Optimal design of snow stake networks to estimate snow depth in an alpine mountain range

Abstract

Monitoring and estimation of snow depth in alpine catchments is needed for a proper assessment of management alternatives for water supply in these water resources systems. The distribution of snowpack thickness is usually approached by using field data that come from snow samples collected at a given number of locations that constitute the monitoring network. Optimal design of this network is required to obtain the best possible estimates. Assuming that there is an existing monitoring network, its optimization may imply the selection of an optimal network as a subset of the existing one (if there are no funds to maintain them) or enlarging the existing network by one or more stations (optimal augmentation problem). We propose an optimization procedure that minimizes the total variance in the estimate of snowpack thickness. The novelty of this work is to treat, for the first time, the problem of snow observation network optimization for an entire mountain range rather than for small catchments as done in the previous studies. Taking into account the reduced data available, which is a common problem in many mountain ranges, the importance of a proper design of these observation networks is even larger. Snowpack thickness is estimated by combining regression models to approach the effect of the explanatory variables and kriging techniques to consider the influence of the stakes location. We solve the optimization problems under different hypotheses, studying the impacts of augmentation and reduction, both, one by one and in pairs. We also analyse the sensitivity of results to nonsnow measurements deduced from satellite information.Finally, we design a new optimal network by combining the reduction and augmentation methods. The methodology has been applied to the Sierra Nevada mountain range (southern Spain), where very limited resources are employed to monitor snowfall and where an optimal snow network design could prove critical. An optimal snow observation network is defined by relocating some observation points. It would reduce the estimation variance by around 600 cm 2 (15%). K E Y W O R D Sestimation uncertainty, regression-kriging snow depth estimation, satellite information, Sierra Nevada (southern Spain), snow observation network optimization, snow stakes

Snow Depth Extraction From Time‐Lapse Imagery Using a Keypoint Deep Learning Model

“…In future iterations of this model, the model could be trained on large databases of publicly available time‐lapse snow pole data sets to achieve accuracy across multiple domains. Many studies reference snow pole data sets (Collados‐Lara et al., 2020; Cosgrove et al., 2021; Currier et al., 2016; Hofmeester et al., 2019; McCreight et al., 2014), but currently no standardized labeling framework or comprehensive database exists to consolidate across sources as in other domains (e.g., ImageNet; Deng et al., 2009). Future work could explore publishing a database of multiple data sets or a standardized labeling framework, lowering the threshold for future model training and testing.…”

“…The labeling demand increases in year‐to‐year studies, and when networks are installed in large watersheds or habitat ranges where upwards of hundreds of cameras have been installed at a time (Genthon et al., 2016; Kohler et al., 2006; Schöner et al., 2009). Furthermore, there are dozens of research groups utilizing time‐lapse cameras with snow poles, but snow pole designs vary in color and pattern depending on the study needs (Collados‐Lara et al., 2020; Cosgrove et al., 2021; Currier et al., 2016; Hofmeester et al., 2019; McCreight et al., 2014). The result is an enormous amount of data and a need for a scalable method to improve automated labeling.…”

Getting your money's worth: Testing the value of data for hydrological model calibration

“…Therefore, there is a need to more thoroughly assess what, when, and where to measure (Figure 1). Geostatistical approaches have been used to determine the optimal locations for temperature (Amorim et al, 2012), precipitation (Pardo‐Igúzquiza, 1998), and snow depth (Collados‐Lara et al, 2020) monitoring networks. Machine learning can also be used to determine the optimal network design (Chen et al, 2022; Oroza et al, 2016).…”

El manto de nieve en la península ibérica: climatología y sensibilidad a la variabilidad climática

“…En cualquier caso, los datos procedentes de las balizas han permitido realizar diversos trabajos científicos que han ayudado a avanzar en el conocimiento de la dinámica del manto de nieve y su relación con la hidroclimatología de la península ibérica. Algunos ejemplos los podemos encontrar en estudios hidrológicos , estudios de la variabilidad de los eventos de nevadas , estudios de tendencias Morán-Tejeda et al, 2013), validación in situ de información por satélite (Juan Collados-Lara et al, 2016), o incluso utilizar la información procedente de los jalones para reproducir el manto de nieve de manera distribuida (Collados-Lara et al, 2020, 2017Nogués-Bravo, 2006, 2005) entre otros. De forma semejante a la red ERHIN, pero en la vertiente francesa de los Pirineos, Meteofrance ha instalado la red NIVOSÊ, con 9 medidores automáticos de espesor del manto de nieve junto a estaciones automáticas de alta montaña.…”