Energy burden, the ratio of energy expenditure to household income, is a critical yet often overlooked measure of economic and environmental inequality in the United States. A high energy burden, 6% or greater, is not just a financial issue; it is a public health and environmental justice concern, as frontline communities often experience greater exposure to pollution, poorer housing efficiency, and heightened vulnerability to extreme weather events. This study uses self-organizing maps (SOMs), an unsupervised neural network, to identify contributing factors and inform policy interventions for energy-burdened communities in the North, South, Midwest, and West census regions, a novel use of this method. It is also among the first to integrate environmental justice indicators, including outdoor air quality metrics and health disparities, as determinants of energy burden. In addition to environmental justice indicators, socioeconomic status, building characteristics, and power outages are explored to assist policymakers, engineers, and advocates working within the energy transition. Results revealed statistically significant ( p < 0.05) differences in these indicators across SOM-defined energy-burden regimes. For the Midwest and South regions, all 45 indicators showed statistical significance, while 44 were significant in the Northeast, and 41 were significant for the West. These findings suggest that high energy-burden regimes tend to coincide with elevated environmental and health risk indicators, which may intensify under climate change.

Abstract We adapt a semi‐Bayesian hierarchical modeling framework to jointly characterize the space–time variability of seasonal precipitation totals and precipitation extremes across the Northern Great Plains (NGP). In this framework, seasonal precipitation totals at each station and year are modeled using a Gamma distribution, while seasonal maximum precipitation extremes are modeled with a Generalized Extreme Value distribution. Parameters for both distributions vary spatially and temporally as linear functions of selected covariates, including Pacific and Atlantic sea surface temperature indices and regional precipitation anomalies. Covariate coefficients are estimated using Maximum Likelihood methods. In the process layer, these ML‐estimated coefficients for both precipitation totals and extremes are spatially modeled through Gaussian multivariate processes, capturing spatial dependencies among stations. Appropriate priors on model hyperparameters complete the Bayesian formulation at this hierarchical level, resulting in conditional posterior distributions. We apply the model to precipitation data from 60 NGP stations spanning the period 1951–2019 under contemporaneous (0‐month), 1‐month, and 2‐month lead times, thereby assessing the predictive utility of teleconnection signals at varying lead times. Model evaluation and cross‐validation results highlight strong performance in capturing historical spatial and temporal variability in precipitation across all lead times, with contemporaneous models showing the highest skill, as expected. Conditional posterior distributions derived from this modeling approach provide probabilistic forecasts of seasonal precipitation totals and extremes, offering critical information for agricultural planning, ecosystem conservation, and water resource management decisions under climate variability and change conditions.

Cloudbursts, defined as sudden, intense rainfall episodes, are increasingly frequent extreme weather events in the Indo-Himalayan region, causing widespread devastation to human life and property; yet understanding their causal mechanisms and improving predictability remains constrained by incomplete knowledge of atmospheric and land-based precursors. Particularly, the role of soil moisture as a vital land-surface component has been underexplored in the context of cloudburst formation. This study hypothesizes that increased soil moisture from agricultural irrigation amplifies atmospheric moisture fluxes via land-atmosphere coupling and contributes to enhanced cloudburst risk. The objective here is to attribute moisture source locations, identify critical pre-event land-atmospheric indicators, and assess soil–atmosphere coupling through the analysis of IMD-specified cloudburst events from 1991 to 2020 using the Indian Land Data Assimilation System (ILDAS) dataset. We employ NOAA's Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) back-trajectory model and create Integrated Vapor Transport (IVT) maps, composited with winds, surface pressure, and sea level pressure, to trace moisture source locations. Pre-event anomaly detection and change-point analysis are performed using the Pruned Exact Linear Time (PELT) algorithm on soil moisture, precipitation, evaporation, and runoff variables across nine spatially proximate grid cells per event. Additionally, extreme percentile threshold exceedances and non-parametric persistence metrics quantify the early-warning potential. Decadal NDVI trends contextualize Land Use/Land Cover (LULC) influences. Results reveal moisture source hotspots in regions undergoing land-use transitions, with steep pressure gradients establishing strong circulation patterns that contribute moisture to multiple cloudburst events. Significant temporal anomalies occur across all four variables, with threshold exceedances and change-point detections ranging from 2 to 10 occurrences per event and anomaly persistence spanning 2 to 8 days for soil moisture. Early warning lead times of 15 to 120 days are identified for soil moisture, precipitation, evaporation, and runoff anomalies preceding the cloudburst events. These findings suggest that further quantifying the causal links among these variables can better help understand soil–atmosphere coupling and substantially improve early warning systems for detecting extreme rainfall events.

Growing challenges of water scarcity and extreme weather events, including floods, significantly impact cities globally. Indian urban agglomerations (UAs) face these challenges, highlighting the pressing need to understand rainfall variability. This study presents a holistic assessment of rainfall dynamics across 63 UAs in Peninsular India. It integrates change-point detection, long-term trends, and the influence of data periods on trend assessments (moving window approach). Distinct change-points were identified in rainfall attributes - annual (AR), annual maximum (AMR), monsoon (MR), and post-monsoon rainfall (PMR), during mid-20th century. AR, MR, and PMR decreased in southwestern UAs (Western Ghats), while AMR trends are more diverse. A moving window approach revealed four temporal trend patterns, highlighting the importance of data periods in trend assessments and aiding identification of water-stressed agglomerations. Observed rainfall variability and climate interactions pose challenges for sustainable development. Thus, the study offers a framework to evaluate spatio-temporal rainfall changes in UAs and provides a scientific basis to integrate rainfall variability into action plans, policy strategies and sustainable solutions.

Abstract Bayesian Causal Networks (BCNs) are adapted to investigate causal factors of hydroclimate variability in the Brahmaputra River Basin (BRB) during the seasonal and sub‐seasonal periods. This is the longest monsoonal river system in India, prone to frequent severe flooding. Directed Acyclic Graphs are constructed to connect response variable (streamflow) with various covariates, which include indices of large‐scale climate drivers, Sea Surface Temperatures (SSTs), across tropical Indo‐Pacific and basin Precipitation, and Bayes Factors are computed for flow regimes to quantify the strength of causal connections. Results show that eastern precipitation exerts strong influence during Early and Peak monsoon (|BF| > 1), while western precipitation dominates during the late season. Niño4 index exhibits substantial influence ((|BF| > 0.5) during Early, Peak and Monsoon season. Bayesian Causal Network was applied to each grid point SSTs and Antecedent SSTs in Directed Acyclic Graph configuration to capture lagged ocean influences. Regions (“hotspots”) in Central Pacific in the Niño4 and Eastern Pacific regions exert strong causality in modulating streamflow variability during Early, Peak and Monsoon season. Composite analysis of Integrated Vapor Transport and pressure anomalies show concomitant patterns. In wet years, strong moisture transport from the Bay of Bengal during Early and Peak seasons is driven by anomalous high‐pressure region over Southeast Asia, and vice‐versa in dry years. SSTs from the hotspots in machine learning model show interesting nonlinear interactions at varying thresholds in enabling skillful probabilistic estimates of seasonal flow categories. The results provide new insights into streamflow variability in BRB and potential for use developing skillful forecasts of streamflow attributes.

Spatial footprint, that is, the length scale of precipitation extremes, directly impacts the affected area and flooding [1, 2]. However, a physical framework for analyzing it has yet to be developed. Here, we investigate the seasonal and spatial distribution of the spatial footprint of precipitation extremes, their observed changes, and the underlying physical processes using the observational records of the last four decades. We show that subtropical arid regions are global hotspots for large-scale precipitation extremes, which are triggered mainly by the breaking of planetary-scale waves. The eddy length scales measure the spatial scales at which the weather systems are most prevalent. In the extratropics (poleward of 30◦ N/S), the eddy length scales exhibit a significant positive correlation with the length scale of precipitation extremes. Our analysis indicates that the eddy length scales offer a useful framework to assess the spatial footprint of precipitation extremes in the extratropics. Although precipitation is scarce in arid regions, sudden deluges with large spatial footprints over just a few hours to days can make them highly vulnerable to flash floods. We emphasize that the frameworks for design rainfall estimation [3–5] should account for the spatial footprint of events in addition to the conventional characteristics of intensity, duration, and frequency.

Wildfires can dramatically alter water quality, resulting in severe implications for human and freshwater systems. However, regional-scale assessments of these impacts are often limited by data scarcity. Here, we unify observations from 1984-2021 in 245 burned watersheds across the western United States, comparing post-fire signals to baseline levels from 293 unburned basins. Organic carbon and phosphorus exhibit significantly elevated levels (p <= 0.05) in the first 1-5 years post-fire, while nitrogen and sediment show significant increases up to 8 years post-fire. During peak post-fire response years, average carbon, nitrogen, and phosphorus concentrations are 3-103 times pre-fire levels, and sediment 19-286 times pre-fire concentrations. Higher responses are linked with greater forested and developed areas, which respectively explain up to 31 and 33% of inter-basin response variability. Overall, this analysis provides strong evidence of multi-year water quality degradation following wildfires in the western United States and highlights the influence of basin and wildfire features. These insights may aid water managers in preparation efforts, increasing resilience of water systems to wildfire impacts.

It is well established that streamflow regimes evolve over decadal time scales (i.e., low frequency) leading to long term shifts in distributions. Similar low frequency variations have also been documented in streamflow predictability. Here we explore connections between streamflow distribution attributes and predictability regimes in the Upper Colorado River Basin. We employ nonlinear dynamical time series analysis methods on streamflow timeseries covering the period 762 - 2019 for six locations in the basin. First, a wavelet spectral analysis is performed to obtain the quasi-periodic 'signal' of the streamflow. The wavelet analysis also provides the temporal variability of the variance of the signal time series. The signal time series is embedded in a D-dimensional space with appropriate lag to reconstruct the phase space of the dynamics - i.e. the attractor. Overall predictability is assessed by quantifying the average divergence trajectories in the phase space using Global Lyapunov Exponents and the temporal variability of predictability via the Local Lyapunov Exponents. Results show clear oscillations in streamflow predictability with periods of both high and low predictability occurring throughout the study period at all gauges. Comparing predictability timeseries across the stream gauges we find that general consistency in high and low predictability periods, although they do not perfectly align temporally. In general, higher (lower) predictability periods are characterized by lower (higher) streamflow variance. While there is not a clear relationship between streamflow magnitude and predictability in general, modern high predictability epochs are characterized by a slightly greater likelihood of dry years and lower likelihood of wet years than other epochs. These findings indicate the potential for statistically significant differences in streamflow signatures between high and low predictability periods. Exploring these fundings further with potential connections to large-scale climate can be helpful in exploiting them for skillful short and medium term flow projections.

We adapt a Bayesian Hierarchical Network Model (BHNM) for modeling and ensemble forecasting of daily river stages at multiple gauges on a rainfed river network. The stage at a gauge on any day is modeled as a Probability Density Function (PDF), with parameters varying temporally and spatially across the gauges. The PDF parameters vary as a function of covariates that include stage, streamflow, and precipitation from previous times at upstream gauges and catchment areas between gauges. This leverages the network structure of the river to capture spatial correlation along with the river basin hydrologic processes encapsulated in them. With suitable priors on the model parameters, likelihood functions, and, using Markov Chain Monte Carlo simulation approach predictive posterior distributions of all the space-time model parameters are obtained and, consequently, that of river stages across the network for any day. The best model, which includes the PDF type and sub-set of covariates, is obtained via objective criteria, Deviance Information Criteria (DIC). The model is demonstrated by applying it to daily stages at four gauges on the monsoon rainfed Narmada River in western India during the peak monsoon period (July-August) of 1978 to 2014 period. Model validation in cross-validation mode shows skillful and reliable forecasts of river stages, including higher stages that correspond to floods relative to a null-model of linear regression. Since river stages are often used in flood disaster management and preparedness efforts, this model and the skillful outputs enhance the potential for effective real time flood warning and mitigation.

The United States is one of the largest energy consumers per capita, requiring households to have adequate energy expenditures to keep up with modern demand regardless of financial cost. This paper investigates energy burden, defined as the ratio of household energy expenditures to household income. There is a lack of research on creating equitable policies for energy-burdened communities, including environmental justice indicators and community characteristics that could be used to predict and understand energy burden, along with socioeconomic status, building characteristics, and power outages, beneficial to policymakers, engineers, and advocates. Here, generalized additive models and random forests are explored for energy burden prediction using the original dataset and principal components, followed by a leave-one-column-out (LOCO) analysis to investigate indicator influence, with 25 identical indicators out of 42 appearing in the top 100 models. The generalized additive models generally outperform the random forests, with the best-performing model yielding a coefficient of determination of 0.92.

Effective management of water resources requires reliable estimates of land surface states and fluxes, including water balance components. But most land surface models run in uncoupled mode and do not produce river discharge at catchment scales to be useful for water resources management applications. Such integrated systems are also rare over India where hydrometeorological extremes have wreaked havoc on the economy and people. So, an Indian Land Data Assimilation System (ILDAS) with a coupled land surface and a hydrodynamic model has been developed and driven by multiple meteorological forcings (0.1 degrees, daily) to estimate land surface states, channel discharge, and floodplain inundation. ILDAS benefits from an integrated framework as well as the largest suite of observation records collected over India and has been used to produce a reanalysis product for 1981-2021 using four forcing datasets, namely, Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2), Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS), ECMWF's ERA-5, and Indian Meteorological Department (IMD) gridded precipitation. We assessed the uncertainty and bias in these precipitation datasets and validated all major components of the terrestrial water balance, i.e., surface runoff, soil moisture, terrestrial water storage anomalies, evapotranspiration, and streamflow, against a combination of satellite and in situ observation datasets. Our assessment shows that ILDAS can represent the hydrological processes reasonably well over the Indian landmass with IMD precipitation showing the best relative performance. Evaluation against ESA-CCI soil moisture shows that MERRA-2 based estimates outperform the others, whereas ERA-5 performs best in simulating evapotranspiration when evaluated against MODIS ET. Evaluations against observed records show that CHIRPS-based estimates have the highest performance in reconstructing surface runoff and streamflow. Once operational, this system will be useful for supporting transboundary water management decision making in the region.

We performed a systematic space-time analysis of monsoon seasonal (Jun-Sep) rainfall and extremes (3-day maximum rainfall) over India for the period 1951–2019. Employing Partition Around Medoid (PAM) clustering technique on the seasonal rainfall and extremes, six spatially coherent regions (clusters) were identified that are contiguous in space and consistent with the topography, which are: Central-West India (CW), Northwest and Northern India (NW), (WG) Western Ghats (WG), Deccan Plateau (DP), Central-East India including Indo-Gangetic plain (CE) and Northeast India (NE). Integrated Vertical Transport (IVT) of moisture composites for wet and dry years for each cluster indicated that Bay of Bengal is the major source of moisture for extreme rainfall for all of India, except for WG. Arabian Sea and Bay of Bengal both provide moisture for the seasonal rainfall for western and eastern halves of India, respectively. Trend analysis revealed decline in seasonal rainfall over CE, NE and WG clusters and increase in extreme rainfall over CW region. These are consistent with increasing IVT trends over Bay of Bengal and decreasing over Arabian Sea and Indian Ocean. Teleconnections to tropical Pacific Sea surface temperatures (SSTs) were reminiscent of El Nino Southern Oscillation (ENSO) patterns, with cooler SSTs in central and eastern Pacific favoring stronger monsoon rainfall and to a lesser extent the extremes. Further, warmer Indian Ocean in recent decades is likely a mediator in the moisture transport by reducing seasonal rainfall and enhancing the extremes. These interesting insights brighten the prospects for skillful forecast of monsoon rainfall and extremes.

Monsoon precipitation is the critical source of freshwater for some of the world’s most densely populated areas, yet extreme precipitation events in these regions present significant risks, including devastating floods and damage to agriculture and infrastructure. Recent events, such as the severe flooding and landslides during the 2023 North India monsoon and the 2022 Pakistan floods,1For example, https://foreignpolicy.com/2022/09/01/pakistan-flooding-crisis-climate-change-governance/ and https://www.theguardian.com/world/2023/jul/10/india-floods-new-delhi-rain-record-deaths (last assessed 30 September 2024).1 underscore the pressing need to better understand and predict these hazards. While the science of monsoons has been studied for decades, with theories centered on global dynamics and moist energy budgets to explain the zonal mean state of monsoon and factors leading to regional differences, one key theme of all these analyses is the spatiotemporal variability of rainfall from the dry to wet seasons. A key challenge is understanding and predicting extreme rainfall incidents during monsoon seasons to help mitigate dire undesired consequences.In the past two decades, nonlinear system dynamics has emerged as a novel and promising approach in climate science research, with complex network analysis becoming one of the most rapidly developing methods. Complex networks offer a powerful tool to uncover interactions among various geographic locations and teleconnection patterns, providing new insights into the behavior of monsoon systems. Since its origin in graph theory, network dynamics has evolved to focus on metrics such as centrality and community detection, which when applied to monsoon precipitation, particularly extremes, reveal coherent structures that were previously unidentified. Notably, network communities have shown strong associations with the major monsoon regions, offering fresh perspectives on monsoon dynamics.This paper synthesizes recent studies on monsoon precipitation, particularly those employing network metrics to understand key physical processes. While both statistical and dynamical models continue to struggle with predicting extreme monsoon precipitation, complex network analysis has identified new predictors related to global monsoon teleconnection patterns. These predictors address non-stationarities caused by climate variability, presenting opportunities to enhance monsoon predictions. Nonlinear system science thus holds significant potential for deepening our understanding of the spatiotemporal variability of global monsoon and extreme weather events.Finally, this paper outlines a future research agenda aimed at addressing key knowledge gaps. These include expanding the regions of study to explore region-to-region teleconnections, enhancing the physical understanding of network metrics, applying coupled networks, investigating the interannual and interdecadal variability of monsoons, and utilizing network diagnostics of climate model evaluation.

Lipid remodeling, the modification of cell membrane chemistry via structural rearrangements within the lipid pool of an organism, is a common physiological response amongst all domains of life to alleviate environmental stress and maintain cellular homeostasis. Whereas culture experiments and environmental studies of phytoplankton have demonstrated the plasticity of lipids in response to specific abiotic stressors, few analyses have explored the impacts of multi-environmental stressors at the community-level scale. Here, we study changes in the pool of intact polar lipids (IPLs) of a phytoplanktonic community exposed to multi-environmental stressors during a ~2-month long mesocosm experiment deployed in the eastern tropical South Pacific off the coast of Callao, Perú. We investigate lipid remodeling of IPLs in response to changing nutrient stoichiometries, temperature, pH, and light availability in surface and subsurface water-masses with contrasting redox potentials, using multiple linear regressions, classification and regression trees, and Random Forest analyses. Notable responses include the proportional increases of certain glycolipids (namely mono- and di-galactosyldiacylglyercols; MG and CO2(aq) availability and high pH are associated with increased DG and sulfoquinovosyldiacylglycerol (SQ) concentrations. DG, respectively) associated with thermal stress as well as the degradation of these lipids under oxygen stress. Reduced Higher production of MG in surface waters corresponds well with their stablished photoprotective and antioxidant mechanisms in thylakoid membranes. Certain phosphatidylglycerol (PG) moieties show strong linear trends with light availability and are known to be important components in electron transport processes of photosystems I and II. IPL remodeling suggests the variable pH, hypoxia, and photoinhibition. These physiological responses reallocate resources from structural or recycling of acyl chains for energy storage in the form of triacylglycerols (TAGs) in response to stressors; like N limitation, extrachloroplastic membrane lipids (i.e., phospholipids and betaine lipids) under high-growth conditions, to thylakoid/plastid membrane lipids (i.e., glycolipids and certain PGs) and TAGs under growth-limiting conditions. Investigation of this lipid remodeling system is necessary to understand how membrane reorganization can affect the pools of cellular C, N, and S, and how it may influence fluxes of biologically relevant elements to higher trophic levels and to the dissolved organic matter pool.

Decision Making Under Deep Uncertainty often uses prohibitively large scenario ensembles to calculate robustness and rank policies’ performance. This paper contributes a framework using subsampling algorithms and space-filling metrics to determine how smaller ensemble sizes impact the accuracy of robustness rankings. Subsampling methods create smaller scenario ensembles of varying sizes. We evaluate ranking sensitivity to the ensemble size and calculate accuracy relative to a baseline ranking. Then, metrics of scenario set quality predict ranking accuracy. Notably, the metrics and subsampling methods do not require additional model simulations. We demonstrate the framework with a case study of shortage policies for Lake Mead in the Colorado River Basin (CRB). Results suggest that fewer scenarios than previous studies can accurately rank Lake Mead policies, and results depend on the type of objective and robustness metric. Smaller ensembles could reduce the computational burden of robustness analyses in the ongoing CRB policy renegotiation.

There is an increasing need for skillful runoff season (i.e., spring) streamflow forecasts that extend beyond a 12-month lead time for water resources management, especially under multiyear droughts and particularly in basins with highly variable streamflow, large storage capacity, proclivity to droughts, and many competing water users such as in the Colorado River Basin (CRB). Ensemble streamflow prediction (ESP) is a probabilistic prediction method widely used in hydrology, including at the National Oceanic and Atmospheric Administration (NOAA) Colorado Basin River Forecasting Center (CBRFC) to forecast flows that the Bureau of Reclamation uses in their water resources operational decision models. However, it tends toward climatology at 5-month and longer lead times, causing decreased skill, particularly in forecasts critical for management decisions. We developed a modeling approach for seasonal streamflow forecasts using a machine learning technique, random forest (RF), for runoff season flows (April 1-July 31 total) at the important gauge of Lees Ferry, Arizona, on the CRB. The model predictors include antecedent basin conditions, large-scale climate teleconnections, climate model projections of temperature and precipitation, and the mean ESP forecast from CBRFC. The RF model is fitted and validated separately for lead times spanning 0 to 18 months over the period 1983-2017. The performance of the RF model forecasts and CBRFC ESP forecasts are separately assessed against observed streamflows in a cross validation mode. Forecast performance was evaluated using metrics including relative bias, root mean square error, ranked probability skill score, and reliability. Measured by ranked probability skill score, RF outperforms a climatological benchmark at all lead times and outperforms CBRFC's ESP hindcasts for lead times spanning 6 to 18 months. For the 6- to 18-month lead times, the RF ensemble median had a root mean square error that was between similar to 410- and similar to 620-thousand acre-feet lower than that of the ESP ensemble median (i.e., RF reduced ensemble median RMSE by -9% to -12% relative to ESP). Reliability was comparable between RF and ESP. More skillful long-lead cross-validated forecasts using machine learning methods show promise for their use in real time forecasts and better informed and efficient water resources management; however, further testing in various decision models is needed to examine RF forecasts' downstream impacts on key water resources metrics like robustness, reliability, and vulnerability.

Of concern to Colorado River management, as operating guidelines post-2026 are being considered, is whether water resource recovery from low flows during 2000-20 is possible. Here, we analyze new simulations from phase 6 of the Coupled Model Intercomparison Project (CMIP6) to determine plausible climate impacts on Colorado River flows for 2026-50 when revised guidelines would operate. We constrain projected flows for Lees Ferry, the gauge through which 85% of the river flow passes, using its estimated sensitivity to meteorological variability together with CMIP6-projected precipitation and temperature changes. The critical importance of precipitation, especially its natural variability, is emphasized. Model projections indicate increased precipitation in the upper Colorado River basin due to climate change, which alone increases river flows by 5%-7% (relative to a 2000-20 climatology). Depending on the river's temperature sensitivity, this wet signal compensates for some, if not all, of the depleting effects of basin warming. Considerable internal decadal precipitation variability (;5% of the climatological mean) is demonstrated, driving a greater range of plausible Colorado River flow changes for 2026-50 than previously surmised from treatment of temperature impacts alone: the overall precipitation-induced Lees Ferry flow changes span from -25% to 140%, contrasting with a range from -30% to -5% from expected warming effects only. Consequently, extreme low and high flows are more likely. Lees Ferry flow projections, conditioned on initial drought states akin to 2000-20, reveal substantial recovery odds for water resources, albeit with elevated risks of even further flow declines than in recent decades.

Snowpack in mountainous areas often provides water storage for summer and fall, especially in the Western United States. In situ observations of snow properties in mountainous terrain are limited by cost and effort, impacting both temporal and spatial sampling, while remote sensing estimates provide more complete spacetime coverage. Spatial estimates of fractional snow covered area (fSCA) at 30m are available every 16 days from the series of multispectral scanning instruments on Landsat platforms. Daily estimates at 463m spatial resolution are also available from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument on the Terra satellite. Fusing Landsat and MODIS fSCA images creates high resolution daily spatial estimates of fSCA that are needed for various uses: to support scientists and managers interested in energy and water budgets for water resources and to understand the movement of animals in a changing climate. Here, we propose a new machine learning approach conditioned on MODIS fSCA, as well as a set of physiographic features, and fit to Landsat fSCA over a portion of the Sierra Nevada USA. The predictions are daily 30m fSCA. The approach relies on two stages of spatially-varying models. The first classifies fSCA into three categories and the second yields estimates within (0, 100) percent fSCA. Separate models are applied and fitted within sub-regions of the study domain. Compared with a recently-published machine learning model (Rittger, Krock, et al., 2021), this approach uses spatially local (rather than global) random forests, and improves the classification error of fSCA by 16%, and fractionally-covered pixel estimates by 18%.

Lakes provide important water resources and many essential ecosystem services. Some of Earth's largest lakes recently reached record-low levels, suggesting increasing threats from climate change and anthropogenic activities. Yet, continuous monitoring of lake levels is challenging at a global scale due to the sparse in situ gauging network and the limited spatial or temporal coverage of satellite altimeters. A few pioneering studies used water areas and hypsometric curves to reconstruct water levels but suffered from large uncertainties due to the lack of high-quality hypsometry data. Here, we propose a novel proxy-based method to reconstruct multi-decadal water levels from 1992 to 2018 for both large and small lakes using Landsat images and ICESat (2003-2009) and recently launched ICESat-2 (2018+) laser altimeters. Using the new method, we evaluate reconstructed levels of 342 lakes worldwide, with sizes ranging from 1 to 81,844 km2. Reconstructed water levels have a median root-mean-square error (RMSE) of 0.66 m, equivalent to 57% of the standard deviation of monthly level variability. Compared with two recently reconstructed water level data sets, the proposed method reduces the median RMSE by 27%-32%. The improvement is attributable to the new method's robust construction of high-quality hypsometry, with a median R2 value of 0.92. Most reconstructed water level time series have a bi-monthly or higher frequency. Given that ICESat-2 and Landsat can observe hundreds of thousands of water bodies, this method can be applied to conduct an improved global inventory of time-varying lake levels and thus inform water resource management more broadly than existing methods. Landsat images and laser altimeters were leveraged to reconstruct multi-decadal lake levels of both large and small lakes Reconstructed water levels were validated against observed levels on 342 global lakes with a median error of 0.66 m Most of the reconstructed lake level time series have a bi-monthly or higher frequency

Climate change and human activities increasingly threaten lakes that store 87% of Earth’s liquid surface fresh water. Yet, recent trends and drivers of lake volume change remain largely unknown globally. Here, we analyze the 1972 largest global lakes using three decades of satellite observations, climate data, and hydrologic models, finding statistically significant storage declines for 53% of these water bodies over the period 1992–2020. The net volume loss in natural lakes is largely attributable to climate warming, increasing evaporative demand, and human water consumption, whereas sedimentation dominates storage losses in reservoirs. We estimate that roughly one-quarter of the world’s population resides in a basin of a drying lake, underscoring the necessity of incorporating climate change and sedimentation impacts into sustainable water resources management.