Most state-of-the-art deep learning systems have their roots in computer vision, which force the remote sensing community to develop ad-hoc procedures for applying deep learning methods in the analysis of remote sensing data. In this Juypuiter notebook, we present Keras Spatial (https://pypi.org/project/keras-spatial), a new python package for pre-processing and augmenting geospatial data for deep learning models. Keras Spatial is composed of loosely-coupled components, which allow users to pre-process geospatial raster data on-the-fly before ingesting them into neural networks. The advantage of using Keras Spatial over more traditional Ad-hoc pipelines are (1) allowing scientists and developers to work in projected coordinates rather than pixels and (2) controlling the sample space and hence avoiding issues such as bias and class imbalance during training. We will demonstrate Keras Spatial using the case study of processing digital elevation data for a segmentation model. We will also demonstrate advanced data pre-processing features of this package, such as accessing remote data sources directly, easy integration of multiple datasets using automatic reprojection and resampling, and decoupling training samples dimensions from the geographic extent to open the door for prediction across different scales.

The Bransfield Strait is a semi enclosed region located between the Antarctic Peninsula and the South Shetland Islands (SSI), where the Bransfield Current (BC) is a relatively warm baroclinic jet (0-250 m) flowing northeastward along the southern slope of the SSI. Recent studies have shown that the BC propagates as a buoyant gravity current, recirculating around the north-eastern tip of the islands while shedding an anticyclonic eddy. However, most previous works are based on summertime measurements and a more comprehensive spatio-temporal view of the regional circulation is still lacking. In this study we provide the first seasonal description of the BC based on an extensive dataset of direct velocity measurements. These measurements were routinely collected along ship tracks from 275 cruises between 1999 and 2014. Seasonally, the horizontal structure of the BC flowing as a coastal jet appears more prominent during spring and summer, when its pathway extends along the island slopes over 180 km at velocities up to 45 cm/s, from Deception Island towards King George Island. During these seasons, the BC is consistently narrower at the beginning of its path (15-20 km wide), and wider before starting its recirculation around the SSI (30 km wide). Through autumn, this pattern is also apparent but less intense, especially south off Livingston-Robert Islands, where subsurface velocities (150 m) are lower at about 10-15 cm/s. The scarcity of winter data confirms the existence of a relatively strong coastal jet with mean velocities about 34 cm/s at least south off Nelson Island. Vertically, the BC also displays a distinctive pattern. At the wake of the islands, the core of the current flows hugging the island slopes with velocities about 35 cm/s at 150 m. Differently, when located at channels between islands, the core of the current migrates offshore flowing at 30-45 cm/s. In terms of zonal volume transport (0-250 m), our estimates indicate that the BC transports about 0.85, 0.93, 0.90 and 1.01 Sv from spring to winter, respectively, at the location where all seasons are covered by data (south off Nelson Island). These results support the BC is a recurrent feature of the circulation in Bransfield Strait, flowing northeastward all along the southern slope of the SSI not only during summer but also during spring and autumn.

Representation of canopy and tree function in earth system models (ESMs) is going through rapid advancements in the last decade. The simplistic big-leaf representation of canopy in model is being expanded and replaced by more complex canopy representations that include multiple functional types, species and age/size stages, multiple patches with different canopy height and leaf area characteristics, and vertically detailed representations with between 2 (light and shade) to n canopy leaf layers. Recently, the hydrodynamic approach to modeling stomata conductance has advanced, and many ESMs include a hydrodynamic version. Under the hydrodynamic approach, stomata respond to water availability in the xylem, and not directly in the soil. The result is a model that include the stem, and xylem dimensions as canopy characteristics for determining the flow and storage of water in the xylem. However, pipe diameter, area and volume do not scale linearly, and thus, the simulations cannot resolve canopy-scale hydrodynamics by pulling all the stem conductive area and volume to a single virtual patch scale, but instead, need to consider the hydrodynamics of a single virtual tree and scale its resulting fluxes to the entire canopy. While that is trivial in a homogeneous canopy, the difference between the tree-level description of the hydrodynamic canopy and the horizontally pulled description of the canopy in the surface-flux and radiation-exchange modules leads to a conundrum. Imagine a mixed canopy: 50% oaks and 50% maples. Assume oaks have 20% more LAI and are 1 m taller than the maples. As far as hydrodynamics are concerned, the evaporative demands of maple follow from light attenuation and leaf area profiles of a maple-like canopy. However the roughness length and aerodynamic resistance, and the resulting wind profile inside the canopy is not characteristic of maple or oak but is a result of the mixed canopy. I will present formulation for a consistent scaling of tree-level canopies to patch-level for mixed forests with multiple species of different functional types. The formulation scales canopy characteristics and the resulting canopy fluxes from tree to forest in an energy and mass-conservative way, and allow a smooth and consistent multi-species canopy description for hydrodynamic models in mixed-forests.

A document by Fatemeh Karandish. Click on the document to view its contents.

While many studies comparing atmospheric reanalysis and surface observations have focused on the similarity of mean fields, trends, or frequencies of extreme events, very few have assessed how similar surface observations and reanalysis data sets are in terms of their specific identification of extreme temperature event days. Here, we assess the similarity between surface observations and three reanalysis products: ERA5, ERA5-LAND, and NARR, in terms of the days on which they identify extreme temperature events. We assess similarity from 1979-2016 for 231 locations in the United States and Canada, assessing Extreme Heat and Cold Event days, as well as their counterpart events that are relative for the time of year. Cold Events have a greater match than Heat Events. ERA5 has the greatest match percentage with station data across the study region. Match percentage is greatest in mid-latitude, continental locations, with poorer performance in coastal areas, and the Arctic.

Floods are among the most destructive natural hazards on Earth. In paleohydrology, sediments are generally considered as one of the best archives to extend flood records to pre-historical timescales. Doing so requires being able to identify flood deposits from sediment archives and decipher between flood types. The latter is particularly important in glacierized regions, where meteorological floods frequently co-occur with Glacial Lake Outburst Floods (GLOFs). In Patagonia, GLOFs are particularly pronounced in the Baker River watershed (48°S), where 23 events occurred between 2008 and 2020. Since 1976, the same region experienced three intense rain-on-snow events. To identify the sedimentary signature of these flood events, ten sediment cores collected in the fjord immediately downstream of the Baker River (Martínez Channel) were investigated and compared to the recent flood history of the river. Results show that sediments accumulate on the fjord head delta at 2.0 to 3.4 cm yr-1 and that GLOF deposits can be distinguished from background sediments by their finer grain size (5.98 ± 0.82 μm) and lower organic carbon content (0.31 ± 0.06%), reflecting the release and transport in suspension of high amounts of glacial rock flour. Our results also show that the rain-on-snow events that occur in summer, and therefore primarily affect the glacierized part of the watershed, have the same sedimentary signature as GLOFs. In contrast, rain-on-snow events occurring in winter have a distinct coarse and organic-rich signature, reflecting sediment input from the non-glacierized part of the watershed. In summary, this study shows that (a) GLOF deposits in fjord sediments are distinct from typical flood turbidites and are best identified by their low grain size and total organic carbon content, and (b) the sedimentary signature of rain-on-snow floods in partially glacierized watersheds depends on the season during which they occur. We anticipate that our findings will contribute to a better interpretation of flood records from partially glacierized watersheds.

The Filchner-Ronne Ice Shelf (FRIS) is characterized by moderate basal melt rates due to the near-freezing waters that dominate the wide southern Weddell Sea continental shelf. We revisited the region in austral summer 2018 with detailed hydrographic and noble gas surveys along FRIS. The FRIS front was characterized by High Salinity Shelf Water (HSSW) in Ronne Depression, Ice Shelf Water (ISW) on its eastern flank and an inflow of modified Warm Deep Water (mWDW) entering through Central Trough. Filchner Trough was dominated by Ronne HSSW-sourced ISW, likely forced by a recently intensified circulation beneath FRIS due to enhanced sea ice production in the Ronne polynya since 2015. Glacial meltwater fractions and tracer-based water mass dating indicate two separate ISW outflow cores, one hugging the Berkner slope after a two-year travel time, and the other located in the central Filchner Trough following a ~six year-long transit through the FRIS cavity. Historical measurements indicate the presence of two distinct modes, in which water masses in Filchner Trough were dominated by either Ronne HSSW-derived ISW (Ronne-mode) or more locally-derived Berkner-HSSW (Berkner-mode). While the dominance of these modes has alternated on interannual time scales, ocean densities in Filchner Trough have remained remarkably stable since the first surveys in 1980. Indeed, geostrophic velocities indicated outflowing ISW-cores along the trough’s western flank and onto Berkner Bank, which suggests that Ronne-ISW preconditions Berkner-HSSW production. The negligible density difference between Berkner- and Ronne-mode waters indicates that each contribute cold dense shelf waters to protect FRIS against inflowing mWDW.

Scalar parameters of various aspects of magnetic reconnection offer easily comparable measures of the characteristics of the diffusion regions surrounding X-lines in a way arguably more general than other attributes which may depend on the coordinate system used. We performed a principal component analysis (PCA) of several scalar parameters including measures of relative magnetic field line (MFL) curvature, ion agyrotropy, energy conversion rate, electric field, and current density in the time spans surrounding and including independently identified ion diffusion regions (IDRs) observed by MMS in the geomagnetic tail. We find that relative MFL curvature is the most bold indicator of an IDR encounter while some more traditional measures vary comparatively little between IDRs and the regions adjacent to them when averaged across events. More granular statistics are also presented and discussed.

Observations of the spatio-temporal structure of turbulent mixing in a large, energetic strait were used to examine the interactions between wind- and tidally-forced boundary layers in a coastal environment. Te Moana-o-Raukawa (Cook Strait) of Aotearoa (New Zealand) is a relatively wide, energetic strait, known to experience substantial tidal currents and wind stress. A turbulence-enabled ocean glider mission measured O(40,000) turbulence samples that passed QAQC including the use of a vehicle-mounted speed through water sensor. The observations were compared to one-dimensional models of turbulence to understand the mechanisms that regulates the vertical structure of mixing. Tidal flows of O(1 m/s) and wind speeds of O(10 m/s) enhance dissipation to ε=O(10^{-5} W/kg) through boundary drag, shear-driven production of turbulent kinetic energy (P) and to a minor extent buoyancy flux (G). The benthic and wind-driven boundary layers behaved reasonably predictably when considering a 1D perspective. The interaction between the two boundary layers depended on mid-water column stratification which is to a large degree an externally-prescribed condition. Transient stratification can stabilize the mean flow (median Ri_g=0.6(>1/4)) and reduce both turbulence intensity (Re_b) and diapycnal diffusivity (K_z) by up to two orders of magnitude in the middle of the water column, insulating bottom and surface mixing-layers. Mid-water dissipation rate levels tend to be associated with marginal dynamical stability (median Ri_g=0.22(~1/4)) and canonical mixing efficiency (median R_f=0.17), while elevated levels are connected to unstable mean flow conditions (median Ri_g=0.14(<1/4)) and reduced mixing efficiency (median R_f=0.1(<0.17)) that promotes turbulence growth.

We demonstrate that an extended eddy-diffusivity mass-flux (EDMF) scheme can be used as a unified parameterization of subgrid-scale turbulence and convection across a range of dynamical regimes, from dry convective boundary layers, over shallow convection, to deep convection. Central to achieving this unified representation of subgrid-scale motions are entrainment and detrainment closures. We model entrainment and detrainment rates as a combination of turbulent and dynamical processes. Turbulent entrainment/detrainment is represented as downgradient diffusion between plumes and their environment. Dynamical entrainment/detrainment are proportional to a ratio of buoyancy difference and vertical velocity scale, partitioned based on buoyancy sorting approaches and modulated by a function of relative humidity difference in cloud layer to represent buoyancy loss owing to evaporation in mixing. We first evaluate the closures offline against entrainment and detrainment rates diagnosed from large-eddy simulations (LES) in which tracers are used to identify plumes, their turbulent environment, and mass and tracer exchanges between them. The LES are of canonical test cases of a dry convective boundary layer, shallow convection, and deep convection, thus spanning a broad range of regimes. We then compare the LES with the full EDMF scheme, including the new closures, in a single column model (SCM). The results show good agreement between the SCM and LES in quantities that are key for climate models, including thermodynamic profiles, cloud liquid water profiles, and profiles of higher moments of turbulent statistics. The SCM also captures well the diurnal cycle of convection and the onset of precipitation.

Despite being featured as a very attractive geographical, climatic, and cultural region, the Mediterranean Basin is also frequently affected by high-impact weather. The latter may induce catastrophic consequences with detrimental socio-economic and ecological impacts. Mediterranean cyclones are responsible for a wide range of weather-induced natural hazards affecting a large variety of socio-economic activities. Therefore, there is a growing interest in understanding Mediterranean cyclones and their impacts. This concern involves the scientific community, the stakeholders, and the general public. Many questions are still open calling for extensive research and discussion. Among these, the lack of systematic quantification of cyclones' contribution to Mediterranean socio-economic losses, and efficient and effective communication of information among scientists in different disciplines, stakeholders, and the general public, are complex and currently under-addressed challenges. To contribute to future development beyond the current state of expertise, this work reviews existing knowledge regarding the socio-economic impacts of Mediterranean cyclones with a focus on particularly vulnerable sectors in the region. As a result of this effort, knowledge gaps, as well as the vulnerability of the main societal sectors are identified. The overarching goal of gathering knowledge on the socio-economic impacts of Mediterranean cyclones is to inform decision-makers on the state of the art, thus contributing to the creation of resilient communities in the face of climate change. This is even more relevant in the context of the growing destructiveness inflicted by Mediterranean cyclones, despite their probable lower frequency in future scenarios.

Antarctic Bottom Water (AABW) production supplies the deep limb of the global overturning circulation and ventilates the deep ocean. While the Weddell and Ross Seas are recognised as key sites for AABW production, additional sources have been discovered in coastal polynya regions around East Antarctica, Vincennes Bay being the latest. Vincennes Bay, despite encompassing two distinct polynya regions, is considered the weakest source, producing Dense Shelf Water (DSW) only just dense enough to contribute to the lighter density classes of AABW found offshore. Importantly, the network of local glaciers and upstream Totten Ice Shelf system are all reportedly thinning and the freshwater input from such melting is likely to influence water mass structure. Accordingly, Vincennes Bay presents an interesting test case for DSW/AABW sensitivity to climate-driven changes in Antarctic coastal oceanography. Here we provide the first detailed observations of the Vincennes Bay shelf region and surrounds, using CTD data from instrumented elephant seals in late summer/early fall. We find that Vincennes Bay has East Antarctica’s warmest recorded intrusions of modified Circumpolar Deep Water (mCDW), intrusions that both hinder sea-ice production and contribute salt to new DSW formation. Warm mCDW is also observed to be driving basal melt in Vincennes Bay, as seal CTD data provide the first direct observational evidence for inflow of basal melt to this region. As the most marginal of AABW sources, Vincennes Bay is a particularly useful region for assessment of the sensitivity of AABW production to changes in climate.

One possible human response to climate change and other environmental stresses is migration. However, migration is complex, multi-causal phenomenon, and the complexity of human migration poses a challenge for researchers who aim to study the effects of environmental changes on population mobility. This project aims to understand how changing environmental conditions and livelihood opportunities impact migration decisions in coastal Bangladesh. We are developing an original agent-based model (ABM) that combines stylized environmental change dynamics with livelihood to understand how these dynamics impact migration decisions as well as what feedbacks may exist between them. The ABM uses agents to represent households, which consist of individuals, within a single origin community. At each step of the model, the household assesses the expected utility of different livelihood options within the community for each of its members, including agricultural work on the household’s land (if any), seeking paid employment within the community, and pursuing non-agricultural livelihood activities. After assessing opportunities within the community, households will decide whether or not a household member should seek opportunities outside the community. The model imposes stochastic droughts that impact crop yields. We report preliminary results from this model, comparing simulated migration patterns to empirical observations. Future work will incorporate more complex and psychologically realistic decision-making heuristics, as well as diverse destination locations and the possibility of return migration.

The misalignment of the observation and predicted waveforms in regional moment tensor inversion is mainly due to seismic models’ incomplete representation of the Earth’s heterogeneities. Current moment tensor inversion techniques, allowing station-specific time shifts to account for the model error, are computationally expensive. Here, we propose a lightweight method to jointly invert moment-tensor parameters and unknown station-specific time shifts utilizing the modern functionalities in deep learning frameworks. A $L_2^2$ misfit function between predicted synthetic and time-shifted observed seismograms is defined in the spectral domain, which is differentiable to all unknowns. The inverse problem is solved by minimizing the misfit function with a gradient descent algorithm. The method’s feasibility, robustness, and scalability are demonstrated on earthquakes in the Long Valley Caldera, California. This work presents an example of fresh opportunities to apply advanced computational infrastructures developed in deep learning to geophysical problems.

Lacustrine paleotemperature reconstructions are important for characterizing past temperature and hydroclimate change, validating multi-proxy reconstructions, and evaluating global climate models. In particular, lake water temperature is often derived from geochemical proxies–including clumped isotopes (Δ47), oxygen isotopes (δ18O), alkenone lipids (Uk’37), and GDGT compounds (TEX86). However, global climate models, constrained by resolution, computational demand, and cost, are designed to simulate large-scale processes, often at the expense of resolving lakes and simulating lake temperature. Consequently, this limitation complicates the comparison of climate model-simulated variables such as air temperature, with lake water temperature or with other proxy variables (e.g., pollen-derived air temperature), and necessitates use of a transfer function to relate lake temperature to air temperature. Previous work developed transfer functions to translate proxy-derived seasonal lake water temperature to mean annual air temperature using ground-based measurements from 88 lakes. In this study, we introduce new lake-to-air temperature transfer functions (for annual, spring-summer, spring, summer, and warmest month) that incorporate lake surface water temperature, and new variables including latitude and elevation, by analyzing climate reanalysis data and long-term satellite observations of surface temperatures for 1395 modern lakes via regression-based inverse modeling. With the use of multiple regression models and a dataset roughly 10 times larger, the error in predictions of mean annual air temperature is reduced by up to 48% compared to previous work. To demonstrate the impact of the new transfer functions, we reconstruct Pliocene-Pleistocene mean annual air temperature from Δ47-derived lake temperatures and compare estimates to model simulations for the Last Glacial Maximum and mid-Piacenzian warm period. The new transfer functions, with reduced error, should enable more accurate paleotemperature reconstructions from proxy-derived lake water temperature and allow for more comprehensive assessments of climate model skill.

Earth's upper mantle rheology controls lithosphere-asthenosphere coupling and thus surface tectonics. Rock deformation experiments and seismic anisotropy measurements indicate that composite rheology (co-existing diffusion and dislocation creep) occurs in the Earth’s uppermost mantle, potentially affecting convection and surface tectonics. Here, we investigate how the spatio-temporal distribution of dislocation creep in an otherwise diffusion-creep-controlled mantle impacts the planform of convection and the planetary tectonic regime as a function of the lithospheric yield strength in numerical models of mantle convection self-generating plate-like tectonics. The low upper-mantle viscosities caused by zones of substantial dislocation creep produce contrasting effects on surface dynamics. For strong lithosphere (yield strength >35 MPa), the large lithosphere-asthenosphere viscosity contrasts promote stagnant-lid convection. In contrast, the increase of upper mantle convective vigor enhances plate mobility for lithospheric strength <35 MPa. For the here-used model assumptions, composite rheology does not facilitate the onset of plate-like behavior at large lithospheric strength.

One of the unresolved and debated topics in climatic and atmospheric sciences is whether soil moisture (SM) anomalies can trigger extreme precipitation (P) events? Researchers have used coupled atmospheric model experiments, models based on simple water budget equations, and water vapour tracing studies; however, a consensus is lacking. Some studies reported that the excess SM anomalies trigger subsequent P; a few also postulate about a negative feedback loop. In the present study, we used a novel Event Coincidence Analysis to investigate this trigger relationship between SM and P. Using SM and P data from 2004-2020, we identified hotspots of SM-P coupling over India. A statistical significance test (α = 0.10) was carried out to ensure that the observed coincidences are not by chance. On increasing, the temporal window from one day to three-day the extent and severity of the hotspots increased significantly. The highest values of trigger coincidence rates are observed over central India (>70%). Our observed results agree with the widely regarded hypothesis of stronger SM-P coupling in transitional regions between wet and dry climates. The results obtained in our work has vast potential for atmospheric forecast purposes, including flood early warning systems for India.

Simultaneous data from two interferometers separated by 16 km and synchronized within 100 ns was collected for a thunderstorm near Langmuir Lab on October 23, 2018. Analysis via triangulation followed by a least-squares fit to time of arrival across all six antennae produced a three-dimensional interferometer data set (3DINTF). Simultaneous Lightning Mapping Array (LMA) data enabled an independent calculation of 3DINTF accuracy, yielding a median location uncertainty of 200 m. This is the most accurate verified result to date for a two-station interferometer. The 3D data allowed profiling the velocity of multiple dart leaders and K leaders that followed the same channel. 3D velocities calculated from the in-cloud initiation site to ground ranged from 3x10^6 m/s to 20x10^6 m/s. Average velocity generally increased with subsequent leaders, consistent with increased conditioning of the channel. Also, all leaders showed a factor of two to three decrease in velocity as they proceeded over 15 km of channel. We speculate that the velocity decrease is consistent with energy lost in the reionization of the channel at the leader tip. This paper includes an appendix providing details of the triangulation technique used.