Spatiotemporal estimation of snow depth using point data from snow stakes,digital terrain models,and satellite data

Snow availability in Alpine catchments plays an important role in water resources management. In this paper, we propose a method for an optimal estimation of snow depth (areal extension and thickness) in Alpine systems from point data and satellite observations by using significant explanatory variables deduced from a digital terrain model. It is intended to be a parsimonious approach that may complement physical‐based methodologies. Different techniques (multiple regression, multicriteria analysis, and kriging) are integrated to address the following issues: We identify the explanatory variables that could be helpful on the basis of a critical review of the scientific literature. We study the relationship between ground observations and explanatory variables using a systematic procedure for a complete multiple regression analysis. Multiple regression models are calibrated combining all suggested model structures and explanatory variables. We also propose an evaluation of the models (using indices to analyze the goodness of fit) and select the best approaches (models and variables) on the basis of multicriteria analysis. Estimation of the snow depth is performed with the selected regression models. The residual estimation is improved by applying kriging in cases with spatial correlation. The final estimate is obtained by combining regression and kriging results, and constraining the snow domain in accordance with satellite data. The method is illustrated using the case study of the Sierra Nevada mountain range (Southern Spain). A cross‐validation experiment has confirmed the efficiency of the proposed procedure. Finally, although it is not the scope of this work, the snow depth is used to asses a first estimation of snow water equivalent resources.     

Diagnosing snow accumulation errors in a rain‐snow transitional environment with snow board observations

Diagnosing the source of errors in snow models requires intensive observations, a flexible model framework to test competing hypotheses, and a methodology to systematically test the dominant snow processes. We present a novel process‐based approach to diagnose model errors through an example that focuses on snow accumulation processes (precipitation partitioning, new snow density, and snow compaction). Twelve years of meteorological and snow board measurements were used to identify the main source of model error on each snow accumulation day. Results show that modeled values of new snow density were outside observational uncertainties in 52% of days available for evaluation, while precipitation partitioning and compaction were in error 45% and 16% of the time, respectively. Precipitation partitioning errors mattered more for total winter accumulation during the anomalously warm winter of 2014–2015, when a higher fraction of precipitation fell within the temperature range where partition methods had the largest error. These results demonstrate how isolating individual model processes can identify the primary source(s) of model error, which helps prioritize future research.     

Snow model sensitivity analysis to understand spatial and temporal snow dynamics in a high‐elevation catchment

In this paper, we addressed a sensitivity analysis of the snow module of the GEOtop2.0 model at point and catchment scale in a small high‐elevation catchment in the Eastern Italian Alps (catchment size: 61 km²). Simulated snow depth and snow water equivalent at the point scale were compared with measured data at four locations from 2009 to 2013. At the catchment scale, simulated snow‐covered area (SCA) was compared with binary snow cover maps derived from moderate‐resolution imaging spectroradiometer (MODIS) and Landsat satellite imagery. Sensitivity analyses were used to assess the effect of different model parameterizations on model performance at both scales and the effect of different thresholds of simulated snow depth on the agreement with MODIS data. Our results at point scale indicated that modifying only the “snow correction factor” resulted in substantial improvements of the snow model and effectively compensated inaccurate winter precipitation by enhancing snow accumulation. SCA inaccuracies at catchment scale during accumulation and melt period were affected little by different snow depth thresholds when using calibrated winter precipitation from point scale. However, inaccuracies were strongly controlled by topographic characteristics and model parameterizations driving snow albedo (“snow ageing coefficient” and “extinction of snow albedo”) during accumulation and melt period. Although highest accuracies (overall accuracy = 1 in 86% of the catchment area) were observed during winter, lower accuracies (overall accuracy < 0.7) occurred during the early accumulation and melt period (in 29% and 23%, respectively), mostly present in areas with grassland and forest, slopes of 20–40°, areas exposed NW or areas with a topographic roughness index of ?0.25 to 0 m. These findings may give recommendations for defining more effective model parameterization strategies and guide future work, in which simulated and MODIS SCA may be combined to generate improved products for SCA monitoring in Alpine catchments.     

Exploration of remotely sensed forest structure and ultrasonic range sensor metrics to improve empirical snow models

Current methods to estimate snow accumulation and ablation at the plot and watershed levels can be improved as new technologies offer alternative approaches to more accurately monitor snow dynamics and their drivers. Here we conduct a meta‐analysis of snow and vegetation data collected in British Columbia to explore the relationships between a wide range of forest structure variables – obtained from Light Detection and Ranging (LiDAR), hemispherical photography (HP) and Landsat Thematic Mapper – and several indicators of snow accumulation and ablation estimated from manual snow surveys and ultrasonic range sensors. By merging and standardizing all the ground plot information available in the study area, we demonstrate how LiDAR‐derived forest cover above 0.5 m was the variable explaining the highest percentage of absolute peak snow water equivalent (SWE) (33%), while HP‐derived leaf area index and gap fraction (45° angle of view) were the best potential predictors of snow ablation rate (explaining 57% of variance). This study reveals how continuous SWE data from ultrasonic sensors are fundamental to obtain statistically significant relationships between snow indicators and structural metrics by increasing mean r² by 20% when compared to manual surveys. The relationships between vegetation and spectral indices from Landsat and snow indicators, not explored before, were almost as high as those shown by LiDAR or HP and thus point towards a new line of research with important practical implications. While the use of different data sources from two snow seasons prevented us from developing models with predictive capacity, a large sample size helped to identify outliers that weakened the relationships and suggest improvements for future research. A concise overview of the limitations of this and previous studies is provided along with propositions to consistently improve experimental designs to take advantage of remote sensing technologies, and better represent spatial and temporal variations of snow. Copyright © 2013 John Wiley & Sons, Ltd.     

Rain on snow induced urban floods in cold maritime climate: Risk,indicators and trends

Urban floods pose a societal and economical risk. This study evaluated the risk and hydro-meteorological conditions that cause pluvial flooding in coastal cities in a cold climate. Twenty years of insurance claims data and up to 97 years of meteorological data were analysed for Reykjavík, Iceland (64.15°N; <100 m above sea level). One third of the city's wastewater collection system is combined, and pipe grades vary from 0.5% to 10%. Results highlight semi-intensive rain (<7 mm/h; ≤3 year return period) in conjunction with snow and frozen ground as the main cause for urban flood risk in a climate which undergoes frequent snow and frost cycles (avg. 13 and 19 per season, respectively). Floods in winter were more common, more severe and affected a greater number of neighbourhoods than during summer. High runoff volumes together with debris remobilized with high winds challenged the capacity of wastewater systems regardless of their age or type (combined vs. separate). The two key determinants for the number of insurance claims were antecedent frost depth and total precipitation volume per event. Two pluvial regimes were particularly problematic: long duration (13–25 h), late peaking rain on snow (RoS), where snowmelt enhanced the runoff intensity, elongated and connected independent rainfall into a singular, more voluminous (20–76 mm) event; shorter duration (7–9 h), more intensive precipitation that evolved from snow to rain. Closely timed RoS and cooling were believed to trigger frost formation. A positive trend was detected in the average seasonal snow depth and volume of rain and snowmelt during RoS events. More emphasis, therefore, needs to be placed on designing and operating urban drainage infrastructure with regard to RoS co-acting with frozen ground. Furthermore, more detailed, routine monitoring of snow and soil conditions is important to predict RoS flood events.     

Inference and uncertainty of snow depth spatial distribution at the kilometre scale in the Colorado Rocky Mountains: the effects of sample size,random sampling,predictor quality,and validation procedures

Historically, observing snow depth over large areas has been difficult. When snow depth observations are sparse, regression models can be used to infer the snow depth over a given area. Data sparsity has also left many important questions about such inference unexamined. Improved inference, or estimation, of snow depth and its spatial distribution from a given set of observations can benefit a wide range of applications from water resource management, to ecological studies, to validation of satellite estimates of snow pack. The development of Light Detection and Ranging (LiDAR) technology has provided non‐sparse snow depth measurements, which we use in this study, to address fundamental questions about snow depth inference using both sparse and non‐sparse observations. For example, when are more data needed and when are data redundant? Results apply to both traditional and manual snow depth measurements and to LiDAR observations. Through sampling experiments on high‐resolution LiDAR snow depth observations at six separate 1.17‐km² sites in the Colorado Rocky Mountains, we provide novel perspectives on a variety of issues affecting the regression estimation of snow depth from sparse observations. We measure the effects of observation count, random selection of observations, quality of predictor variables, and cross‐validation procedures using three skill metrics: percent error in total snow volume, root mean squared error (RMSE), and R². Extremes of predictor quality are used to understand the range of its effect; how do predictors downloaded from internet perform against more accurate predictors measured by LiDAR? Whereas cross validation remains the only option for validating inference from sparse observations, in our experiments, the full set of LiDAR‐measured snow depths can be considered the ‘true’ spatial distribution and used to understand cross‐validation bias at the spatial scale of inference. We model at the 30‐m resolution of readily available predictors, which is a popular spatial resolution in the literature. Three regression models are also compared, and we briefly examine how sampling design affects model skill. Results quantify the primary dependence of each skill metric on observation count that ranges over three orders of magnitude, doubling at each step from 25 up to 3200. Whereas uncertainty (resulting from random selection of observations) in percent error of true total snow volume is typically well constrained by 100–200 observations, there is considerable uncertainty in the inferred spatial distribution (R²) even at medium observation counts (200–800). We show that percent error in total snow volume is not sensitive to predictor quality, although RMSE and R² (measures of spatial distribution) often depend critically on it. Inaccuracies of downloaded predictors (most often the vegetation predictors) can easily require a quadrupling of observation count to match RMSE and R² scores obtained by LiDAR‐measured predictors. Under cross validation, the RMSE and R² skill measures are consistently biased towards poorer results than their true validations. This is primarily a result of greater variance at the spatial scales of point observations used for cross validation than at the 30‐m resolution of the model. The magnitude of this bias depends on individual site characteristics, observation count (for our experimental design), and sampling design. Sampling designs that maximize independent information maximize cross‐validation bias but also maximize true R². The bagging tree model is found to generally outperform the other regression models in the study on several criteria. Finally, we discuss and recommend use of LiDAR in conjunction with regression modelling to advance understanding of snow depth spatial distribution at spatial scales of thousands of square kilometres. Copyright © 2012 John Wiley & Sons, Ltd.     

From soil aggregates to riverine flocs: a laboratory experiment assessing the respective effects of soil type and flow shear stress on particles characteristics

Particles eroded from hillslopes and exported to rivers are recognized to be composite particles of high internal complexity. Their architecture and composition are known to influence their transport behaviour within the water column relative to discrete particles. To‐date, hillslope erosion studies consider aggregates to be stable once they are detached from the soil matrix. However, lowland rivers and estuaries studies often suggest that particle structure and dynamics are controlled by flocculation within the water column. In order to improve the understanding of particle dynamics along the continuum from hillslopes to the lowland river environment, soil particle behaviour was tested under controlled laboratory conditions. Seven flume erosion and deposition experiments, designed to simulate a natural erosive event, and five shear cell experiments were performed using three contrasting materials: two of them were poorly developed and as such can not be considered as soils, whilst the third one was a calcareous brown soil. These experiments revealed that soil aggregates were prone to disaggregation within the water column and that flocculation may affect their size distribution during transport. Large differences in effective particle size were found between soil types during the rising limb of the bed shear stress sequence. Indeed, at the maximum applied bed shear stress, the aggregated particles median diameter was found to be three times larger for the well‐developed soil than for the two others. Differences were smaller in the falling limb, suggesting that soil aggregates underwent structural changes. However, characterization of particles strength parameters showed that these changes did not fully turn soil aggregates into flocs, but rather into hybrid soil aggregate–floc particles. Copyright © 2013 John Wiley & Sons, Ltd.     

A rain on snow climatology and temporal analysis for the eastern United States

ABSTRACT

Rain-on-snow (ROS) has the potential to produce devastating floods by enhancing runoff from snowmelt. Although a common phenomenon across the eastern United States, little research has focused on ROS in this region. This study used a gridded observational snow dataset from 1960–2009 to establish a comprehensive seasonal climatology of ROS for this region. Additionally, different rain and snow thresholds were compared while considering temporal trends in ROS occurrence at four grid cells representing individual locations. Results show most ROS events occur in MAM (March-April-May). ROS events identified with rainfall >1 cm are more frequent near the east coast and events identified with >1 cm snow loss are more common in higher latitudes and/or elevations. Decreasing trends in DJF (December-January-February) ROS events were identified near the coastal areas, with increasing trends in the northern portion of the domain. Significant decreasing trends in MAM ROS are likewise present on a regional scale. Factors playing a role in snowpack depth and rainfall, such as movement of storm tracks in this region, should be considered with future work to discern mechanisms causing the changes in ROS frequency.