests scenes and assessed the variations in VIs in relation to leaf area index (LAI) values over different backgrounds, sun-sensor geometries, and spatial distribution types. The saturation characteristics of 36 VIs were evaluated in combination with simulation results and satellite observations from multiple sensors. The ranking of VI saturation from simulated and satellite results revealed a good agreement. Our results indicated that the simple ratio vegetation index (SR) performed best with the highest saturation point and can well characterize the surface vegetation condition until LAI reaches 4. Besides, we found that the saturation effect of VIs was influenced by soil brightness, sun-sensor geometry, and canopy structure. SR, modified simple ratio (MSR) and normalized green red difference index (NGRDI) were the most susceptible to these disturbing factors, although they had higher resistance to saturation. Modified triangular vegetation index 1 (MTVI1), modified non-linear vegetation index (MNLI), triangular greenness index (TGI), and triangular vegetation index (TriVI) performed well overall, combining the ability to resist saturation and disturbance factors. Appropriate application of VIs can help better understand vegetation responses to climate change and accurately assess ecosystem status. Our results contribute to the understanding of the VI saturation effect and provide a combined model and satellite data experimental workflow in appropriate VI selection to accurately characterize vegetation.

Abstract not available

Monitoring Antarctic icebergs helps us understand the interaction between ocean, atmosphere, and sea ice in the Southern Ocean. Although gigantic icebergs have been the objects of many remote sensing studies, medium icebergs in the Southern Ocean have been rarely monitored or traced. In this study, we develop an iceberg detection and tracking tool particularly for medium and large icebergs (0.4–10 km2), by using Python programming language and Sentinel-1 (S1) imagery, based on Google Earth Engine (GEE). To detect icebergs, we employ the simple non-iterative clustering (SNIC) and region adjacency graph (RAG) merging for object-based image segmentation and train/test the support vector machine (SVM) model with 6432 labeled segments of iceberg or non-icebergs from 40 images of S1 (2019–2021). Radar backscatter features and morphological features of those segments are used as the inputs of the SVM model. After icebergs are detected in two image scenes of different dates, we track the displacements of detected icebergs by comparing their 1-D shape signals. Our model shows ∼99% of accuracy in detecting icebergs and ∼ 93–98% of accuracy in tracking icebergs depending on the day difference between image scenes. When using our tool for the monitoring of icebergs in the Amundsen Sea, we find that the iceberg fraction varies from 2% to 8% in 2021 and most of icebergs move westward with a speed of <0.2 km/day.

Recent studies have highlighted that meltwater in supraglacial lakes (SLs) can be buried during frozen season in the Greenland Ice Sheet (GrIS). Meltwater in buried lakes (BLs) can even persist through the winter, disturbing the englacial thermal regime and providing an important buffer against GrIS's contribution to sea-level rise. However, little is known about the inter-annual BL dynamics in the GrIS, and there is no quantitative statistic about the overall buried percentage. Here, we conduct a satellite-based study to automatically map the winter BLs over the GrIS during 2017–2022 using multi-source optical and synthetic aperture radar (SAR) images on the Google Earth Engine (GEE) platform. To eliminate the interferences from other weak microwave reflecting surfaces, summer SLs are first extracted from Landsat 8 and Sentinel-2 images to determine the potential BL searching areas on winter Sentinel-1 images. A self-adaptive thresholding algorithm is proposed to extract BLs within the dilated summer SLs using histogram-based morphological edge detectors. BLs extracted by the proposed method and visual interpretation show a substantial agreement with a precision of 0.82 and a Kappa coefficient of 0.70. On average, a total buried lake area of 182.27 km2 was observed each winter during the period 2017–2022. BLs were mainly distributed in the Center-West, South-West and North-East Basins, with the majority occurring at elevations between 800 and 1700 m. In 2019–2020, a sudden extension of BLs was observed over the GrIS, especially in the North-East Basin where abnormally high temperatures and surface runoff were recorded. In 2021–2022, a widespread distribution of BLs in the South-West Basin was observed after abnormal snowfall. Overall, about 13% of the GrIS summer SLs can persist through winter, suggesting the potential for meltwater hydrofracture in winter over large areas.

Cashews are grown by over 3 million smallholder farmers in >40 countries worldwide as a principal source of income. Expanding the area of cashew plantations and increasing productivity are critical to improving the livelihood of many smallholder communities. As the third largest cashew producer in Africa, Benin has nearly 200,000 smallholder cashew growers contributing 15% of the country's national export earnings. Expansion of the cashew industry is thus an essential economic driver and a governmental priority in Benin. However, a lack of information on where and how cashew trees grow across the country hinders decision-making that could support increased cashew production and poverty alleviation. By leveraging 2.4-m Planet Basemaps and 0.5-m aerial imagery, two newly developed deep learning algorithms, and large-scale ground truth datasets, we successfully produced the first-of-its-kind national map of cashew in Benin and characterized the expansion of cashew plantations between 2015 and 2021. In particular, we developed a SpatioTemporal Classification with Attention (STCA) model to map the distribution of cashew plantations with 2.4-m multi-temporal Planet Basemaps from 2019 to 2021, which can fully capture texture information from discriminative time steps during a growing season. The U-Net model was employed to map the distribution of cashew plantation with 0.5-m mono-temporal aerial imagery in 2015, which can achieve accurate and fast predictions even with limited training data. We further developed a Clustering Augmented Self-supervised Temporal Classification (CASTC) model to distinguish high-density versus low-density cashew plantations by automatic feature extraction and optimized clustering. Results show that the STCA model has an overall accuracy over 85% based on 1400 ground truth point samples from each year. The CASTC model achieved an overall accuracy of 76% based on 348 ground truth samples of planting density. We found that the cashew area in Benin has almost doubled to 519 ± 20 kha from 2015 to 2021 with 60% of new plantation development coming from cropland or fallow land, while encroachment of cashew plantations into protected areas has increased by 55%. Only about half of cashew plantations were high-density in 2021, suggesting high potential for intensification. Our study illustrates the power of combining high-resolution remote sensing imagery and state-of-the-art deep learning algorithms to better understand tree crops in the heterogeneous smallholder landscape, which can help efficiently allocate limited training and nursery resources for sustainable agricultural development.

Solar-induced chlorophyll fluorescence (SIF) is a rapidly advancing front in modeling global terrestrial gross primary production (GPP). Canopy total SIF emissions (SIFtotal) are mechanistically linked to the plant photosynthesis, and can be estimated from satellite observed SIF (SIFobs) through radiative transfer modeling. However, the current satellite SIFobs and thus SIFtotal are available only at coarse spatial resolutions from several kilometers to tens of kilometers, inhibiting the application at fine spatial scales. Here, we proposed an algorithm to generate both global high-resolution SIFtotal (HSIFtotal) and high-resolution SIFobs (HSIFobs) at 1 km from low-resolution SIFobs (LSIFobs) from the TROPOspheric Monitoring Instrument (TROPOMI), which has a spatial resolution at nadir of 3.5 km by 5.6–7 km. Our statistical method is based on the law of energy conservation and uses satellite derived fraction of absorbed photosynthetically active radiation, fluorescence efficiency, and the escape probability of fluorescence. We evaluated the accuracy of our HSIFtotal using the Orbiting Carbon Observatory-2 SIF (R2 = 0.78). We found that the spatial resolution had clear effects on the relationship between HSIFtotal and GPP. We also compared HSIFtotal to 8-day averaged tower GPP from 135 flux sites and found that they were better correlated when HSIFtotal was averaged over a 1-km radius around the tower than when averaged over a larger radius. Our study provided a unique high-resolution HSIFtotal product, which will advance the estimation of GPP by extrapolating site-level relationships to the global scale.