The Local Air Emissions Tracking Atlas (LOCAETA) is a tool to quantify and understand the co-benefits of industrial decarbonization. It includes data harmonization, emissions scenario development, and modeling. Its interactive platform was designed to help communities and decision-makers across the US understand the long-term air quality and public health impacts relative to various decarbonization approaches, such as carbon capture, electrification, and hydrogen fuel-switching. In this work, we explore the air dispersion modeling component of LOCAETA using HYSPLIT and InMAP for facility PM2.5 emissions. We present a case study of the Denver, Colorado region and connect “chronic-exposure” plumes from several large industrial emitters to available census and household demographic information, which supplements LOCAETA’s public health modeling and source apportionment results. This work can benefit decision-makers in all levels of government, as well as communities advocating on their own behalf, clean energy investors, and stakeholders interested in decarbonization.

The 2017 GOES-R Post Launch Test (PLT) field campaign was conducted to validate the Advanced Baseline Imager (ABI) and Geostationary Lightning Mapper (GLM) instruments on board the GOES-16 satellite. As part of the campaign, a suite of airborne lightning and cloud measurements were collected from the high-altitude ER-2 aircraft, complemented by ground-based lightning and radar data. Of particular interest to an ongoing historical reanalysis of these data is (a) the characterization of optical measurements of lightning from the airborne Fly’s Eye GLM Simulator (FEGS) using ground-based detection of lightning radio emission from Lightning Mapping Arrays (LMAs) and (b) how these relationships vary with cloud and convective properties.The GOES-R PLT campaign included six ER-2 flights over an LMA, capturing lightning that occurred in multi-mode convection in varying regions across the continental United States. Multi-spectral optical measurements from the FEGS are characterized using the spatial and temporal characteristics of matched lightning flashes reconstructed from LMA detections. These relationships are then interpreted using cloud depth inferred from the airborne Cloud Physics Lidar as well as precipitation type and ice water content retrieved from ground-based polarimetric Doppler radars. These data are used to address how the interpretation of optical lightning data and its portrayal of lightning characteristics vary with cloud and convective context.

Water and solute transport in the vadose and saturated zone of crystalline hard rock formations are driven by complex flow phenomena, and modeling these processes at the hillslope scale is challenging. Additionally, sub-seasonal dynamics in climatic parameters can affect the hydraulic properties (hydraulic conductivity and specific yield/ porosity) of the vadose zone media. This necessitates characterizing the permeability architecture while modeling these aquifers’ unsaturated and saturated flow. The present study attempts to understand the permeability architecture of a crystalline hard rock aquifer system with agriculture as the dominant land use in India’s Eastern Ghat mountain range at the hillslope scale. The hydrogeology was investigated using 70 surface geophysical tests, 20 borehole loggings, groundwater level monitoring, and delineation of rock outcrops in a 27.38 km2 study area. The investigation improved the understanding of stratigraphy, weathered thickness, and particle size distribution in the underlying hard rock formations. Groundwater head simulations indicate that water table fluctuations in the stratiform fractured layer result in a significant fall in the water table, in contrast to fluctuations in the saprolite layer. The permeability architecture derived from this description of aquifer response distinctly illustrates groundwater reserves and their susceptibility to exploitation. The framework of this analysis can be applied to examine the implications of climatic effects on weathered hard rock aquifers with similar hydrogeological settings.

Extreme weather poses significant threats to the reliability of diversified electricity generation portfolios across the United States. For example, ice accumulation disables wind turbines, heavy snow disrupts fuel transportation, floods damage infrastructure, and extreme heat reduces thermal plant efficiency. In this work, we examine the impacts of an extensive list of extreme weather events—including heavy snow, blizzards, ice storms, winter storms, extreme cold/wind chill, hurricanes, tornadoes, floods, droughts, heat waves, and wildfires—on wind, solar, natural gas, coal, and other generation sources covering nearly all power plants in the United States, analyzing impacts at temporal resolutions varying from hourly to monthly depending on data availability. For instance, for wind power plants, we integrated monthly reported generation from 1202 facilities across the nation (2008-2022) with physics-informed modeled generation derived from reanalysis-driven simulations, finding that extreme weather events cause substantial reductions in actual generation relative to model predictions for that month: hurricanes reduce generation by 71%, cold/wind chill events by 14%, blizzards by 15%, and ice storms by 13%, demonstrating the severe operational impacts of extreme weather on wind power output. Our comprehensive assessment across all generation technologies reveals systematic vulnerabilities in our power infrastructure and can support development of proactive planning strategies to enhance grid resilience under future climate conditions.

The El Niño Southern Oscillation (ENSO) is a key driver of global climate variability. Through its teleconnections, ENSO influences precipitation and temperature extremes worldwide, yet its influence on compound extremes across agricultural systems is less well explored. Here, we examine ENSO's influence on co-occurring hot-dry (HD) and hot-wet (HW) events across global croplands, given their distinct risks for food production. HD events intensify crop stress by combining heat and moisture deficits, while HW events can increase risks of pests, pathogens, and post-harvest losses. Using the Niño 3.4 index, high-resolution climate data (1901–2024), and crop extent and calendar data, we quantify the localized risk of these compound heat events and assess exposure for all croplands as well as staple crops of maize, rice, soybeans, and wheat. We show increased risk of both HD and HW events across global croplands during  El Niños. For example, the localized risk of HD events more than doubles across parts of global croplands including India, Australia, the Sahel, Brazil and Mexico during El Niño relative to ENSO-Neutral years, resulting in a significantly elevated global cropland exposure to HD events during developing El Niño summers. Localized risk of HW events also more than doubles particularly across India, Argentina, Brazil, and parts of Sahel with significantly elevated cropland exposure in decaying El Niño summers. Of the four staple crops, rice shows the strongest link to ENSO, with global exposure increasing > 30% and > 40% above Neutral years for HD and HW, respectively. These findings demonstrate that ENSO modifies multi-crop risks through compound events, highlighting the need to integrate ENSO risks for compound heat events into early warning systems and country- and crop-specific adaptation strategies to safeguard food security in vulnerable regions.

Relative to the rest of the soil matrix, soil preferential flow (PF) disproportionately influences large-scale processes like groundwater recharge and solute flux via rapid transport of water through soil pore spaces. Despite the ubiquity of PF occurrence across land uses and climates, there is lack of understanding around the feedbacks between PF generation and carbon transport and transformation, both of which result from abiotic (e.g., water content, clay content) and biotic (e.g., root and microbial activity) processes. We can assess the depth distribution of soil carbon parameters in relation to PF occurrence through soil profiles to decipher the effects of PF generation on carbon dynamics. Recent work from our group shows that the probability of PF occurrence increases with rainfall intensity, net primary production, and clay content. In addition, total rainfall amount increases PF propagation to deeper soil layers. We hypothesize that these same factors increase the transport of carbon deeper into the subsoil and influence carbon transformation at depth. Specifically, we hypothesize that high peak rainfall intensity, net primary production, and clay content will be associated with greater availability of extractable organic carbon relative to total soil organic carbon (SOC) at depth due to rapid transport via PF, high root abundance, and high retention of mineral-associated SOC, respectively. To identify the dominant drivers of PF and the resulting dynamic feedbacks with soil carbon, we synthesized metrics of soil PF, carbon, and enzyme activity from observations at eight U.S. sites spanning climatic gradients, land-cover types, and soil orders. We derived PF metrics using a soil water velocity threshold method based on observations of high frequency soil moisture and precipitation data. PF occurred when the observed soil water flux exceeded the saturated hydraulic conductivity of the soil matrix estimated from a pedotransfer function. Preliminary findings indicate that land-cover is a primary control on extractable DOC availability and propagation with soil depth. PF may also be associated with DOC transformations, given the coupling of land cover and net primary production. Understanding which, and to what extent, abiotic and biotic factors promote generation of PF through feedbacks with soil carbon dynamics advances our knowledge of the potential effects of changing climate and land use on water and carbon fluxes through soils.

Strain localization and partitioning play a critical role in the formation of shear zones and broader tectonic processes. However, our understanding of strain localization is still restricted. Various factors contribute to how strain is distributed in a shear zone, such as differences in lithology or mineralogy, grain size reduction, the presence of weak minerals, and the formation of crystallographic preferred orientations. In this study, we examine how mineral composition and grain size reduction influence strain localization within the Santa Rosa Mylonite Zone. The Santa Rosa and San Jacinto Mountains are the youngest and highest peaks within the Peninsular Ranges, which stretch southward into Baja California. Within this region lies the Santa Rosa Mylonite Zone—the focus of our study due to its intricate geologic history and its mechanical link with the San Andreas Fault system. The zone features a range of rock types, including protoliths such as granodiorite and tonalite, as well as various deformed rocks like mylonites, ultramylonites, marbles, limestones, and other tectonites. The close spatial association of these diverse rocks offers a valuable opportunity to explore how strain is partitioned and localized, and how mineral fabrics evolved from protoliths to their deformed counterparts. This study utilizes a combination of optical microscopy, scanning electron microscopy, electron backscatter diffraction, and X-ray diffraction to investigate microstructural and compositional changes in the rocks, with a particular emphasis on the deformation of plagioclase, biotite, amphibole, and quartz. The results will provide a deeper understanding of how strain is distributed within well-developed shear zones and offer broader insights into the processes driving lithospheric deformation.

Citizen science involves the public in scientific research, often through data collection and analysis using digital tools. It helps democratize science and increase the volume of meaningful environmental data. NASA’s GLOBE Observer app enables users to record land cover, cloud, and mosquito observations for public databases. However, as reported by NASA STEM Enhancement in Earth Sciences (SEES) interns, users of the land tool frequently face the loss of their observation data, issues with location accuracy, and low image resolution, all of which negatively impact classification results. To address these challenges, we are prototyping EarthLens, which is a drone and mobile app-based system designed to enhance citizen science land cover monitoring through real-time data collection and improved user experience. The project combines an open-sourced lightweight drone design, equipped with affordable high-resolution cameras and visible/NIR spectrometers (NASA STELLA devices), and an intuitive mobile application that allows users to upload and visualize their data. Through this, they can access geotagged photos, generate interactive maps, and use AI tools to analyze land cover data. Additionally, the app features a disaster overlay tool that utilizes satellite imagery to display before-and-after views of affected areas. Users may also utilize the app as a social platform where they can post their observations and participate in thematic challenges highlighting traits of their communities. EarthLens explores how advanced yet accessible technology can make citizen science more accurate, collaborative, and engaging. By addressing limitations in current platforms like the GLOBE Observer, this prototype aims to expand the role of everyday observers in environmental research and resilience efforts.

Permafrost degradation linked to rapidly rising air temperatures across the Arctic drives geologic hazards, such as weakened ground bearing capacity and differential ground subsidence. This threatens to disrupt the lives of millions of people living in the Arctic by damaging infrastructure needed to sustain over 1000 high-latitude communities on permafrost. Publicly-available geospatial data, such as OpenStreetMap, used to estimate exposure of infrastructure to hazardous permafrost degradation is limited in geographic coverage across the pan-Arctic. This has led to inaccuracies in the estimated spatial distribution of risk. Using a deep learning model trained to detect buildings from sub-meter resolution Maxar satellite imagery of Arctic communities, we mapped a building area of 52 million m2 (454,947 individual buildings) missing from the OpenStreetMap dataset, expanding the mapped area of buildings by 22% at the pan-Arctic scale. Building area was expanded by 45% in Alaska, 23% in Russia, and 13% in Canada. The highest total area of new buildings was mapped in Russia at 42 million m2 (384,850 buildings). Combining OpenStreetMap with this newly mapped building area suggests that 64% and 52% more pan-Arctic building area is exposed to highly-hazardous permafrost degradation by 2041-2060 under the RCP4.5 and RCP8.5 scenarios, respectively.

Seismic waves originate from dynamic rupture propagation in faults. Seen from afar, faults are analogous to shear cracks, and their rupture can be analyzed using the tools of fracture mechanics. However a closer look reveals that faults can also be considered locally as a tribosystem, i.e. as a layered structure which accomodates deformation thought localized shearing in a thin granular layer of fault gouge. These two scales are equally important but are difficult to handle simultaneously in simulations. In this communication, we propose a novel numerical model where this challenge is adressed. The gouge layer is represented using the Discrete Element Method, where each micrometric gouge grain (about 1 million of them in the present case) is explicitely represented and submitted to Newtonian dynamics, based on the forces it receives from its contacting neighbours. This layer is 2 mm-thick, and is confined between two continuum regions simulated using an explicit Meshfree Method. They receive the elastic properties of country rock, and are prestressed in the normal and tangential directions in order to bring the gouge layer just below its peak strength. The resulting fault system has a total length of 64 cm. A labquake is then triggerred from the central point of the fault, and the weakening rheology of the gouge layer allows it to propagate along two rupture fronts, which exhibit specific properties inherited from the frictional response and structure of the gouge. Inclined Riedel bands spontaneously develop at quasi-periodic intervals in the granular layer, and both rupture fronts propagate by leaps when successively activating slip in these structures. They both transition to a supershear regime after a certain sliding distance. This model allows for the first time to observe the behaviour and response of the gouge layer as it endures the propagation of a rupture front. Localization patterns and granular complexity render the rupture irregular and heterogeneous, but a moving average in time in the frame of the crack tip allows to recover stress concentrations and slip velocity patterns which are consistent with the Linear Elastic Fracture Mechanics predictions. Il allows to relate gouge frictional response and rupture dynamics without the need to prescribe an abritrary friction law or to invoque separation of scales.

New and upcoming satellite missions such as NASA’s 2024 Plankton, Aerosol, Cloud Ocean Ecosystem (PACE) and 2028 Surface Biology and Geology (SBG) missions will produce hyperspectral imagery with great potential for monitoring water quality. To better understand how hyperspectral data represent lake biogeochemical processes and states, the LakeView campaign entailed simultaneous airborne hyperspectral imaging and comprehensive in situ sampling of Lake Mendota in Madison, WI. 14 sampling events in March through November, 2024 captured the limnological conditions of Lake Mendota as it passed through seasonal phases that have distinct water optical characteristics. Such phases are driven by ecosystem functions such as food webs, nutrient cycling, and metabolism. To evaluate relationships between water quality and remote sensing reflectance, a canonical correlation analysis was performed on two groups of variables: (1) the optically active water constituents (dissolved organic carbon, total suspended solids, chlorophyll, and phycocyanin) and (2) visible wavelengths. Clusters of hyperspectral data on canonical variate axes align with known phases of Lake Mendota’s annual cycle: non-cyanobacteria phytoplankton in the spring, the clear water phase, and a cyanobacteria-dominated summer into fall. This airborne campaign helps to uncover the spectral shapes and magnitudes correlated with field measurements of coupled biogeochemical variables over time. Such information should drive us closer to monitoring dynamic lake ecosystems with hyperspectral satellites.

Low-frequency (LF) seismic data are crucial for stable full-waveform inversion (FWI), yet remain difficult to acquire due to source bandwidth limitations and noise. We propose a novel physics-aware conditional diffusion framework to extrapolate LF wavefields from observed high-frequency (HF) data. Our method models LF extrapolation as a conditional generation task, using a neural network trained via denoising diffusion to approximate the gradient of the log-probability of LF data given the HF observations, also known as the conditional score function. During sampling, we enforce physical consistency by introducing a soft constraint based on the plane-wave partial differential equation, leveraging the observation that HF and LF wavefields share similar local slopes. This guides the generative process toward kinematically plausible extrapolations. We demonstrate on synthetic datasets that our method improves amplitude coherence and continuity in complex waveform settings. Figure 1 shows a scheme with an overview of the proposed methodology. The approach also supports uncertainty quantification through posterior sampling: each diffusion trajectory yields a different plausible LF reconstruction. This capability is particularly valuable in FWI, where LF data strongly influence the recovery of long-wavelength velocity models and proper initialization helps mitigate local minima. By analyzing variability across LF realizations, we gain insights into inversion uncertainty. Overall, the method combines data-driven generation with physics-based guidance to enable realistic, interpretable LF reconstructions with direct relevance for seismic inversion workflows.

Low-frequency (LF) components are critical for resolving deep structures and mitigating cycle skipping in full-waveform inversion (FWI). However, seismic surveys often lack sufficient LF content below 4 Hz due to source and receiver limitations. We propose and evaluate a data-driven method for LF extrapolation using Bidirectional Long Short-Term Memory (BiLSTM) networks trained exclusively on synthetic data. The model learns to reconstruct missing LF content from conventional airgun data by observing how high-frequency traces relate to their low-frequency-rich counterparts. The training dataset comprises 100,000 synthetic traces generated by convolving randomized reflectivity series with estimated airgun and tuned pulse source (TPS) signatures. Additive noise is introduced before and after convolution to simulate acquisition variability and unmodeled effects. The network is trained to map airgun-band inputs to TPS-band outputs using a hybrid loss function combining time-domain mean squared error with a frequency-domain regularization term, encouraging spectral recovery below 4 Hz. To assess generalization, we apply the trained model to a real field dataset acquired with dual-source acquisition (airgun and TPS). Despite no exposure to real data during training, the BiLSTM predicts low-frequency components that closely match the TPS ground truth. Spectral analysis shows significant enhancement below 4 Hz, and a 3 Hz low-pass filter reveals strong event alignment between the model output and TPS recordings (Figure). These results demonstrate the potential of synthetic training pipelines and recurrent architectures to enrich legacy seismic datasets and support LF-sensitive applications like FWI. The method offers a lightweight, field-deployable tool requiring no changes to acquisition hardware.

Mountainous areas in South Asia are generally prone to natural hazards like floods, landslides, and earthquakes; yet conventional geospatial hazard analysis does not always make the best use of the rich cultural memory embedded in traditional lore. This study proposes an interdisciplinary approach that synthesizes geomythology—consisting of myths, legends, and oral tradition that contain environmental information—into geospatial decision support systems (GDSS) for hazard mitigation. Through the analysis of ancient texts (e.g., the Nilamata Purana), regional ballads, and folklore, we georeferenced and plotted narratives of major hazard-prone events, e.g., the Kedarnath Floods (2013), the Kangra Earthquake (1905), and the Great Assam Earthquake (1897).By superimposing oral tradition on geospatial information—such as fault lines, landslide data sets, and river catchment basins—we generated GIS layers that reveal significant correlations between physical hazards and cultural topics. Sample case studies suggest that temples, sacred landscapes, and mythic warnings (such as Dhari Devi’s wrath or extinguishing Jwala Mukhi’s flame) often coincide with geologically active zones. These markers are de facto ”seismic sentinels” in local folklore, retaining knowledge about hazards over generations. Our methodological integration provides a concept test for the incorporation of traditional ecological knowledge (TEK) into scientific models. The combined framework increases public engagement, facilitates culturally appropriate disaster communication, and broadens the temporal horizons of hazard perception—connecting deep time and proximal risk. This study contributes to emerging work in disaster anthropology and resilience research by demonstrating how myths function as cultural expressions as well as historically grounded sources of empirical insight into environmental risk.

Understanding the basal geometry and thickness of a landslide is crucial for slope stability and risk assessment, design of mitigative measures, and the study of landscape evolution. However, estimating landslide thickness remains a major challenge. Existing methods primarily generate two-dimensional profiles of landslide thickness. Nevertheless, these methods often rely on oversimplified geometric assumptions and do not fully exploit the three-dimensional nature of surface displacement data. In this work, we propose TIKE (Thickness Inversion with Kinematic Elements), a novel method for estimating the location of slip surfaces of landslides based on mass conservation and the integration of surface kinematic and geomorphic units. Building upon the framework introduced by Booth et al. (2013), we present an inversion model that considers spatial inhomogeneity of landslides and explicitly accounts for the kinematic boundaries of landslide sections. This new approach leverages recent advances in inverse theory, utilizing domain-based regularization and the Balancing Principle to automatically determine the appropriate level of regularization. Preliminary results from the well-studied Slumgullion landslide support the validity of the proposed approach. Direct measurements of landslide thickness are used to validate the inversion results. Surface geomorphic features, such as the presence of mini-slides and the locations of stationary ponds, are also used to assess the model predictions. The estimated basal surface reveals a channel-like structure, consistent with prior studies, and pronounced localized depressions beneath stationary ponds that were previously underappreciated. These results indicate that TIKE can substantially improve landslide thickness estimation, especially in data-rich settings with mapped kinematic units, and provide a promising tool for advancing three-dimensional landslide characterization.

In Utah, significant agricultural water demand (75% of available water) combined with arid climate and population growth puts pressure on existing water resources. Also, Utah climate is very variable and strongly correlated with topographic and land-surface contrasts, with prevailing geographical weather patterns. Agricultural water management can vary depending on climatic and landscape conditions, hindering its assessment. To address this, an implemented remote sensing-based water balance will give an insight into agricultural non-consumptive water use, and performance metrics will be presented to look for opportunities of agricultural water diversion reduction under diverse Utah geographies and Irrigation Canal Companies (ICCs). In this study, as part of the USDA-sponsored Secure Water Future project, historical-remote-sensing data (OpenET and gridMET) and Geographic Information Systems (GIS) data were inputs to the water balance model at ICC scale located in three major basins (Bear River Basin, Colorado River Basin, and Sevier River Basin) during the 2020-2023 period. These basins have unique water use challenges such environmental (Bear River Basin/Great Salt Lake), legal (Upper Colorado River Basin), and supply (Sevier River Basin). The proposed water balance approach is run at monthly scale and addresses the change of soil water content depletion at the root zone (∆Dr) with precipitation, irrigation, evapotranspiration (without assumptions on water use efficiencies or losses), capillary rise from groundwater (CR), runoff (RO), and deep percolation (DP). CR, RO, and DP were merged as Water Balance Residual (WBR). Based on WBR and water balance components, performance metrics are evaluated from agronomical and irrigation standpoints considering effects of geography, growing season length, and agricultural water diversion along the season. These metrics will allow for a standard ICC comparison considering the different geographical conditions. Furthermore, the proposed water balance implementation and comparable metrics will help to identify opportunities for agricultural water diversion reduction. Thereby, supporting Utah’s agriculture resilience across the different climatic and landscape conditions existing in the state.

The detection of biosignatures in exoplanet atmospheres through transmission spectroscopy is essential for assessing planetary habitability, yet traditional atmospheric retrieval techniques are computationally intensive and not well-suited for the large amounts of data expected from upcoming missions. The proposed solution is a machine learning pipeline that leverages Convolutional Neural Networks (CNNs) for more efficient atmospheric composition determination from transmission spectra. A training dataset is generated directly in Python by simulating absorption-like features with added Gaussian noise and additional spectral clutter to mimic realistic artifacts. The spectra are represented as transit depth or intensity ((Rp/Rs)2) plotted with respect to wavelength, then standardized and formatted as inputs for the CNN. Data processing followed the Eureka! pipeline: calibration and reduction steps correct detector/systematic effects, transit light curves are extracted and fit, and the resulting wavelength-dependent transit depths are assembled into transmission spectra suitable for model input. However, because fully processing many targets end-to-end is computationally expensive, the majority of real spectra used in this project were sourced from the NASA Exoplanet Archive, providing a larger set of already-published transmission spectra for training and evaluation. The molecules considered in this work were H2O, CO2, and CH4. The model was trained to predict the presence of these compounds directly from the processed spectra. The trained models were then applied to real transmission spectra sourced primarily from the NASA Exoplanet Archive and their predictions were validated against published atmospheric results to assess model performance. Results show that the CNN can reliably detect H2O, CO2, and CH4 signatures, achieving >90% region-level accuracy across molecules with many predictions made at high confidence (p > 0.65). A one-sample proportion (binomial) test yields p levels p << 0.1 which further indicates performance as significantly better than chance for each molecule-specific model.  

Real-time, spatially dense networks of sensors measuring CO2 concentrations hold great promise for obtaining fine-grained attribution of the location, sectors, and trends of CO2 emissions in cities. Without these sensors, emissions are tracked via emissions inventories, which are expensive to maintain and based on assumptions such as accurate reporting by emitters. Treating the emissions inventory as the Bayesian prior updated by sensor observations grounds the posterior emissions in real measurements. However, this method is prior-dependent, computationally expensive, and impractical where emissions inventories are unreliable or unavailable. We present a novel benchmark dataset for machine learning-based CO2 emissions mapping from sparse ground-based sensor networks, addressing a gap in supervised learning resources for atmospheric inverse modeling. Our dataset consists of 1 km resolution CO2 concentration simulations over CONUS using WRF-Chem paired with Vulcan v3.0 emission fields, structured as city-sized patches (48 km × 48 km) with 72-hour temporal windows and augmented with spatial masking to simulate realistic sensor coverage (5-100% pixel coverage). We evaluate baseline models including linear regression, XGBoost, and a modified U-Net architecture. The U-Net significantly outperforms tabular models and maintains the ability to identify large point sources at 50% coverage but losing fine spatial detail below 25% coverage. Validation on real sensor data from Houston and the Bay Area reveals challenges in translating from simulated to real-world conditions, with 10-20 sensors in each city providing insufficient coverage for detailed emission mapping, highlighting the need for denser networks and improved models to enable practical urban emissions monitoring from sparse observations.