Natural flood risk management in the tropics: Prospects in the biodiverse archipelagic nation of the Philippines 

On 10-11 September 2023, Storm Daniel, an exceptionally deep, early-season Mediterranean cyclone triggered an extreme flood in the city of Derna, Libya. The flood destroyed two dams that were built to protect the city, resulting in more than 5,000 casualties, making it one of the deadliest hydrometeorological disasters on record. Here, we aim to unravel (a) whether the catastrophic impact was inevitable, caused by an extremely rare storm, or was it foredoomed because of inadequate dam design, (b) how pivotal were the dam bursts in the tragic outcome, and (c) what insights this event provides for risk management and mitigation strategies elsewhere. Based on atmospheric reanalysis, satellite-based precipitation data, and rainfall-runoff modelling, we estimate the return period of both rainfall and flood to be only a few decades, meaning that the event was not as extreme as could have been implied by its disastrous consequences. Hydraulic simulations of the flood's stream power reveal that the damage caused by the collapse of the dams was approximately 20 times higher than would have been if dams were not built in the Wadi. Our methodological framework, which is based on globally available data, is applicable to risk assessment in other regions 1 as well. It underscores the importance of uncertainty-aware frequency analysis and suggests that distributed flood prevention and warning techniques should be favourable in mitigating flood risks.

Model Spread and Bias in Snowpack in the American Mountain WestTravis Aerenson,a Daniel McCoy,aGregory Elsaesser,bca University of Wyoming, Laramie, WYb Columbia University, New York, NYC NASA-GISS, New York, NYCorresponding author : Travis Aerenson, taerenso@uwyo.eduABSTRACTAs the climate is rapidly changing, we are faced with questions about which regions will become drier and which wetter. In many regions such as the American Mountain West the amount of moisture available for agriculture, drinking water, and industry is dependent on the depth of the wintertime snowpack that melts throughout the spring and summer, filling the downstream rivers and reservoirs. We provide an overview on how snowpack in the Mountain West relates to interannual climate variability in observations and reanalysis and how snowpack in reanalysis has changed during the past century of warming. We then compare these changes with the historical simulations from an ensemble of CMIP6 Global Climate Models (GCMs) to determine how skillful they are at predicting snowpack in the Mountain West, which models are the most skillful, and how we can constrain future predictions by selecting only the models with the best simulation of historical snowpack. We perform a decomposition of Snow Water Equivalent (SWE) trends and variability in reanalysis and CMIP6 models to determine if the model deficiencies in their SWE simulation are due to issues relating to their simulation of precipitation, fraction of precipitation occurring as snow, or the melt-rate of the snowpack. Finally, we construct machine learning emulator models trained on reanalysis that are useful diagnostic tools for the snowpack biases and show that most of the spread in the GCMs’ prediction of historical snowpack change is due to spread in their simulation of regional surface temperature and moisture convergence.1. IntroductionAs the Earth is warming due to anthropogenic emissions of heat-trapping gasses, we anticipate numerous changes to global circulations and local water cycle that will alter regional moisture availability and may cause regional water stress (e.g., Held & Soden, 2006; Pendergrass & Hartmann, 2014; Reed et al., 2023). Hence, it is important to determine if a region is expected to become more moist or arid, and if any adaptive moisture storage methods may be necessary to provide ample water for drinking, agriculture, and industry in a warmer future. In this study we examine the trends and variability in moisture availability in the American Mountain West, defined by the region bounded by 37° to 48° latitude and 247° to 256° longitude, which roughly encompasses the mountainous regions of Colorado, Utah, Wyoming, and Montana, with 3 primary questions in mind: 1) how does large scale variability in the atmospheric and oceanic circulations impact regional moisture availability in the Mountain West, 2) how has moisture in this region changed during the historical global warming during the 20th century, and 3) what aspects of GCMs could be improved to most effectively increase the fidelity of their predictions of regional water stress?The American Mountain West is a continental climate with complex terrain, where most of the run-off throughout the year originally precipitates as wintertime snow (Li et al., 2017). The seasonal snowpack acts as a reservoir, allowing the wintertime precipitation to run-off throughout the spring and summer. Understanding how precipitation and snow-storage is changing is important for future water management plans (Middelkoop et al., 2001), so we focus our analysis on changes in snow water equivalent (hereafter SWE). As the climate warms, one might anticipate a decrease in SWE as it becomes too warm for the surface to support seasonal snowpack for a greater portion of the year. The influence of changing precipitation patterns on SWE also needs to be considered, where there may be increase in precipitation (and snowfall) that can increase the SWE in a warming climate. Due to the complexity of these drivers of future SWE, it is important to thoroughly evaluate the causes of previous trends in SWE and model predictions of future SWE changes with global warming. Long-term snow-monitoring stations have reported widespread snow declines throughout the western United States, including maritime climates such as the Pacific Northwest and California (Chavarria & Gutzler, 2018; Clow, 2010; Fritze et al., 2011; Harpold et al., 2012; Knowles et al., 2006; Mote, 2003), however the observed changes in the intercontinental regions such as the Mountain West are less consistent compared with maritime regions (Harpold et al., 2012; Regonda et al., 2005). In this study we use a combination of in situ and satellite observations, reanalysis, and state-of-the-art global climate model (hereafter GCM) simulations to determine if simulations of historical SWE in the Mountain West region are consistent with observations. We then use a machine learning model trained on historical reanalysis to determine which aspects of the GCMs contribute most to their bias and spread in SWE predictions.Throughout this paper we use the relationship between SWE in the Mountain West and the phase of the El Niño Southern Oscillation (ENSO) to benchmark if models are correctly simulating the relationship between SWE in the Mountain West and large-scale climate variability. ENSO is an interannual oscillation of sea-surface temperature in the equatorial eastern pacific and is the largest mode of interannual variability in the historical sea-surface temperature record (Battisti et al., 2019; Chen & Wallace, 2015). It has numerous impacts on atmospheric circulation (Bayr et al., 2014; Bjerknes, 1969; Burgers & van Oldenborgh, 2003; Oort & Yienger, 1996; Wang & Su, 2015) and some studies have linked weather patterns and hydrology over the continental United States to the phase of ENSO (Hidalgo & Dracup, 2003; Singh et al., 2021). There have also been a handful of studies that have shown that the climate’s response to ENSO may be a useful proxy for understanding the climate’s response to global warming (Klein & Hall, 2015; Norris et al., 2022). We expect that models which can simulate the response in the Mountain West to ENSO with high fidelity are likely most skillful at simulating future climate change within that region, so we evaluate the efficacy of our models’ predictions of SWE in the Mountain West based on if they correctly predict the relationship with ENSO. After which, we use the models to evaluate the response to past warming, and what we might expect from future warming. We note that the analysis used in this study could easily be applied to other regions too, so we hope this paper lays the framework for future assessments of models’ ability to forecast regional SWE changes.This paper is organized as follows, in section 2 we describe the data used in this study, including observational datasets, reanalysis, and GCM output. Then section 3 describes a decomposition to determine how three mechanisms contribute to the SWE change and the machine learning framework that is described and validated. The main results of this study are discussed in section 4, followed by concluding remarks in section 5.2. Dataa. Observational DatasetsIn this paper, we use a combination of satellite estimates of precipitation and ground-based measurements of SWE to evaluate if the reanalysis and GCM response to climate variability is consistent with observations.Precipitation estimates from the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG; Huffman et al., 2020) product serve as our precipitation observations over the Mountain West. IMERG incorporates, merges, and inter-calibrates various spaceborne infrared (IR) and microwave retrievals, which are then calibrated by ground-based rain gauges. Most of the spaceborne observations were either used in the Tropical Rainfall Measuring Mission Multi-Scale Precipitation Analysis (TMPA; Huffman et al., 2007), or were launched in the follow-on mission, the Global Precipitation Measurement (GPM) constellation, which was meant to improve the accuracy and resolution of TMPA and expand its measurement region from 35° N-S to 65° N-S (Hou et al., 2014). We use the sixth version (V06) of the IMERG dataset, which includes substantial updates to the morphing system that relies on a Lagrangian approach to propagate precipitation vectors into regions without instantaneous precipitation observations (Tan et al., 2019). The newest V06 dataset allows for precipitation to be estimated globally (year 2000 – present) at 0.1°x0.1° resolution, and when compared to surface observations, V06 performs considerably better than previous versions (Pradhan et al., 2022). Though IMERG continues to advance, it still exhibits biases including an underestimation of precipitation over mountainous regions (Anjum et al., 2018; Asong et al., 2017; Dezfuli et al., 2017; Huang et al., 2018; Kim et al., 2017; Sharifi et al., 2016). Due to this known bias, we do not compare the precipitation averages in GCMs and reanalysis directly against the precipitation average measured by IMERG. Instead, we evaluate if the GCMs and reanalysis correctly simulate the relationship between ENSO and precipitation variability in the Mountain West relative to IMERG. In their review of IMERG performance Pradhan et al. (2022) showed that over mountainous terrain there is considerable bias in the mean-precipitation, but with sufficient aggregation (monthly or greater) it can capture ground-based observed variability.In this study we are interested in seasonal snowpack, and so to evaluate the reanalysis and GCMs predictions of SWE we use the SNOw TELemetry (SNOTEL) network of snowpack related climate sensors operated by the Natural Resources Conservation Service of the United States Department of Agriculture that are situated throughout the mountainous terrain of the Western United States. Some of their earliest uses in research was by McGinnis (1997), who estimated the impact of climate change on Colorado snowpack with a machine-learning model trained on the SNOTEL network. The SNOTEL network has since been used throughout the scientific literature as the measure of snowpack in North America (e.g., Fassnacht et al., 2003; Molotch & Bales, 2006; Schneider & Molotch, 2016; Serreze et al., 1999). Using ground-based observations to estimate changes throughout an entire region has some inherent problems; namely that there is typically not sufficient sampling over the entire region to accurately measure the average SWE level in the entire Mountain West. For this reason, like our usage of IMERG, we focus on the SNOTEL variability as opposed to its average value during any given period. SNOTEL observations are also subject to biases related to uneven sampling. Specifically, the SNOTEL locations are not evenly spaced, and do not sample all aspects, slope angles, or regions evenly, so an average over all SNOTEL sites would over-weight the regions that are well sampled and under-weight those with poor sampling. Hence, we do not average the SNOTEL sites over the Mountain West, and instead look at them on a station-by-station basis. Additionally, since the start of the SNOTEL project numerous sites have been added and some decommissioned, so we limit our scope to the SNOTEL sites within the Mountain West that have been consistently providing data from 1980 to 2015.b. ReanalysisOften, observations cannot be directly compared with GCM output due to a combination of the aforementioned observational biases and their relatively short observational periods in comparison to the timescale of climate change (decades/centuries). Hence, we use the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5; Hersbach et al., 2020), as a comparison point to determine the efficacy of GCM predictions of the previous century. The ERA5 dataset is a global climate model that assimilates roughly 24 million observations per day from a variety of satellite instruments, and in situ observations of 10 m winds over sea, 2m humidity over land, surface pressure, measurement of upper-air temperature, humidity and wind from dropsondes, radiosondes, and aircraft measurement, and rain rate from ground-based radar-gauge composite observations (Hersbach et al., 2020). All variables are assimilated together to yield a simulated atmosphere that is as consistent as possible with that of the real world. Later in this manuscript we show that ERA5 precipitation and SWE in the Mountain West are rather consistent with observations, enabling us to perform direct comparisons between GCM output and reanalysis to determine the deficiencies in GCMs and their potential causes.c. Global Climate ModelsTo evaluate our current ability to simulate SWE in the Mountain West we also use GCMs that are not constrained by observations (as in ERA5). We primarily use output from two types of simulations performed as a part of the sixth phase of the Coupled Model Intercomparison Project (CMIP6; Eyring et al., 2016). We use the historical simulations, which are fully coupled model simulations forced by the historical levels of atmospheric constituents from the year 1850 to 2014 and were performed as a part of the CMIP6 DECK experiments (Eyring et al., 2016). We also use output from the Scenario Model Intercomparison Project (ScenarioMIP), which provides multi-model projections of future climate based on possible future emissions and land-use changes produced with integrated assessment models (O’Neill et al., 2016). The full list of CMIP6 models that we use is provided in the Supplementary Materials .In addition to the CMIP6 models, we also include the latest model developed by the NASA (National Aeronautics and Space Administration) Goddard Institute for Space Science (NASA-GISS), Model E version 3 (hereafter GISS-E3), which has not yet been published on the CMIP database. This model provides unique insight into the drivers of model spread because NASA-GISS has produced three equally viable versions of their climate model, each characterized by different parameter tunings, but with emergent mean climatological states that broadly match satellite observations. Later in this paper we highlight the results of these three versions of the GISS-E3 model to indicate how much model spread could be a result of parametric uncertainty as opposed to structural errors.3. Methodsa. SWE DecompositionThroughout this study we decompose changes in SWE (either due to variability or climate change) into the portion of the anomaly due to changes in precipitation rate, the fraction of precipitation which occurs as snow, and the fraction of the previous snowfall that remains in the snowpack (and has not melted, evaporated, or blown away). We apply this decomposition to changes in SWE driven by internal variability (ENSO) or climate change. The decomposition method used here was initially introduced by Räisänen (2008), where they show that SWE of seasonal snowpacks can be expressed as a product of the fraction of accumulated snowfall that remains on the ground (\(G\)), and the amount of accumulated snowfall, which is the product of the precipitation (\(P\)) and fraction of precipitation that occurs as snow (\(F\)) integrated from sometime before the start of the snowy season (in our case we use August) to a selected time during the snowy season, as expressed in Equation 1.\begin{equation} \begin{matrix}SWE=G\int_{\text{Aug}}^{t}{\text{FPd}t^{{}^{\prime}}}\ \#\left(1\right)\\ \end{matrix}\nonumber \\ \end{equation}Then in Equation 2 the SWE anomaly is decomposed into the portion of the anomaly due to precipitation anomaly (first term on RHS of Equation 2), the fraction of the precipitation to occur as snow anomaly (second term on RHS of Equation 2), the fraction of the accumulated snowfall to remain on the ground at time t (third term on RHS of Equation 2), and the non-linear combinations of \(G\), \(F\), and \(P\) which is typically at least one order of magnitude less than the other terms (Räisänen, 2008, 2023).\begin{equation} \begin{matrix}\Delta SWE=\overline{G}\int{\overline{F}\Delta}Pdt^{\prime}+\overline{G}\int\Delta F\overline{P}dt^{\prime}+\Delta G\int{\overline{F}\ \overline{P}dt^{{}^{\prime}}}+\frac{1}{4}\Delta G\int{\Delta F\Delta\text{Pd}t^{{}^{\prime}}\ }\ \#(2)\\ \end{matrix}\nonumber \\ \end{equation}In Equation 2 the overbar denotes an average over the entire time domain and \(\Delta\) denotes the anomaly in some subset of the time domain (either a specific set of years or composites of El Niño and La Niña years). Throughout this paper, Equation 2 is applied to each month of the year individually, meaning that \(\Delta\) denotes the anomaly of a specific month in a specific year (or average over some subset of years) from the monthly average over all years within the domain. This allows us to show how global warming and climate variability impact each term in Equation 2 at months of the year. Unfortunately, we cannot perform this calculation with observational datasets because we do not have any observational retrievals of \(F\).b. Gaussian Process EmulationWe use machine learning to determine if deficiencies in GCMs’ SWE and their inter-model spread are due to their simulation of surface temperature and moisture convergence, or if they are due to the numerous small-scale processes that can impact SWE, such as aerosol snowpack effects, evaporation and sublimation rates, and runoff. We use Gaussian Process Emulation (hereafter GPE), a machine-learning algorithm that has robust uncertainty estimates and performs particularly well in cases with relatively small amounts of training data (Watson-Parris et al., 2021a). To train and use our GPE models we use the Earth System Emulator (ESEm) software package, which is designed for use with geospatial data, and allows for complex calibration and emulation to be done in a simple analysis framework (Watson-Parris et al., 2021b). Using ESEm, we train a GPE model to predict March (the typical start of the melting season) SWE and its components from Equation 2 using only surface temperature and moisture convergence from October to March as the predictor variables. We train the model on ERA5 output from the year 1940 to 2015, where a subset of 20 randomly selected years is removed from the training data to test the efficacy of the GPE model.We argue that \(P-E\) is an acceptable approximation for moisture convergence. Moisture convergence is not explicitly archived for GCMs participating in CMIP6, so we approximate moisture convergence as\(P-E\), which does neglect changes in column water vapor; however, changes in column water vapor are typically small compared to \(P-E\). We validate our use of \(P-E\) with ERA5 (which does directly save output of the true moisture convergence). In the Supplementary Materials we show that the seasonal cycle and ENSO variability of\(P-E\) and moisture convergence in ERA5 are quite similar, and in fact, we find that from 1940 to 2015 the mean values of ERA5 moisture convergence and \(P-E\) have a correlation coefficient of 0.95 and during Autumn, Winter and Spring are statistically indistinguishable at\(2\sigma\) confidence.GPE requires selection of a kernel, which can be thought of as a “best guess” of the functional form of the relationship between the input predictors and the output predictand. We select our kernel by looping through all possible additive combinations of the kernels available in ESEm and choosing that for each component of (and total) SWE where the relationship between the predicted SWE and the ERA5 SWE for the testing dataset explains more than 50% of the SWE variance (r2 greater than 0.5), where the slope of the linear relationship is closest to 1, and where a sufficient number of the GPE predictions for the testing data cross the 1:1 line.

Biomass-based marine CO2 removal (mCDR) aims to harness photosynthetic organisms to remove excess CO2 from the atmosphere and sequester that fixed carbon in a long-lived marine reservoir. This strategy would contribute to a portfolio of climate mitigation efforts. To guide decision-making around testing, deploying, and regulating mCDR, we need to better understand how the deep sea and the broader Earth system would respond to increased biomass addition. The central processes driving this response are sensitive to choices about biomass type and storage site, and they stretch across spatial and temporal scales from microns to kilometers and from minutes to millenia. To organize this immense interdisciplinary challenge, we define five generalizable phases of a biomass-based mCDR project: inputs, placement, short-term response, long-term response, and functional stability. Each phase is associated with high-priority research objectives that could be achieved through thoughtful integration of direct field measurements, investigations of analog sites, experiments, and/or models. In-situ and laboratory experiments can be particularly powerful for isolating key processes; for example, in-situ “closed-system” bottle incubations can amplify small signals and reduce uncertainties created by complex physical flows. Regardless of approach, the overarching goal of biomass-based mCDR research is to develop a process-based understanding of biomass sequestration that is robust enough to project the likely outcomes of alternative choices related to mCDR. Beyond assessing carbon storage and ensuring regulatory compliance, future field experiments should prioritize generating the data required to improve models for impacts of biomass-based mCDR on the deep ocean at climatically-relevant scales.

The present-day Martian crust is likely to be dominated by the basaltic secondary crust formed via the extrusion of basaltic magma from the partially molten Martian mantle. However, it is still unknown if the Mars possessed other crustal types. In this work, we investigate the growth of the Martian crust using the 1-D parameterized mantle convection model. In order to account for the crustal thickness, crustal Th abundance and mantle potential temperature at the present-day, a silica-rich primitive crust thicker than 6-10 km must be formed before 4.5 Ga. During consequential evolution, the primitive crust of the northern hemisphere would be destroyed by the Borealis impact at ∼ 4.48 Ga, but the primitive crust of the southern hemisphere still survives and is buried below the ejecta of the Borealis impact and the basaltic secondary crust. Besides, the water content of the Martian mantle depends strongly on the predominant creep style therein. The dominance of diffusion creep corresponds to a Martian mantle drier than the terrestrial mantle, whereas the dominance of dislocation creep in the Martian mantle may allow an Earth-like water content in the Martian mantle.

Accurate modeling of thermospheric impact of high solar and geomagnetic activities is crucial for safeguarding our space-based infrastructure. However, current modeling capabilities are still unable to accurately predict thermospheric density, which is a key parameter for calculating satellite drags. In this paper, we performed global sensitivity analysis (GSA) for nitric oxide (NO)-related chemical reaction rates using the Global Ionosphere Thermosphere Model (GITM) during solar maximum and solar minimum conditions. We have performed GSA and uncertainty quantification (UQ) for the first time in GITM. GITM is a computationally expensive model; therefore, we employed a Gaussian process (GP)-based surrogate model to approximate the thermospheric states of GITM and inexpensively generate samples for Monte-Carlo-based Sobol analysis. We computed first-order (main effect) and total-order (total effect) Sobol’ sensitivity indices to quantify how the uncertainty associated with NO-related chemical reaction rate coefficients in GITM influences the variance of the NO density, NO cooling rate, temperature, and neutral density. Our study identified the most influential reaction rates the uncertainty of which contribute to the most uncertainty in estimating thermospheric states in GITM and provided important information for UQ within GITM to accurately estimate the thermospheric density. Our findings suggest that reducing the uncertainty in the reaction rates, particularly for RR43 ($NO + hv ightarrow N ({}^4 S) + O$), RR44 ($N({}^4 S) + O_2 ightarrow NO + O$), and RR5 ($N_2^+ + O ightarrow NO^+ ({}^2 D) + N ({}^2 D)$), should be prioritized to fix GITM’s response to variations in F10.7 solar flux.

Fast emulators of comprehensive climate models are used to explore the impact of anthropogenic emissions in future climate. A new approach to emulators is introduced that predicts distributions of coarse-grained monthly averaged variables as a multivariate Gaussian distribution. The emulator is trained with a state-of-the-art climate model and serves as a good first-order representation for many statistics of future climates. The emulator is applied to statistics of surface temperature and relative humidity for illustrative purposes, but the approach can be applied to any other variable of interest as long the multivariate Gaussian approximation captures the bulk of the distribution. Importantly the emulator accounts for the internal variability of the system, allowing one to examine shifts in distributions of climate variables. In this sense the work can be considered as an extension of pattern scaling emulators that focus on the evolution of the mean rather than the distribution of climate variables.

Lunar chronology functions are calibrated from Apollo samples that have a well-defined radiometric age and are correlated to a region with a known crater density. Different chronology functions have been constructed based on different crater counts and different curve fits to crater-counting data. The Neukum Production Function (NPF) is the most commonly used chronology function for the Moon, but there also exists the Robbins Production Function (RPF). The RPF is notable for featuring a much more rapid drop off in impactor flux during the Imbrian period. Using an impact bombardment model, we modeled the distribution of impact melt at the Apollo sites for the Imbrian period. Large sub-basin craters like the 250 km Iridum sub-basin may be the source of some melts at Apollo sites, but it is very difficult to determine specific craters of origin for impact melts from this time period. The distribution of Apollo melt ages is more compatible with the steep decline in impactor flux during the Imbrian characterizing the RPF than the more gradual decline of the NPF.

Kinematic conditions at the water table are commonly used in analytical models to interpret aquifer pumping test data. However, the existing kinematic conditions often lead to an underestimation of specific yield during the drainage process due to the assumption of instantaneous gravity drainage, which does not align with estimates from soil column experiments. This study derives a new water table condition on the basis of a full unsaturated-saturated flow model. This new condition not only integrates the hydraulic properties of the unsaturated zone that explicitly provides a comprehensive insight into the physical implications, but intriguingly retains mathematical tractability. This new condition is applied to a groundwater flow model for a pumping problem, and the drawdown solution is derived via integral transformations. The proposed solution is unitized to analyze the field data from a series of pumping test. The results demonstrate that the proposed model provides a better fit to measured data, particularly in scenarios where an unsaturated flow exerts a significant effect. Overall, the newly derived water table condition can be a handy tool to characterize the aquifer properties in an unconfined aquifer.

Low-level mixed-phase clouds occur frequently and persistently in the central Arctic, and thus play a key role in climate feedback mechanisms, air mass transformations, and sea ice melt. Turbulent entrainment at cloud top driven by radiative cooling modulates these clouds by affecting the boundary-layer heat budget. However, reliable measurements of this small-scale process are scarce. This study presents new insights into entrainment in radiatively-driven cloudy mixed-layers at high latitudes based on a library of daily large-eddy simulations covering the full MOSAiC drift. The simulations are based on measurements, cover a periodic and homogeneously-forced small domain representing conditions observed at the Polarstern Research Vessel, and resolve Arctic turbulence and clouds to a high degree. Approximately 1 out of 3 simulated days contain cloud mass in liquid phase. Drift-average heat budget analyses show that the full column is roughly in radiative-advective equilibrium. In contrast, the cloudy boundary layer is not, being dominated by cloud top cooling. Warming by top-entrainment only partially counters this cooling, at efficiencies of about 30 %. Such efficiencies are much lower compared to previous findings for subtropical warm marine stratocumulus. Interestingly, a few outlying MOSAiC cases show similarly high efficiencies. Analysis of turbulence energetics reveals that high entrainment efficiency can be achieved in two ways: surface coupling and strong local wind shear. The former explains the high efficiencies in the subtropics, while the latter drives the highest efficiencies encountered during MOSAiC. In general, these findings emphasize the important role played by wind shear in boosting entrainment efficiency.

Monitoring and detecting marine volcanic activities are key for scientific understanding and disaster prevention. However, this is difficult because they are hidden under water. Near Torishima Island in the Izu Islands, Japan, intensified seismic activity was observed during October 2023, including a mysterious tsunami-triggering earthquake on October 8 (UTC), which was considered to be linked to a volcanic activity. On October 20, 2023, aerial surveys confirmed an 80-km stretch of floating pumice near Torishima Island. This study conducted a Lagrangian back-tracking drift simulation using the ocean current data and surface wind data to trace the origin of the pumice while clarifying the theoretical basis of Lagrangian back-tracking from the Bayesian perspective. Results indicate that the pumice drifted southward from around extensional back-arc basins near Myojinsho Reef and Sumisujima Island approximately 3–5 days before its discovery. These findings are consistent with independent observations such as biological traces and the geochemical characteristics of sampled floating pumice, which is considered identical to that found on October 20 by an airplane. This indicates the presence of unknown volcanic activity around back-arc basins west of the major active volcanic zone. This study demonstrates the utility of combining drift simulations with geochemical and biological data to identify the sources of marine volcanic events, particularly in regions where direct observations are limited. The results of this study contribute to our understanding of volcanic mechanisms and their potential hazards.

Great volumes of water are carried downward into the mantle transition zone (MTZ, 410-670 km depth) by subducting slabs. If this water is later drawn upward, the resulting mantle melting may generate intraplate volcanism (IPV). Despite its importance, the amount and spatial distribution of water within the MTZ, and its impact on IPV, are poorly constrained. Here we use a series of plate tectonic reconstructions to estimate rates and positions of water injection into the MTZ by subducted slabs during the past 400 Myr. This allows us to construct maps of heterogeneous MTZ hydration, which we then compare to IPV locations since 200 Ma. We find a statistically significant correlation between wet regions of the MTZ and locations of IPV at the surface, but only if water remains stored in the MTZ for periods of 30-100 Myr after being carried there by slabs. We find that 42-68% of IPV is underlain by wet MTZ, with higher correlations associated with longer MTZ residence time, slower slab sinking rates, and longer time periods between MTZ hydration and IPV eruption. The correlation is highest during the Jurassic, when more extensive slab interaction with the MTZ caused a wider area of the MTZ to become hydrated. Parts of the MTZ near the western Pacific, southern Africa, and western Europe, have remained dry by avoiding wet slabs. Hydrous upwellings rising from the MTZ, some driven by interactions with subducting slabs, may be responsible for IPV rising from wet MTZ regions.

This study presents time variability of metastable orthohelium, He(23S), in polar region associated with a geomagnetic storm for the first time. Continuous 17-nights dataset of He(23S) airglow brightness at 1083 nm was obtained from a short-wavelength infrared imaging spectrograph (NIRAS-2) at the KHO, Svalbard (78.1°N, 16.0°E). The observed He(23S) airglow brightness clearly displayed responses to the geomagnetic storm in different time scales. The He(23S) airglow brightness has began to decrease sharply within an hour of sudden commencement of the storm, and it was gradually decreased over the next few days and then recovered slowly. It agreed to helium density variations at 500-km altitude calculated by MSIS. The depletion of He(23S) was mainly caused by enhanced Penning ionization due to upwelling N2 from the lower atmosphere; this was consistent with decreased O/N2 ratio in MSIS and TIMED/GUVI measurements, and electron density depletion in F region observed by EISCAT Svalbard Radar (ESR). Additionally, sudden increases in He(23S) airglow brightness were clearly found associated with intermittent electron precipitations observed by the ESR. Therefore, direct impact by precipitating electron injected from the space to the polar upper atmosphere can play significant roles in production of He(23S). The NIRAS-2 measurements have successfully demonstrated that column density of He(23S) from the upper thermosphere to the exosphere was drastically changed by forcing both from the lower atmosphere and from the space. He(23S) measurements will definitely improve our understanding of thermosphere-ionosphere coupling system and extend the coverage of space weather forecasting up to the exobase.

Leveraging the novel kilometer-scale global storm-resolving model GFDL X-SHiELD, we investigate the impact of a 4 K increase in sea surface temperatures on Northern Hemisphere midlatitude cyclones, over a three-year-long period. X-SHiELD simulations show a poleward shift in midlatitude cyclone tracks under +4 K warming, consistent with CMIP model projections. Thanks to X-SHiELD’s high resolution and explicit deep convection, we provide a detailed analysis of the warm and cold sectors of midlatitude cyclones, which are typically underrepresented in coarser CMIP models. Compositing the 100 most intense midlatitude cyclones in the North Atlantic, we find that the warm sector exhibits wind speed and precipitation increases of up to 15% and 20% per degree of warming, respectively, while changes in the cold sector are less pronounced. This study demonstrates X-SHiELD’s potential to provide a realistic-looking picture into the evolving risks posed by midlatitude cyclones in a warmer world.

Training data: 1.6 million 2x8x8 patches of SAR data randomly sampled from 105 bursts of OPERA RTC-S1; 30m resolution, with two channels: VV (Vertical transmit and Vertical receive) and VH (Vertical transmit and Horizontal receive). Validation Sites: On May 24, 2024, a large landslide in Papua New Guinea buried a village and killed between 650 and 2,000 people. In February 2024, a series of wildfires broke out in the Valparaiso region of Chile. At least 137 people were killed and over 23,000 hectares were burned.

A document by Itai Zilberstein. Click on the document to view its contents.

• On a weekly timescale the river discharge is the best predictor for salt intrusion length. • Variations of salt intrusion length on a timescale of a few days are driven by coastal setup. • Treating distributary side branches separately is crucial to incorporate the importance of downstream salinity variability.

Subglacial drainage models represent water flow at the ice--bed interface through coupled distributed and channelized systems to determine water pressure, discharge and drainage system geometry. While they are used to understand processes such as the relationship between surface melt and ice flow, the combination of the number of uncertain model parameters and their computational cost makes it difficult to adequately explore the high-dimensional parameter space and construct robust model predictions. Here, we develop Gaussian Process (GP) emulators that make fast predictions accompanied by uncertainty of subglacial drainage model outputs. Using a truncated principal component basis representation, we construct a GP emulator for daily representation of subglacial water pressure. We also explore emulation of scalar variables describing drainage efficiency and configuration. We train the emulators using ensembles of up to 512 simulations varying eight parameters of the Glacier Drainage System (GlaDS) model on a synthetic domain intended to represent an ice-sheet margin. The emulators make predictions ~1000 times faster than GlaDS simulations, with error <3% for the water pressure field and ~5--9% for drainage efficiency and configuration. We apply the emulators to explore the eight-dimensional input space by computing variance-based parameter-sensitivity indices, finding that three parameters (ice-flow coefficient, bed bump aspect ratio and the subglacial cavity system conductivity) explain 90% of the water pressure variance. The GP emulator approach described here is well-suited to integrate observational data with models to make calibrated, credible predictions of subglacial drainage.