The detection and analysis of gravitational waves (GW) has opened new windows on the universe in recent years. Our team has developed the GWSkyNet and GWSkyNet-Multi models, which are designed to operate with the limited data publicly available from the LIGO-Virgo collaboration. We employ a combination of neural networks and explainable AI techniques to predict the nature of the detected signals. The XAI part helps us understand the strengths and biases of our models.

Abstract The global gravitational gradient field has not been observed since the decommission of GOCE in 2013. Based on the foundational work of Peidou and Pagiatakis (2019), we advance the concept of GRACE gradiometer mode (GM) for the purpose of using GRACE, GRACE-FO and future gravity space missions as ‘gradiometer missions’. Certainly, the GRACE missions have never carried a gradiometer, it is only the concept of GM that creates a fictitious gradiometer system very similar to GOCE, only the GRACE ‘gradiometers’ have long and variable baselines, an unprecedented paradigm for space-based gravitational gradiometry that extends the bandwidth of the gradient solution. In this contribution, we develop a new configuration for GM that views an individual satellite as the ‘gradiometer’, in order to directly use Level 1A accelerometer measurements in a 10 Hz sampling interval. We apply the new method in geodynamically active regions around the globe, and we demonstrate that using GRACE-C as a ‘gradiometer’ in the single-satellite gradiometer mode (SS-GM) produces higher-fidelity gravitational gradient estimates, clearly delineating tectonic plate boundaries and subduction zones in the Himalayas and North Africa regions, the Aleutian trench, the Java trench, and the Peru-Chile trench. Over Canada, we see the delineation of the Canadian shield, and the effect of glacial isostatic adjustment (GIA) is apparent. We also observe well-known signals resembling terrestrial water storage (TWS) changes in Africa, among others, demonstrating the usefulness of the GM for a wide variety of geoscience applications.

We evaluate cloud simulations using satellite simulators against multiple observational datasets. These simulators have been run within the Geophysical Fluid Dynamics Laboratory’s Atmosphere Model version 4.0 (AM4.0), as well as an alternative configuration where a two-moment Morrison-Gettelman bulk cloud microphysics with prognostic precipitation (MG2) is applied, denoted as AM4-MG2. The modelled cloud spatial distributions, vertical profiles, phase partitioning, cloud-to-precipitation transitions, and radiative effects from AM4.0 and AM4-MG2 compare reasonably well with satellite observations. Model biases include the underestimate of total and low-level clouds, especially optically thin/intermediate clouds, but the overestimate of optically thick clouds, indicating “too few, too bright’ biases. These biases compensate each other and result in reasonable estimates of cloud radiative effects. The underestimate of low-level clouds is associated with too frequent and too early drizzle/precipitation formation. The precipitation bias is improved in AM4-MG2, where the autoconversion scheme initiates the precipitation more realistically.

Forces that govern the motion of planetary flows originate from interactions occurring at scales that are orders of magnitude smaller than the flow itself. These mesoscale and sub–mesoscale eddies contribute to large-scale flow mixing, particle entrainment, and tracer transport. Therefore, oceanic simulations must accurately characterize their contributions to capture the flow dynamics. In this work, a stochastic representation of the primitive equations, derived within the framework of modeling under location uncertainty (LU), is proposed. This framework decomposes the velocity into a smooth–in–time large–scale component and a random field, accounting for unresolved mesoscale and sub–mesoscale motions. The LU framework is unique in its derivation, as it is obtained from physical conservation principles through a stochastic version of the Reynolds transport theorem. Two data-driven approaches are developed to model the random noise using proper orthogonal decomposition and dynamic mode decomposition methods. For higher resolution, model based noise as well as mixture of model based and data driven noise with observational constraints are proposed and compared. Numerical simulations demonstrate that the LU framework improves flow prediction for gyre circulation compared to its deterministic counterpart. Enhanced flow mixing, jet characteristics, and tracer transport are observed. Additionally, terms in the LU framework are analyzed for their contribution to structuring the flow.

Global river models are crucial for simulating and forecasting floods, especially in regions lacking dedicated flood forecasting systems. Reservoir operations significantly affect flow simulations, impacting forecast accuracy. In this study, we improved a storage-outflow-type global reservoir operation scheme of the global river hydrodynamics model CaMa-Flood, which calculates reservoir outflow based on reservoir storage. First, a non-linear storage-outflow function using convex and concave curves was implemented to enhance flood control and reduce the number of required parameters. Second, a gradual transition in the storage-outflow function was proposed to avoid simulated outflow instability. Third, the outflow during the water-use phase was recalibrated to reflect water availability in dry periods, rather than using annual mean discharge. A case study involving 85 dams in Mainland Southeast Asia was conducted to evaluate the efficacy of the three key modifications by comparing the proposed reservoir operation scheme against three degraded schemes. Using a linear storage-outflow function caused 43% of the 61 reservoirs with a flood control capacity greater than 100 million cubic meters (MCM) to overutilize their flood control capacity during the non-flood water-excess phase. The proposed non-linear function reduced the number of such reservoirs to 27%. In addition, the gradual transition of the storage-outflow function eliminated unrealistic outflow oscillation. Furthermore, parameter estimation for water-use outflow based on reservoir water-use capacity and dry period river discharge enabled better utilization of reservoir capacity during dry seasons. Overall, the proposed reservoir operation scheme holds promise for enhancing the accuracy of global flood simulation within global river models.

Robust increases in temperatures will occur globally across the twenty-first century; however, for some regions, sign changes in rainfall remain uncertain. Navigating this uncertainty is crucial in addressing drought-related challenges faced by climate-exposed sectors. New Zealand represents these challenges, exhibiting significant model uncertainty in warm-season rainfall change. Here, we examine how temperature-driven increases in potential evapotranspiration interact with contrasting storylines of future rainfall to explore drought outcomes for New Zealand. In our drying storyline, we find that increasing temperatures accompanied by less rainfall bring forward the onset of drought several months and delay its termination. In the wetting storyline, increases in rainfall partly offset the temperature-induced drying effect, leading to minor reductions in soil moisture. Examining extreme years, the average hydrological year in the future becomes comparable to the driest years of the current climate, while the worst future events exhibit unprecedented drought severity.

A document by Derek Keir. Click on the document to view its contents.

Stratospheric water vapor (SWV) is a greenhouse gas that has a significant, yet uncertain, impact on the Earth’s climate through its radiative effect and feedback. As the climate changes, it is thus critical to monitor and understand changes in SWV. NASA’s Microwave Limb Sounder (MLS) aboard the Aura satellite has observed SWV since 2004 but will soon reach end of life. The Stratospheric Aerosol and Gas Experiment (SAGE) missions observe SWV as well, with the SAGE III instrument operating on the International Space Station (ISS) since 2017. We use the constituent data assimilation capabilities of NASA’s Goddard Earth Observing System (GEOS) to demonstrate that the up to 30 SAGE III/ISS profiles each day provide a useful constraint on SWV from 500 to 1500 K over the observed midlatitudes and tropics. We conclude that by assimilating SAGE III/ISS SWV into GEOS we can continue to monitor SWV after Aura MLS.

China has been experiencing severe ozone (O3) pollution, and the high concentration of O3 exposure can damage crop production, which will threaten the national food security. To better estimate the impact of O3 on crop production loss (CPL), we collect O3 data from CNEMC (China National Environmental Monitoring Centre) and CMIP6 (Coupled Model Intercomparison Project Phase 6) models to quantify O3-induced production losses in the four main crops (i.e., wheat, rice, maize, and soybean) during the historical (2013-2018) and future (2019-2099) periods. Results confirm that the impacts of O3 damage on CPLs become more and more serious during 2013-2018 in China, with the six-year mean losses of 12.2 Mt (8.5%) for wheat, 8.4 Mt (3.8%) of rice, 4.3 Mt (1.6%) of maize, and 0.7 Mt (4.8%) of soybean. The worsened O3 pollution under SSP3-7.0 scenario during the late-century will cause 283.6% (170.3%, 479.4%, 168.0%) increase in wheat (rice, maize, soybean) production losses relative to that in 2018, while the improved O3 air quality under SSP1-2.6 scenario can decrease 73.8% (69.3%, 80.8%, 75.5%) wheat (rice, maize, soybean) yield losses during the late-century relative to 2018. China has the largest population and food consumption worldwide, and the persistent O3 deterioration does have exacerbated the loss of agricultural productions. Therefore, stricter control measures are urgently needed to improve O3 air quality for ensuring national and even per capita food security in future.

The impact of land use on ecosystems has reached critical levels, jeopardizing biosphere integrity. A key indicator that quantifies, monitors, and analyses such impacts is the Human Appropriation of Net Primary Production (HANPP). Assessing HANPP requires integrating data from sources such as remote sensing and census statistics, as well as modelled data like potential Net Primary Production (NPP), which reflects NPP without land use. Although the availability of global land cover data at high spatial detail from remote sensing has improved, with resolutions reaching 30 arcseconds (about 1 km) and higher, global NPP results from Dynamic Global Vegetation Models (DGVMs) are still unavailable at this resolution. This spatial mismatch causes uncertainties, as simple interpolation methods fail to capture fine-scaled productivity patterns. We here present a parsimonious method to downscale NPP, using the Miami NPP model with temperature and precipitation data as readily available auxiliary information at high spatial resolution. Our method uses a moving window approach with Gaussian convolution to minimize downscaling artefacts. We demonstrate this Smooth Auxiliary Data (SAD) downscaling approach by downscaling potential NPP results from the LPJ-GUESS-DGVM model for the year 2010 from 30arcmin to 30arcsec resolution. This approach, requiring low computational cost, generates fine-scaled productivity patterns and aligns with alternative models for smaller geographic units, offering a solution until high-resolution DGVM results become feasible.

Unsustainable rates of groundwater (GW) depletion make GW management a priority. Effective GW management is hindered by the uncertainty in the predictions of aquifer models, but the increase of geodetic surface deformation data can improve aquifer characterization. We integrate surface deformation measurements into a Bayesian inference framework to infer aquifer permeability in a poroelastic model. We demonstrate the applicability of this technique using a Nevada pumping test with both Interferometric Synthetic Aperture Radar (InSAR) and Global Positioning System (GPS) surface deformation data. We infer the lateral permeability variations in the aquifer with high spatial resolution and identify the information content of each data set. For the Nevada case, a single InSAR surface deformation map provides significantly more information than multiple GPS time series. As the availability of global InSAR data continues to grow, geomechanical inversion of geodetic surface deformation data will become a valuable technique for aquifer characterization and management.

Differences in slab dehydration and fluid transport influence the distribution of magmatism and seismicity in subduction zones. Southwest Japan, particularly Kyushu and Shikoku/Chugoku, exhibits significant along-arc variation in the spatial distribution of Quaternary arc volcanoes and adakites as well as sub-arc tectonic tremors. Using two-dimensional numerical modeling, we quantitatively evaluate melting and fluid transport via a weak hydrous layer on the slab interface since the subduction initiation to clarify the observed key differences. In Kyushu, our model shows that dehydration of the colder subducting slab is mostly completed at sub-arc depths of 80–120 km, triggering intense flux melting in the mantle wedge without slab melting for adakite and an absence of sub-forearc fluid percolation for tremors. Whereas, the model for Shikoku/Chugoku shows shallower sub-forearc dehydration of the warmer slab (<80 km), allowing less flux melting, slab melting for adakite, and sub-forearc fluid percolation for tremors.

The biological pump, a fundamental process governing atmospheric CO2, rapidly transfers particulate inorganic and organic carbon (PIC and POC) from surface waters to the deep sea but is inherently highly variable in space and time, and thus poorly observed. Here we synthesize PIC and POC data from satellites, CTD-profiled optical sensors (birefringence and transmissometer), and from in-situ pumps samples from GEOTRACES transects spanning 20,000 km from the North Pacific to Southern Ocean. High resolution profile data from PIC sensors revealed strong subsurface maxima in the deepest euphotic zone waters of oligotrophic gyres; furthermore, data showed high concentrations of PIC penetrating to >500 m south of the Subarctic Front (45°N – 35°N), at the equator, and north of the Antarctic Polar Front (45°S – 55°S) indicating high carbon export in these regions. We developed a new interpolation scheme for satellite data that improved matchups with ship observations. North of the Antarctic Polar Front (APF), PIC sensor data was generally well aligned with sample PIC; however, a positive bias of satellite PIC was found in poor retrieval regions. South of the APF, both satellite and birefringence sensor greatly overestimated PIC by factors of >25 and 12, respectively, compared to sample PIC which averaged 15nM. The unanticipated discovery of a strong non-carbonate particle birefringence source coupled with a microscopic investigation of pump samples leads us hypothesize that internal reflection within bubbles and/or body structures of heavily silicified colony-forming diatoms (Fragilariopsis and Pseudo-nitzschia) is the cause for both anomalous birefringence and backscattered satellite radiances.

The ability to optically detect reflected and emitted photons from vegetation has transformed our understanding of forest ecophysiology, enabling detailed assessments of photosynthesis and evapotranspiration across scales. This chapter explores the links between physiological processes, remote sensing signals and broader ecosystem fluxes. By integrating remote sensing data across scales (leaf, tower, aircraft, and satellite), we capture the interactions between forest structure, function, and environmental conditions. We address scaling challenges and consider the convergence and divergence of remote sensing signals and forest ecophysiological processes. Convergence in plant traits creates predictable relationships between canopy structure and function, facilitating estimates of forest processes at large scales, while divergence arises from stressors and nuanced interactions at finer scales. Ultimately, we aim to clarify when and why these phenomena emerge and their significance for characterizing forest functional dynamics.

A document by Caleb J Robbins. Click on the document to view its contents.

The concept of the ecological niche is fundamental to understanding species distributions but it often overlooks the critical role of demography in shaping said distributions. Conversely, demographic theory has traditionally neglected how vital rates vary across environments, limiting our understanding of population dynamics across species’ ranges. Over 50 years ago, Maguire proposed conceptualising the ecological niche as composed of multiple “demographic niches” corresponding to separate vital rates such as survival, development, and reproduction. Despite its potential for ecological research, Maguire’s perspective never really took off. Here, we introduce the “demographic niche concept” (DNC), a contemporary evolution of Maguire’s perspective, integrating recent advances in niche theory and demography to formalise the DNC. We first outline the theoretical foundations of the DNC, focusing on vital rate variation along environmental gradients and its role in shaping ecological niches. We then formalise the definition of the DNC, explore how demographic niches and their boundaries are determined, and introduce a framework for applying the DNC in environmental and geographic spaces. Importantly, we present the DNC as a tool for conservation planning by improving our ability to identify critical areas where specific demographic processes are vital for species persistence or absent under global change. Throughout this review, we also explore methods for modelling and integrating demographic niches and discuss their relevance for addressing global challenges such as climate change. By integrating the ecological niche with demographic theory, we propose testable hypotheses on how and why vital rates vary within niche space, the current frontier of the DNC. Finally, we identify key knowledge gaps and suggest research directions to address these gaps. By advancing the conceptual and methodological tools of the DNC, this review establishes a foundation for future empirical research and applications, offering new directions for ecological theory and conservation amidst a changing world.

A document by Biqing Zhu. Click on the document to view its contents.

Data-driven artificial intelligence (AI)-based weather prediction (AIWP) models have demonstrated significant potential in weather forecasts, facilitating paradigm shift of prediction from a deductive to an inductive inference. However, this shift raises concerns regarding the performance of the AIWP models in severe weather forecasting. Tropical cyclones (TCs) are one of the most typical cases of severe weather forecasting. In this study, we compare forecasts of Western Pacific TCs in 2023 produced by the AIWP model, Pangu-Weather, with those generated by numerical weather prediction (NWP) models, specifically the European Centre for Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction (NCEP), in the operational context. We analyze the impact of different initial conditions on the AIWP model Pangu-Weather, in TC forecasting. Our analysis includes statistical evaluation of forecast skills related to TC activity, track, intensity, and a case study on the physical structure of TCs. The Pangu-Weather model exhibits superior forecast skills compared to the NWP model regarding TC tracks and environmental variables within TC activity domains, particularly at longer leading times. However, the overly smooth forecasts from Pangu-Weather lead to significant underestimations of intensity and a weakened dynamic-thermodynamic structure of TCs. Additionally, Pangu-Weather shows reduced sensitivity to initial conditions concerning TC structure and intensity, potentially attributable to the limitations of the training dataset and deep learning model employed. Enhancing the application of higher-quality initial conditions and the exploring hybrid models that integrate physical processes with data-driven methods could significantly improve the effectiveness of AIWP models in severe weather forecasting.