Abstract. The fine scale distribution of snow is important for avalanche forecasting and biological refugia. Here, we test how historical context provided by 3 m snow cover observations derived from PlanetScope commercial satellite imagery could be used to downscale fractional snow covered area (fSCA) observed by the MODerate resolution Imaging Spectroradiometer (MODIS), Visible Infrared Imaging Radiometer Suite (VIIRS), and the Harmonized Landsat and Sentinel-2 (HLS) product using previously published approaches. We evaluate this over Colorado and California montane meadows using 1) probabilistic snow cover maps, and 2) random forest machine learning models. We then compare versus a downscaling benchmark that relies only on terrain characteristics, eliminating the need for PlanetScope observations. Provided these three approaches, downscaling snow cover using random forest models performed best on average, largely because these models corrected annually consistent snow cover biases between PlanetScope and coarser resolution fSCA observations. The approach used to downscale influenced 3 m snow cover most, followed closely by the accuracy of the fSCA estimate that snow cover was downscaled from. Snow cover downscaled from HLS using only terrain indices was often similar or better than snow cover downscaled from MODIS and VIIRS using context from PlanetScope. This demonstrates how a limited historical record of commercial satellite observations can be used to estimate the fine-scale pattern of snow cover in many regions, but also when publicly accessible remote sensing retrievals and information about the terrain may obviate the need for commercial observations.

In snow-dominated mountain belts, understanding how runoff hydrology and landslide hazard will respond to climate change requires the integration of climate science, hydrology, and geomorphology. In this study, we use the DHSVM distributed hydrology model coupled with a Landlab shallow landslide probability model (LandslideProbability) to assess future shallow landslide hazard in the North Cascades Mountains of Washington, United States where high-relief terrain leads to distinct rain-dominated, transient (rain and snow) and snow-dominated precipitation zones. Three future climate scenarios that represent median, low and high warming scenarios from a collection of Coupled Model Intercomparison Project 5 models of future climate are used to force the hydrology model. We find that, for all scenarios, landslide hazard increases in the transient and rain-dominated zones but counter to our expectations, decreases in the snow dominated zone. In the rain-dominated and transient zones, peak saturation is driven by rainfall inputs during the fall-winter period and increased future precipitation rates directly translate to increased recharge. In contrast, in the snow-dominated zone, future peak saturation is driven by snowmelt water inputs, which historically occurred during the early summer months, when temperature and total incoming radiation are high. In the future, warmer temperatures cause the snowpack to melt before reaching the high-melt-energy summer months which results in a decrease in melt that exceeds the increase in precipitation and overall lower recharge rates.

Global climate models often simulate atmospheric conditions incorrectly due to their coarse grid resolution, flaws in their dynamics, and biases resulting from parameterization schemes. Here we document the magnitude and extent of minimum temperature biases in the CMIP6 model ensemble, relative to ERA5. Bias in the southern Cascadia region (i.e. Pacific Northwestern United States and southwestern British Columbia, Canada, spanning from the coast to the Rocky Mountains) stands out relative to the rest of North America, with some models showing a bias in excess of -10°C in the 1st percentile of daily winter minimum temperature. During the coldest minimum temperature days, the CMIP6 models show an anomalous high in mean sea level pressure in the Northeast Pacific – an atmospheric blocking pattern that is also present in ERA5. While this atmospheric blocking pattern is typically concurrent with cold temperatures across much of North America, terrain barriers such as the Rockies and Cascades prevent the cold air from reaching the Pacific Northwest in observation and reanalysis. Our results suggest that the bias in CMIP6 minimum temperatures is a result of unresolved topography in the Rockies and Cascade mountain ranges, such that the terrain does not adequately block cold air advection from the interior of the continent.

Snow water equivalent (SWE) distribution at fine spatial scales (≤ 10 m) is difficult to estimate due to modeling and observational constraints. However, the distribution of SWE throughout the spring snowmelt season is often correlated to the timing of snow disappearance. Here, we show that snow cover maps generated from PlanetScope’s constellation of Dove Satellites can resolve the 3 m date of snow disappearance across seven alpine domains in California and Colorado. Across a 5-year period (2019 – 2023), the average uncertainty in the date of snow disappearance, or the period of time between the last date of observed snow cover and the first date of observed snow absence, was 3 days. Using a simple shortwave-based snowmelt model calibrated at nearby snow pillows, the PlanetScope date of snow disappearance could be used to reconstruct spring snow water equivalent (SWE). Relative to lidar SWE estimates, the SWE reconstruction had a spatial coefficient of correlation of 0.75, and SWE spatial variability that was biased by 9%, on average. SWE reconstruction biases were then improved to within 0.04 m, on average, by calibrating snowmelt rates to track the spring temporal evolution of fractional snow cover observed by PlanetScope, including fractional snow cover over the full modeling domain, and across domain subsections where snowmelt rates may differ. This study demonstrates the utility of fine-scale and high-frequency optical observations of snow cover, and the simple and annually repeatable connections between snow cover and spring snow water resources in regions with seasonal snowpack.

Although Artificial Intelligence (AI) projects are common and desired by many institutions and research teams, there are still relatively few success stories of AI in practical use for the Earth science community. Many AI practitioners in Earth science are trapped in the prototyping stage and their results have not yet been adopted by users. Many scientists are still hesitating to use AI in their research routine. This paper aims to capture the landscape of AI-powered geospatial data sciences by discussing the current and upcoming needs of the Earth and environmental community, such as what practical AI should look like, how to realize practical AI based on the current technical and data restrictions, and the expected outcome of AI projects and their long-term benefits and problems. This paper also discusses unavoidable changes in the near future concerning AI, such as the fast evolution of AI foundation models and AI laws, and how the Earth and environmental community should adapt to these changes. This paper provides an important reference to the geospatial data science community to adjust their research road maps, find best practices, boost the FAIRness (Findable, Accessible, Interoperable, and Reusable) aspects of AI research, and reasonably allocate human and computational resources to increase the practicality and efficiency of Earth AI research.

Abstract Mountain meadows are an essential part of the alpine–subalpine ecosystem; they provide ecosystem services like pollination and are home to diverse plant communities. Changes in climate affect meadow ecology on multiple levels, for example, by altering growing season dynamics. Tracking the effects of climate change on meadow diversity through the impacts on individual species and overall growing season dynamics is critical to conservation efforts. Here, we explore how to combine crowd‐sourced camera images with machine learning to quantify flowering species richness across a range of elevations in alpine meadows located in Mt. Rainier National Park, Washington, USA. We employed three machine‐learning techniques (Mask R‐CNN, RetinaNet and YOLOv5) to detect wildflower species in images taken during two flowering seasons. We demonstrate that deep learning techniques can detect multiple species, providing information on flowering richness in photographed meadows. The results indicate higher richness just above the tree line for most of the species, which is comparable with patterns found using field studies. We found that the two‐stage detector Mask R‐CNN was more accurate than single‐stage detectors like RetinaNet and YOLO, with the Mask R‐CNN network performing best overall with mean average precision (mAP) of 0.67 followed by RetinaNet (0.5) and YOLO (0.4). We found that across the methods using anchor box variations in multiples of 16 led to enhanced accuracy. We also show that detection is possible even when pictures are interspersed with complex backgrounds and are not in focus. We found differential detection rates depending on species abundance, with additional challenges related to similarity in flower characteristics, labeling errors and occlusion issues. Despite these potential biases and limitations in capturing flowering abundance and location‐specific quantification, accuracy was notable considering the complexity of flower types and picture angles in this dataset. We, therefore, expect that this approach can be used to address many ecological questions that benefit from automated flower detection, including studies of flowering phenology and floral resources, and that this approach can, therefore, complement a wide range of ecological approaches (e.g., field observations, experiments, community science, etc.). In all, our study suggests that ecological metrics like floral richness can be efficiently monitored by combining machine learning with easily accessible publicly curated datasets (e.g., Flickr, iNaturalist).

ABSTRACT The use of artificial intelligence (AI) and machine learning (ML) methods in the geosciences can be categorized into three types, those that: (1) accelerate computationally expensive Earth system models; (2) fill the vacuum where numerical and physics-based models struggle; and (3) enable and enlighten data-driven discoveries. To achieve these tasks, many cyberinfrastructure (CI) systems are required. This chapter reviews the cutting-edge CI aiding the implementation of AI in the geosciences. Each technique presented is evaluated to assist geoscientists in determining how appropriate it is. Use cases in the subdomains of seismology, hydrology, and climatology are introduced to help readers understand the workflows. Challenges and future opportunities for CI development center on big data, provenance, interoperability, and heterogeneity due to the scale and complexity that future AI models in the geosciences will require.

Mountain snowpack provides critical water resources for forest and meadow ecosystems that are experiencing rapid change due to global warming. An accurate characterization of snowpack heterogeneity in these ecosystems requires snow cover observations at high spatial resolutions, yet most existing snow cover datasets have a coarse resolution. To advance our observation capabilities of snow cover in meadows and forests, we developed a machine learning model to generate snow-covered area (SCA) maps from PlanetScope imagery at about 3-m spatial resolution. The model achieves a median F1 score of 0.75 for 103 cloud-free images across four different sites in the Western United States and Switzerland. It is more accurate (F1 score = 0.82) when forest areas are excluded from the evaluation. We further tested the model performance across 7,741 mountain meadows at the two study sites in the Sierra Nevada, California. It achieved a median F1 score of 0.83, with higher accuracy for larger and simpler geometry meadows than for smaller and more complexly shaped meadows. While mapping SCA in regions close to or under forest canopy is still challenging, the model can accurately identify SCA for relatively large forest gaps (i.e., 15m < DCE < 27m), with a median F1 score of 0.87 across the four study sites, and shows promising accuracy for areas very close (>10m) to forest edges. Our study highlights the potential of high-resolution satellite imagery for mapping mountain snow cover in forested areas and meadows, with implications for advancing ecohydrological research in a world expecting significant changes in snow.

EDITORIAL article Front. Water, 31 August 2023Sec. Water and Critical Zone Volume 5 - 2023 | https://doi.org/10.3389/frwa.2023.1274748

Integrating Artificial Intelligence (AI) techniques with remote sensing holds great potential for revolutionizing data analysis and applications in many domains of Earth sciences. This review paper synthesizes the existing literature on AI applications in remote sensing, consolidating and analyzing AI methodologies, outcomes, and limitations. The primary objectives are to identify research gaps, assess the effectiveness of AI approaches in practice, and highlight emerging trends and challenges. We explore diverse applications of AI in remote sensing, including image classification, land cover mapping, object detection, change detection, hyperspectral and radar data analysis, and data fusion. We present an overview of the remote sensing technologies, methods employed, and relevant use cases. We further explore challenges associated with practical AI in remote sensing, such as data quality and availability, model uncertainty and interpretability, and integration with domain expertise as well as potential solutions, advancements, and future directions. We provide a comprehensive overview for researchers, practitioners, and decision makers, informing future research and applications at the exciting intersection of AI and remote sensing.

Many watershed‐scale and land surface models incorporate snowmelt modules with simplified representations of the snowpack with three or fewer layers. These modeling choices were traditionally made to reduce model complexity and computational demand while still being able to simulate large model domains. However, these simple snow layering schemes may not always simulate snow processes and the effects of climate change across a range of climatic and geographic conditions. Here we evaluate simple snow layering schemes (having two to five layers, commonly found in watershed‐scale and land surface models) against a synthetic benchmark with up to 100 layers at three locations with different climate conditions using the SUMMA modeling framework. We evaluate 10 different layering configurations of two to five layers with variable thicknesses and show that the effect of the layering scheme varies with site conditions. We find that the layer configuration is more important at a cold high elevation site in the Sierra Nevada, California (∼1.4°C annual average temperature), and at a warm site in the French Alps (∼6.5°C), and less important at a site in Idaho (∼5.0°C). The top layer thickness of the simpler snow layering configurations also influences the simulated snow surface and snowpack temperatures and timing of snowmelt. Our tests showed that the five‐layer model with thin layers near the surface was closest to the benchmark (median NSE = 0.99), and therefore we recommend using multiple snow layers for reliable simulations of snow accumulation and melt across a range of climates.

Previous studies have documented how forests influence snow at fine spatial scales, but none have documented the influence that existing forest-snow variability has on streamflow. To test how much forest-controlled snow variability influences streamflow, a tiling parameterization based on classifications from high-resolution (1-3 m) vegetation maps was incorporated into the Distributed Hydrology and Soil Vegetation Model (DHSVM). Within each grid cell (90-150 m), the tiling parameterization simulated forest-snow variability with four independently evolving snowpacks. Each tile had unique radiation conditions to represent conditions underneath the canopy, in exposed areas, and along north- and south-facing forest edges. This tiled parameterization was used to test where and when detailed forest-snow modelling should be considered further and where and when the impacts are too small to be worth the effort. To test this, tiled model simulations of streamflow were compared to non-tiled model simulations in the Sierra Nevada, CA, the Jemez Mountains, NM, and the Eastern Cascades, WA. In Tuolumne, CA, the tiled model simulated little difference in grid cell average SWE, and late-season streamflow decreased by only 3%-4% compared to the non-tiled model. In Jemez, NM, the tiled model decreased late-season streamflow by 18% due to increased sublimation. In Chiwawa, WA, the tiled model increased late-season streamflow by 15% due to high shortwave radiation attenuation and less longwave radiation enhancement from the forest. Furthermore, within the Chiwawa, a substantial silvicultural practice was synthetically implemented to increase the north-facing edge's fractional area. This silvicultural experiment, which used the same fractional forest area in all simulations increased late-season streamflow by 35% compared to tiled model simulations that did not represent forest edges. In conclusion, representing forest-SWE variability had an effect on late-season streamflow in some watersheds but not in others based on the fractional area of the forest edges, forest characteristics, and climate conditions.

Improving high-resolution (meter-scale) mapping of snow-covered areas in complex and forested terrains is critical to understanding the responses of species and water systems to climate change. Commercial high-resolution imagery from Planet Labs, Inc. (Planet, San Francisco, CA, USA) can be used in environmental science, as it has both high spatial (0.7–3.0 m) and temporal (1–2 day) resolution. Deriving snow-covered areas from Planet imagery using traditional radiometric techniques have limitations due to the lack of a shortwave infrared band that is needed to fully exploit the difference in reflectance to discriminate between snow and clouds. However, recent work demonstrated that snow cover area (SCA) can be successfully mapped using only the PlanetScope 4-band (Red, Green, Blue and NIR) reflectance products and a machine learning (ML) approach based on convolutional neural networks (CNN). To evaluate how additional features improve the existing model performance, we: (1) build on previous work to augment a CNN model with additional input data including vegetation metrics (Normalized Difference Vegetation Index) and DEM-derived metrics (elevation, slope and aspect) to improve SCA mapping in forested and open terrain, (2) evaluate the model performance at two geographically diverse sites (Gunnison, Colorado, USA and Engadin, Switzerland), and (3) evaluate the model performance over different land-cover types. The best augmented model used the Normalized Difference Vegetation Index (NDVI) along with visible (red, green, and blue) and NIR bands, with an F-score of 0.89 (Gunnison) and 0.93 (Engadin) and was found to be 4% and 2% better than when using canopy height- and terrain-derived measures at Gunnison, respectively. The NDVI-based model improves not only upon the original band-only model’s ability to detect snow in forests, but also across other various land-cover types (gaps and canopy edges). We examined the model’s performance in forested areas using three forest canopy quantification metrics and found that augmented models can better identify snow in canopy edges and open areas but still underpredict snow cover under forest canopies. While the new features improve model performance over band-only options, the models still have challenges identifying the snow under trees in dense forests, with performance varying as a function of the geographic area. The improved high-resolution snow maps in forested environments can support studies involving climate change effects on mountain ecosystems and evaluations of hydrological impacts in snow-dominated river basins.