Mapping land cover in highly heterogeneous landscapes is challenging, and classifications have inherent limitations where the spatial resolution of remotely sensed data exceeds the size of small objects. For example, classifications based on 30-m Landsat data do not capture urban or other heterogeneous environments well. This limitation may be overcome by quantifying the subpixel fractions of different land cover types. However, the selection process and transferability of models designed for subpixel land cover mapping across biomes is yet challenging. We asked to what extent (a) locally trained models can be used for sub-pixel land cover fraction estimates in other biomes, and (b) training data from different regions can be combined into spatially generalized models to quantify fractions across global biomes. We applied machine learning regression-based fraction mapping to quantify land cover fractions of 18 regions in five biomes using Landsat data from 2022. We used spectral-temporal metrics to incorporate intra-annual temporal information and compared the performance of local, spatially transferred, and spatially generalized models. Local models performed best when applied to their respective sites (average mean absolute error, MAE, 9–18%), and also well when transferred to other sites within the same biome, but not consistently so for out-of-biome sites. However, spatially generalized models that combined input data from many sites worked very well when analyzing sites in many different biomes, and their MAE values were only slightly higher than those of the respective local models. A weighted training data selection approach, preferring training data with a lower spectral distance to the image data to be predicted, further enhanced the performance of generalized models. Our results suggest that spatially generalized regression-based fraction models can support multi-class sub-pixel fraction estimates based on medium resolution satellite images globally. Such products would have great value for environmental monitoring in heterogeneous environments and where land cover varies along spatial or temporal gradients.
File: Schug-1-s2.0-S0034425724002785-main.pdf
Remote sensing data are important for assessing ecological change, but their value is often restricted by their limited temporal coverage. Major historical events that affected the environment, such as those associated with colonial history, World War II, or the Green Revolution are not captured by modern remote sensing. In the present article, we highlight the potential of globally available black-and-white satellite photographs to expand ecological and conservation assessments back to the 1960s and to illuminate ecological concepts such as shifting baselines, time-lag responses, and legacy effects. This historical satellite photography can be used to monitor ecosystem extent and structure, species’ populations and habitats, and human pressures on the environment. Even though the data were declassified decades ago, their use in ecology and conservation remains limited. But recent advances in image processing and analysis can now unlock this research resource. We encourage the use of this opportunity to address important ecological and conservation questions.
File: biae002.pdf
More frequent and widespread large fires are occurring in the western United States (US), yet reliable methods for predicting these fires, particularly with extended lead times and a high spatial resolution, remain challenging. In this study, we proposed an interpretable and accurate hybrid machine learning (ML) model, that explicitly represented the controls of fuel flammability, fuel availability, and human suppression effects on fires. The model demonstrated notable accuracy with a F1‐score of 0.846 ± 0.012, surpassing processdriven fire danger indices and four commonly used ML models by up to 40% and 9%, respectively. More importantly, the ML model showed remarkably higher interpretability relative to other ML models. Specifically, by demystifying the “black box” of each ML model using the explainable AI techniques, we identified substantial structural differences across ML fire models, even among those with similar accuracy. The relationships between fires and their drivers, identified by our model, were aligned closer with established fire physical principles. The ML structural discrepancy led to diverse fire predictions and our model predictions exhibited greater consistency with actual fire occurrence. With the highly interpretable and accurate model, we revealed the strong compound effects from multiple climate variables related to evaporative demand, energy release component, temperature, and wind speed, on the dynamics of large fires and megafires in the western US. Our findings highlight the importance of assessing the structural integrity of models in addition to their accuracy. They also underscore the critical need to address the rise in compound climate extremes linked to large wildfires.
File: Earth-s-Future-2024-Li-Projecting-Large-Fires-in-the-Western-US-With-an-Interpretable-and-Accurate-Hybrid-Machine.pdf
The Atlantic meridional overturning circulation (AMOC) has caused significant climate changes over the past 90 000 years. Prior work has hypothesized that these millennial-scale climate variations effected past and contemporary biodiversity, but the effects are understudied. Moreover, few biogeographic models have accounted for uncertainties in palaeoclimatic simulations of millennial-scale variability. We examine whether refuges from millennial-scale climate oscillations have left detectable legacies in the patterns of contemporary species richness in eastern North America. We analyse 13 palaeoclimate estimates from climate simulations and proxy-based reconstructions as predictors for the contemporary richness of amphibians, passerine birds, mammals, reptiles and trees. Results suggest that past climate changes owing to AMOC variations have left weak but detectable imprints on the contemporary richness of mammals and trees. High temperature stability, precipitation increase, and an apparent climate fulcrum in the southeastern United States across millennial-scale climate oscillations aligns with high biodiversity in the region. These findings support the hypothesis that the southeastern United States may have acted as a biodiversity refuge. However, for some taxa, the strength and direction of palaeoclimate-richness relationships varies among different palaeoclimate estimates, pointing to the importance of palaeoclimatic ensembles and the need for caution when basing biogeographic interpretations on individual palaeoclimate simulations.
Climate warming enables easier access and operation in the Arctic, fostering industrial and urban development. However, there is no comprehensive pan-Arctic overview of industrial and urban development, which is crucial for the planning of sustainable development of the region. In this study, we utilize satellite-derived artificial light at night (ALAN) data to quantify the hotspots and the development of light-emitting human activity across the Arctic from 1992 to 2013. We find that out of 16.4 million km2 analyzed a total area of 839,710 km2 (5.14%) is lit by human activity with an annual increase of 4.8%. The European Arctic and the oil and gas extraction regions in Russia and Alaska are hotspots of ALAN with up to a third of the land area lit, while the Canadian Arctic remains dark to a large extent. On average, only 15% of lit area in the Arctic contains human settlement, indicating that artificial light is largely attributable to industrial human activity. With this study, we provide a standardized approach to spatially assess human industrial activity across the Arctic, independent from economic data. Our results provide a crucial baseline for sustainable development and conservation planning across the highly vulnerable Arctic region.
File: akandil-et-al-2024-artificial-light-at-night-reveals-hotspots-and-rapid-development-of-industrial-activity-in-the-arctic.pdf
Heterogeneous vegetation supports higher species richness than homogenous vegetation, which is why efficiently identifying heterogenous vegetation can be useful for biodiversity conservation. Satellite remote-sensing data provide an opportunity to generate vegetation heterogeneity metrics and to explore the phenology of vegetation patterns. Phenoclusters are vegetation types with similar phenological characteristics, and valuable for capturing vegetation habitat heterogeneity patterns. Our goal was to map phenoclusters for Wisconsin, USA, at 10-m spatial resolution based on land surface phenology metrics from EVI (Enhanced Vegetation Index) Sentinel-2 data. We characterized each phenocluster based on landcover composition and structure, phenology timing, and environmental factors, and compared them to bird species richness. We also calculated the diversity of phenoclusters at multiple spatial extents. We identified 14 phenoclusters in Wisconsin, each with distinct landcover composition and structure, and unique phenological characteristics. Our remotely-sensed phenoclusters effectively captured environmental gradients, with elevation and temperature emerging as the most important driving variables. Furthermore, the phenoclusters successfully captured bird biodiversity patterns, especially richness of forest and grassland specialist. Our results identified phenological patterns among Wisconsin’s forests, shrublands, and grasslands, capturing phenological timing both among and within the same tree species. Phenoclusters are a valuable tool for capturing vegetation habitat heterogeneity, phenology diversity and biodiversity patterns, as well as climate change effects.
File: Silveira-et-al_2024_WisconsinPhenoclusters.pdf
Snow cover dynamics (i.e. depth, duration and variability) are dominant drivers of ecological processes during winter. For overwintering species, changes and gradients in snow cover may impact survival and population dynamics (e.g. facilitating survival via thermal refugia or limiting survival via reduced resource acquisition). However, snow cover dynamics are rarely used in species distribution modelling, especially for over-wintering birds. Currently, we lack understanding of how snow cover gradients affect overwintering bird distributions and which functional traits drive these associations at regional and continental scales. Using observations from eBird, a global community science network, we explored the effects of snow cover dynamics on continental pat-terns of occurrence and counts for 150 bird species. We quantified the relative impor-tance, species-specific responses and trait-based relationships of bird occurrence and abundance patterns to ecologically relevant snow cover dynamics across the United States. Snow cover dynamics were important environmental predictors in species dis-tributions models, ranking within the top three predictors for most species occurrence (> 90%) and count (> 79%) patterns across the contiguous United States. Species exhibited a gradient of responses to snow cover from snow association to snow avoid-ance, yet most birds were limited by long, persistent snow seasons. Duration of winter and percent frozen ground without snow structured species distributions in the east-ern USA, whereas snow cover variability was a stronger driver in the western USA. Birds associated with long, persistent snow seasons had traits associated with greater dispersal capacity and dietary diversity, whereas birds inhabiting regions with variable snow cover were generally habitat generalists. Our results suggest that various snow cover dynamics are important ecological filters of species distributions during winter. Global climate change is rapidly degrading key characteristics of seasonal snow cover. A changing cryosphere may elicit variable distributional changes for many overwinter-ing birds, potentially accelerating range shifts and novel community assemblages.
File: Ecography-2022-Keyser-Snow-cover-dynamics-an-overlooked-yet-important-feature-of-winter-bird-occurrence-and.pdf
In this work, we propose a deep learning-based method to perform semiparametric regression analysis for spatially dependent data. To be specific, we use a sparsely connected deep neural network with rectified linear unit (ReLU) activation function to estimate the unknown regression function that describes the relationship between response and covariates in the presence of spatial dependence. Under some mild conditions, the estimator is proven to be consistent, and the rate of convergence is determined by three factors: (1) the architecture of neural network class, (2) the smoothness and (intrinsic) dimension of true mean function, and (3) the magnitude of spatial dependence. Our method can handle well large data set owing to the stochastic gradient descent optimization algorithm. Simulation studies on synthetic data are conducted to assess the finite sample performance, the results of which indicate that the proposed method is capable of picking up the intricate relationship between response and covariates. Finally, a real data analysis is provided to demonstrate the validity and effectiveness of the proposed method.
File: SpatialStatistics_VR_2023.pdf
Global changes in climate and land use are threatening natural ecosystems, biodiversity, and the ecosystem services people rely on. This is why it is necessary to track and monitor spatiotemporal change at a level of detail that can inform science, management, and policy development. The current constellation of multiple Landsat and Sentinel-2 satellites collecting imagery at predominantly ≤30-m spatial resolution affords an opportunity for the generation of global medium- resolution products every few days. Our goal is to both identify the information needs and provide direction towards the generation of a suite of global, high-level, systematically-generated, medium-resolution products designed for both management and science. Our vision builds on the success of the NASA MODIS/VIIRS product suite, while recognizing the unique strengths of medium-resolution satellite data given their higher spatial resolution and longer time series. We propose a suite of 13 essential products that enable the characterization of the current state and changes in the biosphere, cryosphere, and hydrosphere, and would fill information needs identified by the Committee on Earth Observation Satellites for the Global Climate Observing System and the Global Terrestrial Observing System, by the National Research Council of the US National Academies in the decadal survey, and by others. These products are: land cover, land cover change, burned area, forest loss, vegetation indices, phenology, dynamic habitat indices, albedo, land surface temperature, snow cover, ice extent, surface water extent, and evapotranspiration. Furthermore, we provide a list of desirable products poised for addition to the essential products (e.g., crop type, emissivity, and ice sheet velocity). Lastly, we suggest aspirational products requiring further algorithm development (e.g., forest structure and crop yield). For the identified essential products, algorithms are in place, making it feasible to begin generating products systematically. These products should be accompanied by quality and accuracy assessments undertaken following consensus protocols. Five decades after the first Landsat satellite, and two decades after the MODIS products were first produced, it is time now for readily available, standardized, and consistent high-level products built upon medium-resolution imagery, thereby fulfilling the promise and the vision that inspired the Landsat program since its inception.
File: Radeloff_RSE_2024.pdf
Socioeconomic shocks can cause regime shifts in land use, but even during shocks, and when land use change is widespread, some areas persist in their land use. The question is what makes these areas more resistant. Our research goal was to find out what explains where arable farming persisted despite a major socioeconomic shock of forced post-war displacements. Our study area were 291 villages in the Polish Carpathians where abandonment due to the forced displacement of the Ukrainian population after WWII was widespread. We compared prewar arable land with 1990 CORINE Land Cover data to quantify land-use change throughout the socialist period. We applied logistic regression with economically relevant environmental and access-related variables, and assessed the explanatory power of our models and relative importance of determinants. Forty years after forced displacements, arable farming persisted only in a small portion of what had been farmed in the 1940s (16 %), while the majority of former arable land converted to forests (54 %) or grasslands (22 %). Arable farming persisted mainly in areas with high accessibility that had oak-hornbeam forest as potential natural vegetation, on less steep slopes, and at lower elevations. Our models predicting agricultural abandonment leading to reforestation performed well (R2 = 0.57), but our model of persistent agriculture had low explanatory power (R2 = 0.26) as did models of conversion to grassland (R2 = 0.24). We therefore conclude that agricultural persistence is driven by different factors than agricultural land abandonment. In the long term, after arable farming ceases, areas can either be completely abandoned or convert to less intensive grassland use. These long-term changes have strong effects on biodiversity and ecosystem services, but are not well predicted by environmental and access-related determinants. Our findings can help to develop strategies and policies for areas affected by agricultural land abandonment caused by depopulation, and other socioeconomic shocks, and highlight the need to understand not only why arable land is abandoned, but also what determines its long-term fate.
File: Kaim_LandUsePolicy_2023.pdf
