Search

Global and local modeling is essential for image super-resolution tasks. However, current efforts often lack explicit consideration of the cross-scale knowledge in large-scale earth observation scenarios, resulting in suboptimal single-scale representations in global and local modeling. The key motivation of this work is inspired by two observations: 1) There exists hierarchical features at the local and global regions in remote sensing images, and 2) they exhibit scale variation of similar ground objects (e.g. cross-scale similarity). In light of these, this paper presents an effective method to grasp the global and local image hierarchies by systematically exploring the cross-scale correlation. Specifically, we developed a Cross-scale Self-Attention (CSA) to model the global features, which introduces an auxiliary token space to calculate cross-scale self-attention matrices, thus exploring global dependency from diverse token scales. To extract the cross-scale localities, a Cross-scale Channel Attention (CCA) is devised, where multi-scale features are explored and progressively incorporated into an enriched feature. Moreover, by hierarchically deploying CSA and CCA into transformer groups, the proposed Cross-scale Hierarchical Transformer (CHT) can effectively explore cross-scale representations in remote sensing images, leading to a favorable reconstruction performance. Comprehensive experiments and analysis on four remote sensing datasets have demonstrated the superiority of CHT in both simulated and real-world remote sensing scenes. In particular, our CHT outperforms the state-of-the-art approach (TransENet) in terms of PSNR by 0.11 dB on average, but only accounts for 54.8% of its parameters.

Global and local modeling is essential for image super-resolution tasks. However, current efforts often lack explicit consideration of the cross-scale knowledge in large-scale earth observation scenarios, resulting in suboptimal single-scale representations in global and local modeling. The key motivation of this work is inspired by two observations: 1) There exists hierarchical features at the local and global regions in remote sensing images, and 2) they exhibit scale variation of similar ground objects (e.g. cross-scale similarity). In light of these, this paper presents an effective method to grasp the global and local image hierarchies by systematically exploring the cross-scale correlation. Specifically, we developed a Cross-scale Self-Attention (CSA) to model the global features, which introduces an auxiliary token space to calculate cross-scale self-attention matrices, thus exploring global dependency from diverse token scales. To extract the cross-scale localities, a Cross-scale Channel Attention (CCA) is devised, where multi-scale features are explored and progressively incorporated into an enriched feature. Moreover, by hierarchically deploying CSA and CCA into transformer groups, the proposed Cross-scale Hierarchical Transformer (CHT) can effectively explore cross-scale representations in remote sensing images, leading to a favorable reconstruction performance. Comprehensive experiments and analysis on four remote sensing datasets have demonstrated the superiority of CHT in both simulated and real-world remote sensing scenes. In particular, our CHT outperforms the state-of-the-art approach (TransENet) in terms of PSNR by 0.11 dB on average, but only accounts for 54.8% of its parameters. [ABSTRACT FROM AUTHOR]

Accurate crop classification is of vital importance for agricultural water management. Most researchers have achieved crop classification by model optimization in the same temporal and regional domain by adjusting the value of input features. This study aims to improve the accuracy of crop classification across temporal and spatial domains. Sentinel-2 satellite imagery is employed for crop classification training and prediction in selected farming areas of Heilongjiang Province by calculating vegetation indices and constructing sequential input feature datasets. The HUNTS filtering method was used to mitigate the influence of cloud cover, which increased the stability and completeness of the input feature data across different years. To address the issue of shifts in the input feature values during cross-scale classification, this study proposes the hypothesis testing distribution method (HTDM). This method balances the distribution of input feature values in the test set even without knowing the crop distribution, thereby enhancing the accuracy of the classification test set. The results indicate that the HTDM significantly improves prediction accuracy in cases of substantial image quality variance. In 2022, the recognition accuracy for crop types at all farms processed by the HTDM was above 87%, showcasing the strong robustness of the HTDM. [ABSTRACT FROM AUTHOR]

Semantic segmentation of remote sensing (RS) images plays a vital role in a variety of fields, including urban planning, natural disaster monitoring, and land resource management. Due to the complexity and low resolution of RS images,many approaches have been proposed to handle the related task. However, these previously developed approaches dedicate to contextual interaction but ignore the cross-scale semantic correlation and multi-scale boundary information. Therefore, we propose a Cross-scale Graph Interaction Network (CGIN) to address semantic segmentation problems of RS images, which consists of a semantic branch and a boundary branch. In the semantic branch, we first apply atrous convolution to extract multi-scale semantic features of RS images. Particularly, based on the multi-scale semantic features, a Cross-scale Graph Interaction (CGI) module is introduced, which establishes cross-scale graph structures and performs adaptive graph reasoning to capture the cross-scale semantic correlation of RS objects. In the boundary branch, we propose a Multiscale Boundary Feature Extraction (MBFE) module that utilizes atrous convolutions with different dilation rates to extract multi-scale boundary features. Finally, to address the problem of sparse boundary pixels in the fusion process of the two branches, we propose a Multi-scale Similarity-guided Aggregation (MSA) module by calculating the similarity of semantic features and boundary features at the corresponding scale, which can emphasize the boundary information in semantic features. Our proposed CGIN outperforms state-of-the-art approaches in numerical experiments conducted on two benchmark remote sensing datasets. [ABSTRACT FROM AUTHOR]

Context: Coral reef resilience is the product of multiple interacting processes that occur across various interacting scales. This complexity presents challenges for identifying solutions to the ongoing worldwide decline of coral reef ecosystems that are threatened by both local and global human stressors. Objectives: We highlight how coral reef resilience is studied at spatial, temporal, and functional scales, and explore emerging technologies that are bringing new insights to our understanding of reef resilience. We then provide a framework for integrating insights across scales by using new and existing technological and analytical tools. We also discuss the implications of scale on both the ecological processes that lead to declines of reefs, and how we study those mechanisms. Methods: To illustrate, we present a case study from Kāneʻohe Bay, Hawaiʻi, USA, linking remotely sensed hyperspectral imagery to within-colony symbiont communities that show differential responses to stress. Results: In doing so, we transform the scale at which we can study coral resilience from a few individuals to entire ecosystems. Conclusions: Together, these perspectives guide best practices for designing management solutions that scale from individuals to ecosystems by integrating multiple levels of biological organization from cellular processes to global patterns of coral degradation and resilience. [ABSTRACT FROM AUTHOR]

Macroecology research seeks to understand ecological phenomena with causes and consequences that accumulate, interact, and emerge across scales spanning several orders of magnitude. Broad‐extent, fine‐grain information (i.e., high spatial resolution data over large areas) is needed to adequately capture these cross‐scale phenomena, but these data have historically been costly to acquire and process. Unoccupied aerial systems (UAS or drones carrying a sensor payload) and the National Ecological Observatory Network (NEON) make the broad‐extent, fine‐grain observational domain more accessible to researchers by lowering costs and reducing the need for highly specialized equipment. Integration of these tools can further democratize macroecological research, as their strengths and weaknesses are complementary. However, using these tools for macroecology can be challenging because mental models are lacking, thus requiring large up‐front investments in time, energy, and creativity to become proficient. This challenge inspired a working group of UAS‐using academic ecologists, NEON professionals, imaging scientists, remote sensing specialists, and aeronautical engineers at the 2019 NEON Science Summit in Boulder, Colorado, to synthesize current knowledge on how to use UAS with NEON in a mental model for an intended audience of ecologists new to these tools. Specifically, we provide (1) a collection of core principles for collecting high‐quality UAS data for NEON integration and (2) a case study illustrating a sample workflow for processing UAS data into meaningful ecological information and integrating it with NEON data collected on the ground—with the Terrestrial Observation System—and remotely—from the Airborne Observation Platform. With this mental model, we advance the democratization of macroecology by making a key observational domain—the broad‐extent, fine‐grain domain—more accessible via NEON/UAS integration. [ABSTRACT FROM AUTHOR]

Agent-based modelling (ABM) simulates Social-Ecological-Systems (SESs) based on the decision-making and actions of individual actors or actor groups, their interactions with each other, and with ecosystems. Many ABM studies have focused at the scale of villages, rural landscapes, towns or cities. When considering a geographical, spatially-explicit domain, current ABM architecture is generally not easily translatable to a regional or global context, nor does it acknowledge SESs interactions across scales sufficiently; the model extent is usually determined by pragmatic considerations, which may well cut across dynamical boundaries. With a few exceptions, the internal structure of governments is not included when representing them as agents. This is partly due to the lack of theory about how to represent such as actors, and because they are not static over the time-scales typical for social changes to have significant effects. Moreover, the relevant scale of analysis is often not known a priori, being dynamically determined, and may itself vary with time and circumstances. There is a need for ABM to cross the gap between micro-scale actors and larger-scale environmental, infrastructural and political systems in a way that allows realistic spatial and temporal phenomena to emerge; this is vital for models to be useful for policy analysis in an era when global crises can be triggered by small numbers of micro-level actors. We aim with this thought-piece to suggest conceptual avenues for implementing ABM to simulate SESs across scales, and for using big data from social surveys, remote sensing or other sources for this purpose. [ABSTRACT FROM AUTHOR]

Abstract: China's Natural Forest Protection Program (NFPP) and Sloping Land Conversion Program (SLCP), introduced in 1998 and 1999, respectively, are integral parts of the world's largest reforestation effort. State-reported forest cover data indicate effective policy implementation through net forest cover expansion but overlook the scale-dependence of and spatial variation in forest cover change patterns and also lack reliable data on small-scale and illegal logging. As a result, there is considerable uncertainty over the spatial distribution of forest cover change and ultimately the policies' effectiveness at increasing forest cover. This research uses Landsat Thematic Mapper imagery-derived multitemporal Tasseled Cap variables and a decision tree classifier to map short- and long-term forest cover change across three administrative levels in the priority conservation area of Diqing Tibetan Autonomous Prefecture in Yunnan Province. Results indicate a 73% reduction in the rate of forest cover loss and a more than doubled rate of forest cover gain from 1990–1999 to 1999–2009 across the prefecture, both of which support a positive assessment of policy implementation. However, prefectural results are countered by spatially disparate forest cover gain and loss trends at the county- and township-level in the decade following the policies' introductions. Further, more than half of Diqing's townships, mainly those in the prefecture's south where tourism has been rapidly developing, saw continued net forest cover loss attributable to small-scale timber harvesting for tourism-driven construction. This research thus exposes cross-scale spatially disparate forest cover change indicative of highly differentiated policy implementation effectiveness, and shows the pattern by which regional development has redirected, rather than reduced, forest cover loss, contrary to the goals of the NFPP and SLCP. [Copyright &y& Elsevier]