Vegetation plays an essential role in improving urban environments and enhancing the physical and mental health of residents. However, rapid urbanization has exerted complicated influence on ...vegetation conditions, which remain poorly understood. To assess the impacts of urbanization on the vegetation structure and function in the urban area, we quantified the changes in impervious surface area (ISA) and assessed the impacts of urbanization on vegetation greenness (enhanced vegetation index (EVI)), and gross primary production (GPP) in megacity Shanghai during 2000–2016. The results show that 38.0% and 28.0% decreasing trends of EVI and GPP occurred in peri-urban and rural areas due to land use and land cover conversion, whereas 2.8% and 4.6% increasing trends of EVI and GPP occurred in the central city during 2000–2016 in Shanghai. In addition, the enhancement of EVI and GPP owing to the indirect impact of urbanization increased as the impervious surface coverage (ISC) gradient rose and peaked when the ISC reached ~0.8, which compensated for vegetation loss by 24.6% and 17.0%, respectively. The compensation was more stable and significant in peri-urban areas than urban and rural areas. This study provides detailed data and insights on the impacts of urbanization on vegetation, which may help stakeholders to make better management plans for urban vegetation.
•We characterized the pattern of urbanization in Shanghai during 2000–2016.•We explored the spatial-temporal relationships between vegetation and urbanization.•Land conversion caused vegetation and carbon loss in peri-urban and rural areas.•The enhancement of vegetation peaked at the impervious surface coverage of 0.8.•The enhancement could compensate for vegetation loss by around 20%.
The deadly threat that landslide has brought about is drawing more and more attention to analyze the mechanisms of landslides and the relationship between landslides and climate change. Due to the ...limited record of historical landslides in developing countries, relevant research is mostly conducted in developed countries. Owing to the publicly available global long time-series Landsat images, such unbalance can be avoided by proposing a practical landslide detection model, especially in terms of national scale. This paper takes the advantage of google earth engine platform to synthesize the annual Landsat images covering the national scale of Nepal into one image and builds an end-to-end contour-based landslide detection deep learning framework. The framework consists of two parts, one is potential landslide detection using vegetation index and degradation of DEM, the other is exact landslide detection using semantic segmentation deep learning model based on the contour regions extracted from the detected potential landslide. The proposed method is applied to detect landslides of Nepal in the year of 2015 and achieves a satisfactory performance with 65% recall and 55.35% precision. The performance is 44% higher accurate than similarly published works, validating its promising applicability in practical landslide detection for national cases.
•Resist the imbalanced distribution of landslides using contour-based method.•Avoid feature engineering by applying deep learning framework.•Enlarge the practical applicability by evaluation in Nepal.
Understanding how irrigated areas change over time is vital to effectively manage limited agricultural water resources, but long-term, high-resolution, and spatially explicit datasets are rare. The ...High Plains Aquifer (HPA) in the central United States is one of the largest and most stressed aquifer systems in the world. It supports a $20 billion economy, but groundwater use is unsustainable over much of the aquifer. Emerging cloud computing tools like Google Earth Engine (GEE) make it possible to use the full Landsat record to monitor regional systems like the HPA with high spatial and temporal resolution over multiple decades. However, challenges remain to develop irrigation classification methods that are robust to a wide range of climate conditions and crop types, evolving management, and missing data. Here, we addressed these challenges to produce an annual, moderately high resolution (30 m) irrigation map time series from 1984 to 2017 over the aquifer. Leveraging GEE's extensive data catalog, we combined Landsat imagery, environmental covariables, and a large heterogeneous ground truth dataset to create a single random forest classifier applied annually to the entire region. Following classification, we applied the Bayesian Updating of Land-Cover (BULC) algorithm to fill imagery gaps and reduce commission errors in the provisional irrigation time series. Novel neighborhood greenness indices contributed to an overall 91.4% map accuracy across years; county statistics (r2 = 0.86) were similarly well-matched. Trend analysis of irrigated area through time identified regions of stable, expanding, and declining irrigated area. Given declining aquifer storage, we estimate that up to 24% of currently irrigated area may be lost this century. To date, the map dataset is the longest, highest resolution large-scale record of where and when irrigation occurs. It is freely available for stakeholders, managers, and researchers to inform policies and management decisions, as well as for use in hydrology, agronomy, and climate models.
•Stressed systems like the High Plains Aquifer need irrigation data for management.•Google Earth Engine enables annual tracking of irrigated area over multiple decades.•Data gaps in the early Landsat record can be addressed using Bayesian statistics.•New neighborhood greenness indices contributed to 91% map accuracy across 34 years.•This is the longest, highest resolution large-scale record of irrigation to date.
Efficient Neural Architecture Search methods based on weight sharing have shown good promise in democratizing Neural Architecture Search for computer vision models. There is, however, an ongoing ...debate whether these efficient methods are significantly better than random search. Here we perform a thorough comparison between efficient and random search methods on a family of progressively larger and more challenging search spaces for image classification and detection on ImageNet and COCO. While the efficacies of both methods are problem-dependent, our experiments demonstrate that there are large, realistic tasks where efficient search methods can provide substantial gains over random search. In addition, we propose and evaluate techniques which improve the quality of searched architectures and reduce the need for manual hyper-parameter tuning.
This study explores the extent to which bibliometric indicators based on counts of highly-cited documents could be affected by the choice of data source. The initial hypothesis is that databases that ...rely on journal selection criteria for their document coverage may not necessarily provide an accurate representation of highly-cited documents across all subject areas, while inclusive databases, which give each document the chance to
stand on its own merits
, might be better suited to identify highly-cited documents. To test this hypothesis, an analysis of 2515 highly-cited documents published in 2006 that Google Scholar displays in its
Classic Papers
product is carried out at the level of broad subject categories, checking whether these documents are also covered in Web of Science and Scopus, and whether the citation counts offered by the different sources are similar. The results show that a large fraction of highly-cited documents in the Social Sciences and Humanities (8.6–28.2%) are invisible to Web of Science and Scopus. In the Natural, Life, and Health Sciences the proportion of missing highly-cited documents in Web of Science and Scopus is much lower. Furthermore, in all areas, Spearman correlation coefficients of citation counts in Google Scholar, as compared to Web of Science and Scopus citation counts, are remarkably strong (.83–.99). The main conclusion is that the data about highly-cited documents available in the inclusive database Google Scholar does indeed reveal significant coverage deficiencies in Web of Science and Scopus in several areas of research. Therefore, using these selective databases to compute bibliometric indicators based on counts of highly-cited documents might produce biased assessments in poorly covered areas.
Touchscreen-based devices such as smartphones and tablets are gaining popularity, but their rich input capabilities pose new development and testing complications. To alleviate this problem, we ...present an approach and tool named Reran that permits record-and-replay for the Android smartphone platform. Existing GUI-level record-and-replay approaches are inadequate due to the expressiveness of the smartphone domain, in which applications support sophisticated GUI gestures, depend on inputs from a variety of sensors on the device, and have precise timing requirements among the various input events. We address these challenges by directly capturing the low-level event stream on the phone, which includes both GUI events and sensor events, and replaying it with microsecond accuracy. Moreover, Reran does not require access to app source code, perform any app rewriting, or perform any modifications to the virtual machine or Android platform. We demonstrate RERAN's applicability in a variety of scenarios, including (a) replaying 86 out of the Top-100 Android apps on Google Play; (b) reproducing bugs in popular apps, e.g., Firefox, Facebook, Quickoffice; and (c) fast-forwarding executions. We believe that our versatile approach can help both Android developers and researchers.
Google Colaboratory (also known as Colab) is a cloud service based on Jupyter Notebooks for disseminating machine learning education and research. It provides a runtime fully configured for deep ...learning and free-of-charge access to a robust GPU. This paper presents a detailed analysis of Colaboratory regarding hardware resources, performance, and limitations. This analysis is performed through the use of Colaboratory for accelerating deep learning for computer vision and other GPU-centric applications. The chosen test-cases are a parallel tree-based combinatorial search and two computer vision applications: object detection/classification and object localization/segmentation. The hardware under the accelerated runtime is compared with a mainstream workstation and a robust Linux server equipped with 20 physical cores. Results show that the performance reached using this cloud service is equivalent to the performance of the dedicated testbeds, given similar resources. Thus, this service can be effectively exploited to accelerate not only deep learning but also other classes of GPU-centric applications. For instance, it is faster to train a CNN on Colaboratory's accelerated runtime than using 20 physical cores of a Linux server. The performance of the GPU made available by Colaboratory may be enough for several profiles of researchers and students. However, these free-of-charge hardware resources are far from enough to solve demanding real-world problems and are not scalable. The most significant limitation found is the lack of CPU cores. Finally, several strengths and limitations of this cloud service are discussed, which might be useful for helping potential users.