This study presents a 35-year snow phenology record for the Yukon River Basin (YRB), developed using a random forest (RF) model at a 3.125 km resolution, capturing detailed trends in snowmelt onset and snow-off. The RF model, incorporating dynamic daily variables, improves upon traditional threshold-based methods by reducing sensitivity to transient thaw events and atmospheric noise. Model evaluation against station observations yielded a mean absolute error (MAE) of 10.5 d and a root mean square error (RMSE) of 13.7 d for snowmelt onset. For snow-off, the model achieved an MAE of 18.1 d and an RMSE of 20.7 d. This approach successfully mapped snow phenology across the diverse YRB landscape, providing valuable insight into how variations in snow cover align with regional climate patterns. Challenges such as sample bias due to limited ground-based data coverage highlight the need to expand in situ measurements to improve model performance further. Trend analysis segmented by two timeframes, 1988–2005 and 2006–2023, revealed distinct climate impacts on snow phenology. During 1988–2005, high snowfall and stable temperatures resulted in hastened snowmelt onset and lengthened snowmelt durations, reflecting early-season snow abundance. In contrast, from 2006–2023, warming spring and summer temperatures corresponded to progressively earlier snowmelt onset and snow-off. These shifts in snowmelt patterns align with a lengthened snow-free season, indicating an increasing influence of warmer temperatures on the snowpack. This RF-derived dataset provides an essential tool for tracking climate-driven snow changes, offering insights into hydrologic and ecologic dynamics in the YRB under accelerating climate change.

Rain-on-snow (ROS) events are characterized by liquid precipitation or condensation onto snow surfaces that can lead to snowmelt and the formation of ice layers. ROS events can directly alter the physical structure and thermal properties of the snowpack, leading to rapid melting and runoff-induced flooding, reduced snow insulation, and permafrost degradation. However, tracking ROS events and regional trends remain uncertain due to limited ground measurements and lack of long-term satellite ROS observations representing vast and remote boreal-Arctic landscapes. We quantified ROS dynamics over Alaska by developing a daily 6 km resolution ROS record using an established gradient ratio polarization approach and 30 year (1988–2017) satellite observations from the Special Sensor Microwave Imager/Image Sounder (SSMI/S) sensors. The data record captured well-documented ROS events and showed high consistency ( R 0.94) with alternative ROS predictions from the Advanced Microwave Scanning Radiometer-EOS/2 sensors. Analysis showed an overall increasing ROS frequency with significant trends mainly identified during early winter. Notable rises in ROS frequency were also detected in mid-to-high elevation ranges (>400 m above sea level (ASL)), while this increase diminished at higher elevations (>1000 m ASL). Our analysis further confirmed that the warming climate plays a fundamental role in driving these ROS events, with significantly positive correlations between ROS frequency and air temperature. However, the significant correlations did not extend to the Eastern and Western Gulf climate zones of southern Alaska, where the ROS retrievals were likely affected by coastal ocean contamination of SSMI/S observations.

A machine learning task can be viewed as a sequential pipeline of different algorithmic choices, including data preprocessing, model selection, and hyper-parameter tuning. Automated machine learning selects this sequence in an automated manner. While such approaches are natural in supervised settings, they remain challenging for unsupervised tasks such as outlier detection because of the lack of availability of label-centric feedback. In this paper, we present an instance-level metalearning approach for outlier detection. This approach learns how outlier instances are related to normal points in many labeled data sets to create a supervised meta-model. This meta-model is then used on a new (unlabeled) data set to predict outliers. We show the robustness of our approach on several benchmarks from the OpenML repository.

Lakes in Southwest Alaska are a critical habitat to many species and provide livelihoods to many communities through subsistence fishing, transportation, and recreation. Consistent and reliable data are rarely available for even the largest lakes in this sparsely populated region, so data-intensive methods utilizing long-term observations and physical data are not possible. To address this, we used optical remote sensing (MODIS 2002–2016) to establish a phenology record for key lakes in the region, and we modeled lake-ice formation and breakup for the years 1982–2022 using readily available temperature and solar radiation-based predictors in a survival modeling framework that accounted for years when lakes did not freeze. Results were validated with observations recorded at two lakes, and stratification measured by temperature arrays in three others. Our model provided good predictions (mean absolute error, freeze-over = 11 days, breakup = 16 days). Cumulative freeze-degree days and cumulative thaw-degree days were the strongest predictors of freeze-over and breakup, respectively. Lake volume appeared to mediate lake-ice phenology, as ice-cover duration tended to be longer and less variable in lower-volume lakes. Furthermore, most lakes < 10 km3 showed a trend toward shorter ice seasons of −1 to −6 days/decade, while most higher-volume lakes showed undiscernible or positive trends of up to 2 days/decade. Lakes > 20 km3 also showed a greater number of years when freeze-over was neither predicted by our model (37 times, n = 200) nor observed in the MODIS record (19 times, n = 60). While three lakes in our study did not commonly freeze throughout our study period, four additional high-volume lakes began experiencing years in which they did not freeze, starting in the late 1990s. Our study provides a novel approach to lake-ice prediction and an insight into the future of lake ice in the Boreal region.

Glaciers are important sentinels of a changing climate, crucial components of the global cryosphere and integral to their local landscapes. However, many of the commonly used methods for mapping glacier change are labor-intensive and limit the temporal and spatial scope of existing research. This study addresses some of the limitations of prior approaches by developing a novel deep-learning-based method called GlacierCoverNet. GlacierCoverNet is a deep neural network that relies on an extensive, purpose-built training dataset. Using this model, we created a record of over three decades long at a fine temporal cadence (every two years) for the state of Alaska. We conducted a robust error analysis of this dataset and then used the dataset to characterize changes in debris-free glaciers and supraglacial debris over the last ~35 years. We found that our deep learning model could produce maps comparable to existing approaches in the capture of areal extent, but without manual editing required. The model captured the area covered with glaciers that was ~97% of the Randolph Glacier Inventory 6.0 with ~6% and ~9% omission and commission rates in the southern portion of Alaska, respectively. The overall model area capture was lower and omission and commission rates were significantly higher in the northern Brooks Range. Overall, the glacier-covered area retreated by 8425 km2 (−13%) between 1985 and 2020, and supraglacial debris expanded by 2799 km2 (64%) during the same period across the state of Alaska.

Sequential decision making under uncertainty has many prac- tical real-world applications, ranging from supply-chain management to autonomous driving. Reinforcement learning has recently emerged as a promising technique for solving these kinds of problems. However, reinforcement learning algo- rithms are known to be highly sensitive to their internal hyperparameters which require significant expert manual effort for tuning their values. This in turn limits the widespread ap-plicability of reinforcement learning based solutions in prac- tice. In this paper, we introduce a new automated decision optimization system called AutoDO for end-to-end solving of sequential decision making problems. More specifically, the system not only automatically selects the best reinforcement learning algorithm but it also finds the best configuration of its hyperparameters using a search strategy based on limited discrepancy search coupled with Bayesian optimization. Fur-thermore, our system supports several different flavours of reinforcement learning including online and offline as well as model-free and model-based algorithms. We experiment with classical control benchmarks as well as a more realistic inventory management problem. Our results show that the pro- posed system is competitive and often outperforms existing state-of-the-art in terms of the quality of the solutions found.

The Yukon River basin encompasses over 832,000 km2 of boreal Arctic Alaska and northwest Canada, providing a major transportation corridor and multiple natural resources to regional communities. The river seasonal hydrology is defined by a long winter frozen season and a snowmelt-driven spring flood pulse. Capabilities for accurate monitoring and forecasting of the annual spring freshet and river ice breakup (RIB) in the Yukon and other northern rivers is limited, but critical for understanding hydrologic processes related to snow, and for assessing flood-related risks to regional communities. We developed a regional snow phenology record using satellite passive microwave remote sensing to elucidate interactions between the timing of upland snowmelt and the downstream spring flood pulse and RIB in the Yukon. The seasonal snow metrics included annual Main Melt Onset Date (MMOD), Snowoff (SO) and Snowmelt Duration (SMD) derived from multifrequency (18.7 and 36.5 GHz) daily brightness temperatures and a physically-based Gradient Ratio Polarization (GRP) retrieval algorithm. The resulting snow phenology record extends over a 29-year period (1988–2016) with 6.25 km grid resolution. The MMOD retrievals showed good agreement with similar snow metrics derived from in situ weather station measurements of snowpack water equivalence (r = 0.48, bias = −3.63 days) and surface air temperatures (r = 0.69, bias = 1 day). The MMOD and SO impact on the spring freshet was investigated by comparing areal quantiles of the remotely sensed snow metrics with measured streamflow quantiles over selected sub-basins. The SO 50% quantile showed the strongest (p < 0.1) correspondence with the measured spring flood pulse at Stevens Village (r = 0.71) and Pilot (r = 0.63) river gaging stations, representing two major Yukon sub-basins. MMOD quantiles indicating 20% and 50% of a catchment under active snowmelt corresponded favorably with downstream RIB (r = 0.61) from 19 river observation stations spanning a range of Yukon sub-basins; these results also revealed a 14–27 day lag between MMOD and subsequent RIB. Together, the satellite based MMOD and SO metrics show potential value for regional monitoring and forecasting of the spring flood pulse and RIB timing in the Yukon and other boreal Arctic basins.

A large number of time series forecasting models including traditional statistical models, machine learning models and more recently deep learning have been proposed in the literature. However, choosing the right model along with good parameter values that performs well on a given data is still challenging. Automatically providing a good set of models to users for a given dataset saves both time and effort from using trial-and-error approaches with a wide variety of available models along with parameter optimization. We present AutoAI for Time Series Forecasting (AutoAI-TS) that provides users with a zero configuration (zero-conf) system to efficiently train, optimize and choose best forecasting model among various classes of models for the given dataset. With its flexible zero-conf design, AutoAI-TS automatically performs all the data preparation, model creation, parameter optimization, training and model selection for users and provides a trained model that is ready to use. For given data, AutoAI-TS utilizes a wide variety of models including classical statistical models, Machine Learning (ML) models, statistical-ML hybrid models and deep learning models along with various transformations to create forecasting pipelines. It then evaluates and ranks pipelines using the proposed T-Daub mechanism to choose the best pipeline. The paper describe in detail all the technical aspects of AutoAI-TS along with extensive benchmarking on a variety of real world data sets for various use-cases. Benchmark results show that AutoAI-TS, with no manual configuration from the user, automatically trains and selects pipelines that on average outperform existing state-of-the-art time series forecasting toolkits.

A large number of time series forecasting models including traditional statistical models, machine learning models and more recently deep learning have been proposed in the literature. However, choosing the right model along with good parameter values that performs well on a given data is still challenging. Automatically providing a good set of models to users for a given dataset saves both time and effort from using trial-and-error approaches with a wide variety of available models along with parameter optimization. We present AutoAI for Time Series Forecasting (AutoAI-TS) that provides users with a zero configuration (zero-conf ) system to efficiently train, optimize and choose best forecasting model among various classes of models for the given dataset. With its flexible zero-conf design, AutoAI-TS automatically performs all the data preparation, model creation, parameter optimization, training and model selection for users and provides a trained model that is ready to use. For given data, AutoAI-TS utilizes a wide variety of models including classical statistical models, Machine Learning (ML) models, statistical-ML hybrid models and deep learning models along with various transformations to create forecasting pipelines. It then evaluates and ranks pipelines using the proposed T-Daub mechanism to choose the best pipeline. The paper describe in detail all the technical aspects of AutoAI-TS along with extensive benchmarking on a variety of real world data sets for various use-cases. Benchmark results show that AutoAI-TS, with no manual configuration from the user, automatically trains and selects pipelines that on average outperform existing state-of-the-art time series forecasting toolkits.

AI automation tools need machine-readable hyperparameter schemas to define their search spaces. At the same time, AI libraries often come with good human-readable documentation. While such documentation contains most of the necessary information, it is unfortunately not ready to consume by tools. This paper describes how to automatically mine Python docstrings in AI libraries to extract JSON Schemas for their hyperparameters. We evaluate our approach on 119 transformers and estimators from three different libraries and find that it is effective at extracting machine-readable schemas. Our vision is to reduce the burden to manually create and maintain such schemas for AI automation tools and broaden the reach of automation to larger libraries and richer schemas.

Snowoff (SO) date—defined as the last day of observed seasonal snow cover—is an important governor of ecologic and hydrologic processes across Alaska and Arctic-Boreal landscapes; however, our understanding and capacity for the monitoring of spatial and temporal variability in the SO date is still lacking. In this study, we present a 6.25 km spatially gridded passive microwave (PMW) SO data record, complimenting current Alaskan SO records from Moderate Resolution Imaging Spectrometer (MODIS) and Landsat, but extending the SO record an additional 13 years. The PMW SO record was validated against in situ snow depth observations and showed favorable accuracy (0.66–0.92 mean correlations; 2–10 day mean absolute errors) for the major climate regions of Alaska. The PMW SO results were also within 10 days of finer spatial scale SO observational records, including Interactive Multisensor Snow and Ice Mapping System (IMS), MODIS, and Landsat, for a majority (75%) of Alaska. However, the PMW record showed a general SO delay at higher elevations and across the Alaska North Slope, and earlier SO in the Alaska interior and southwest regions relative to the other SO records. Overall, we assign an uncertainty +/−11 days to the PMW SO. The PMW SO record benefits from the near-daily temporal fidelity of underlying brightness temperature (Tb) observations and reveals a mean regional trend in earlier SO timing (−0.39 days yr−1), while significant (p < 0.1) SO trend areas encompassed 11% of the Alaska domain and ranged from −0.11 days yr−1 to −1.31 days yr−1 over the 29-year satellite record. The observed SO dates also showed anomalous early SO dates during markedly warm years. Our results clarify the pattern and rate of SO changes across Alaska, which are interactive with global warming and contributing to widespread permafrost degradation, changes in regional hydrology, ecosystems, and associated services. Our results also provide a robust means for SO monitoring from satellite PMW observations with similar precision as more traditional and finer scale observations.

Livestock production is a socioeconomic linchpin in Mongolia and is affected by large-scale livestock die-offs. Colloquially known as dzuds, these die-offs are driven by anomalous climatic events, including extreme cold temperatures, extended snow cover duration (SCD) and drought. As average temperatures across Mongolia have increased at roughly twice the global rate, we hypothesized that increasing cold season surface melt including soil freeze/thaw (FT), snowmelt, and icing events associated with regional warming have become increasingly important drivers of dzud events as they can reduce pasture productivity and inhibit access to grazing. Here, we use daily brightness temperature (Tb) observations to identify anomalous surface melt and icing events across Mongolia from 2003–2016 and their contribution to dzuds relative to other climatic drivers, including winter temperatures, SCD, and drought. We find a positive relationship between surface melt and icing events and livestock mortality during the fall in southern Mongolia and during the spring in the central and western regions. Further, anomalous seasonal surface melt and icing events explain 17–34% of the total variance in annual livestock mortality, with cold temperatures as the leading contributor of dzuds (20–37%). Summer drought showed the greatest explanatory power (43%) but overall had less statistically significant relationships relative to winter temperatures. Our results indicate that surface melt and icing events will become an increasingly important driver of dzuds as annual temperatures and livestock populations are projected to increase in Mongolia.

Snow covers Arctic and boreal regions (ABRs) for approximately 9 months of the year, thus snowscapes dominate the form and function of tundra and boreal ecosystems. In recent decades, Arctic warming has changed the snowcover's spatial extent and distribution, as well as its seasonal timing and duration, while also altering the physical characteristics of the snowpack. Understanding the little studied effects of changing snowscapes on its wildlife communities is critical. The goal of this paper is to demonstrate the urgent need for, and suggest an approach for developing, an improved suite of temporally evolving, spatially distributed snow products to help understand how dynamics in snowscape properties impact wildlife, with a specific focus on Alaska and northwestern Canada. Via consideration of existing knowledge of wildlife-snow interactions, currently available snow products for focus region, and results of three case studies, we conclude that improving snow science in the ABR will be best achieved by focusing efforts on developing data-model fusion approaches to produce fitfor-purpose snow products that include, but are not limited to, wildlife ecology. The relative wealth of coordinated in situ measurements, airborne and satellite remote sensing data, and modeling tools being collected and developed as part of NASA's Arctic Boreal Vulnerability Experiment and SnowEx campaigns, for example, provide a data rich environment for developing and testing new remote sensing algorithms and retrievals of snowscape properties.