Search for: All records

Abstract In this study, we investigate the air temperature response to land-use and land-cover change (LULCC; cropland expansion and deforestation) using subgrid land model output generated by a set of CMIP6 model simulations. Our study is motivated by the fact that ongoing land-use activities are occurring at local scales, typically significantly smaller than the resolvable scale of a grid cell in Earth system models. It aims to explore the potential for a multimodel approach to better characterize LULCC local climatic effects. On an annual scale, the CMIP6 models are in general agreement that croplands are warmer than primary and secondary land (psl; mainly forests, grasslands, and bare ground) in the tropics and cooler in the mid–high latitudes, except for one model. The transition from warming to cooling occurs at approximately 40°N. Although the surface heating potential, which combines albedo and latent heat flux effects, can explain reasonably well the zonal mean latitudinal subgrid temperature variations between crop and psl tiles in the historical simulations, it does not provide a good prediction on subgrid temperature for other land tile configurations (crop vs forest; grass vs forest) under Shared Socioeconomic Pathway 5–8.5 (SSP5–8.5) forcing scenarios. A subset of simulations with the CESM2 model reveals that latitudinal subgrid temperature variation is positively related to variation in net surface shortwave radiation and negatively related to variation in the surface energy redistribution factor, with a dominant role from the latter south of 30°N. We suggest that this emergent relationship can be used to benchmark the performance of land surface parameterizations and for prediction of local temperature response to LULCC. 

In the space physics community, processing and combining observational and modeling data from various sources is a demanding task because they often have different formats and use different coordinate systems. The Python package GeospaceLAB has been developed to provide a unified, standardized framework to process data. The package is composed of six core modules, including DataHub as the data manager, Visualization for generating publication quality figures, Express for higher-level interfaces of DataHub and Visualization , SpaceCoordinateSystem for coordinate system transformations, Toolbox for various utilities, and Configuration for preferences. The core modules form a standardized framework for downloading, storing, post-processing and visualizing data in space physics. The object-oriented design makes the core modules of GeospaceLAB easy to modify and extend. So far, GeospaceLAB can process more than twenty kinds of data products from nine databases, and the number will increase in the future. The data sources include, e.g., measurements by EISCAT incoherent scatter radars, DMSP, SWARM, and Grace satellites, OMNI solar wind data, and GITM simulations. In addition, the package provides an interface for the users to add their own data products. Hence, researchers can easily collect, combine, and view multiple kinds of data for their work using GeospaceLAB. Combining data from different sources will lead to a better understanding of the physics of the studied phenomena and may lead to new discoveries. GeospaceLAB is an open source software, which is hosted on GitHub. We welcome everyone in the community to contribute to its future development. 

Properties of molecules are indicative of their functions and thus are useful in many applications. With the advances of deep-learning methods, computational approaches for predicting molecular properties are gaining increasing momentum. However, there lacks customized and advanced methods and comprehensive tools for this task currently.

Here, we develop a suite of comprehensive machine-learning methods and tools spanning different computational models, molecular representations and loss functions for molecular property prediction and drug discovery. Specifically, we represent molecules as both graphs and sequences. Built on these representations, we develop novel deep models for learning from molecular graphs and sequences. In order to learn effectively from highly imbalanced datasets, we develop advanced loss functions that optimize areas under precision–recall curves (PRCs) and receiver operating characteristic (ROC) curves. Altogether, our work not only serves as a comprehensive tool, but also contributes toward developing novel and advanced graph and sequence-learning methodologies. Results on both online and offline antibiotics discovery and molecular property prediction tasks show that our methods achieve consistent improvements over prior methods. In particular, our methods achieve #1 ranking in terms of both ROC-AUC (area under curve) and PRC-AUC on the AI Cures open challenge for drug discovery related to COVID-19.

Our source code is released as part of the MoleculeX library (https://github.com/divelab/MoleculeX) under AdvProp.

Supplementary data are available at Bioinformatics online.

 

Climate change is destabilizing permafrost landscapes, affecting infrastructure, ecosystems, and human livelihoods. The rate of permafrost thaw is controlled by surface and subsurface properties and processes, all of which are potentially linked with each other. However, no standardized protocol exists for measuring permafrost thaw and related processes and properties in a linked manner. The permafrost thaw action group of the Terrestrial Multidisciplinary distributed Observatories for the Study of the Arctic Connections (T-MOSAiC) project has developed a protocol, for use by non-specialist scientists and technicians, citizen scientists, and indigenous groups, to collect standardized metadata and data on permafrost thaw. The protocol introduced here addresses the need to jointly measure permafrost thaw and the associated surface and subsurface environmental conditions. The parameters measured along transects include: snow depth, thaw depth, vegetation height, soil texture, and water level. The metadata collection includes data on timing of data collection, geographical coordinates, land surface characteristics (vegetation, ground surface, water conditions), as well as photographs. Our hope is that this openly available dataset will also be highly valuable for validation and parameterization of numerical and conceptual models, and thus to the broad community represented by the T-MOSAiC project. 

Abstract. Infrastructure built on perennially frozen ice-richground relies heavily on thermally stable subsurface conditions. Climate-warming-induced deepening of ground thaw puts such infrastructure at risk offailure. For better assessing the risk of large-scale future damage to Arcticinfrastructure, improved strategies for model-based approaches are urgentlyneeded. We used the laterally coupled 1D heat conduction model CryoGrid3to simulate permafrost degradation affected by linear infrastructure. Wepresent a case study of a gravel road built on continuous permafrost (Daltonhighway, Alaska) and forced our model under historical and strong futurewarming conditions (following the RCP8.5 scenario). As expected, the presenceof a gravel road in the model leads to higher net heat flux entering theground compared to a reference run without infrastructure and thus a higherrate of thaw. Further, our results suggest that road failure is likely aconsequence of lateral destabilisation due to talik formation in the groundbeside the road rather than a direct consequence of a top-down thawing anddeepening of the active layer below the road centre. In line with previousstudies, we identify enhanced snow accumulation and ponding (both aconsequence of infrastructure presence) as key factors for increased soiltemperatures and road degradation. Using differing horizontal modelresolutions we show that it is possible to capture these key factors and theirimpact on thawing dynamics with a low number of lateral model units,underlining the potential of our model approach for use in pan-Arctic riskassessments. Our results suggest a general two-phase behaviour of permafrost degradation:an initial phase of slow and gradual thaw, followed by a strong increase inthawing rates after the exceedance of a critical ground warming. The timing ofthis transition and the magnitude of thaw rate acceleration differ stronglybetween undisturbed tundra and infrastructure-affected permafrost ground. Ourmodel results suggest that current model-based approaches which do notexplicitly take into account infrastructure in their designs are likely tostrongly underestimate the timing of future Arctic infrastructure failure. By using a laterally coupled 1D model to simulate linearinfrastructure, we infer results in line with outcomes from more complex 2Dand 3D models, but our model's computational efficiency allows us to accountfor long-term climate change impacts on infrastructure from permafrostdegradation. Our model simulations underline that it is crucial to considerclimate warming when planning and constructing infrastructure on permafrost asa transition from a stable to a highly unstable state can well occur withinthe service lifetime (about 30 years) of such a construction. Such atransition can even be triggered in the coming decade by climate change forinfrastructure built on high northern latitude continuous permafrost thatdisplays cold and relatively stable conditions today. 

This study applies an indicators framework to investigate climate drivers of tundra vegetation trends and variability over the 1982–2019 period. Previously known indicators relevant for tundra productivity (summer warmth index (SWI), coastal spring sea-ice (SI) area, coastal summer open-water (OW)) and three additional indicators (continentality, summer precipitation, and the Arctic Dipole (AD): second mode of sea level pressure variability) are analyzed with maximum annual Normalized Difference Vegetation Index (MaxNDVI) and the sum of summer bi-weekly (time-integrated) NDVI (TI-NDVI) from the Advanced Very High Resolution Radiometer time-series. Climatological mean, trends, and correlations between variables are presented. Changes in SI continue to drive variations in the other indicators. As spring SI has decreased, summer OW, summer warmth, MaxNDVI, and TI-NDVI have increased. However, the initial very strong upward trends in previous studies for MaxNDVI and TI-NDVI are weakening and becoming spatially and temporally more variable as the ice retreats from the coastal areas. TI-NDVI has declined over the last decade particularly over High Arctic regions and southwest Alaska. The continentality index (CI) (maximum minus minimum monthly temperatures) is decreasing across the tundra, more so over North America than Eurasia. The relationship has weakened between SI and SWI and TI-NDVI, as the maritime influence of OW has increased along with total precipitation. The winter AD is correlated in Eurasia with spring SI, summer OW, MaxNDVI, TI-NDVI, the CI and total summer precipitation. This winter connection to tundra emphasizes the role of SI in driving the summer indicators. The winter (DJF) AD drives SI variations which in turn shape summer OW, the atmospheric SWI and NDVI anomalies. The winter and spring indicators represent potential predictors of tundra vegetation productivity a season or two in advance of the growing season.

 

The key idea of current deep learning methods for dense prediction is to apply a model on a regular patch centered on each pixel to make pixel-wise predictions. These methods are limited in the sense that the patches are determined by network architecture instead of learned from data. In this work, we propose the dense transformer networks, which can learn the shapes and sizes of patches from data. The dense transformer networks employ an encoder-decoder architecture, and a pair of dense transformer modules are inserted into each of the encoder and decoder paths. The novelty of this work is that we provide technical solutions for learning the shapes and sizes of patches from data and efficiently restoring the spatial correspondence required for dense prediction. The proposed dense transformer modules are differentiable, thus the entire network can be trained. We apply the proposed networks on biological image segmentation tasks and show superior performance is achieved in comparison to baseline methods.