Snow is an essential component of the hydrological cycle [1
]. It serves as a reliable water resource and the dynamics of snow cover play a vital function in rebalancing the global energy and water budget [2
]. Due to the unique characteristics of snow, snow cover functions as an energy bank, radiation shield, insulator, reservoir, and water transport medium in the global climate-ecosystems [3
]. Environmental agents interact with snow in complex ways. To predict these interactions between forest structure and snow accumulation and melting, factors such as air temperatures, incoming shortwave and longwave radiation, snow albedo, precipitation, wind speed, wind redistribution, and relative humidity, as well as snow interception by forest canopy or vegetation structure, influencing the snowpack [4
], have to be determined.
Accumulation and ablation of seasonal snowpack within forested areas exhibit very different dynamics compared to snow within open field areas [5
]. Forests reduce shortwave radiation and increase longwave radiation due to trees as important emitters of longwave radiation. Because the decrease in shortwave radiation in forests plays a bigger role for the entire energy balance than the increase in longwave radiation, the snowmelt is much slower in forests compared to open regions [4
Snow albedo that depends on the snow properties especially snow grain size is generally lower in forests than in open areas [7
]. Wind redistribution, coincident with a precipitation event, can quickly alter accumulation patterns in open sites [8
]. In the forests, snow storage and snow redistribution depend mainly on forest type (coniferous, deciduous) [4
], although the importance of forest structure has also been highlighted as a significant factor influencing the under-canopy snow accumulation patterns and timing of snowmelt [9
]. In case of frequent snowfall, snow interception increases. The differences between snow storages accumulated in open areas and in forests were reported from many world’s regions [10
]. The mentioned studies reported by 30–50% lower snow storages in the coniferous forest compared to the adjacent open area depending on canopy structure. Similar differences in snow storages in forests compared to open areas have been also proved by [4
] based on measurements at the same study area as the study presented in this paper. Under freezing and low-wind conditions, the intercepted snow remain in the canopies longer and enhance sublimation directly to the atmosphere. By rising air temperatures sublimation causes faster upload of intercepted snow on the forest ground intensifying the ablation rates of the snow cover [3
]. Alternatively, with increasing forest cover, the interception of snow in the canopies reduces the amount of snow that accumulates on the ground, while shading reduces snow ablation compared to an open site [16
]. It is certain that the effect of forest on snow accumulation and ablation acts to both intensify and reduce the energy budget of snowpack, consequently creating much greater heterogenous snow cover patterns compared to open sites [11
Therefore, efforts to measure snow depth distributions at different places and different scales are crucial to simulating snow accumulation and ablation processing using, for instance, numerical models from conceptual to physically based distributed models or hydrological models that simulate snowpack quantity parameters (snow water equivalent (SWE) and snow depth), as summarized in [2
]. These hydrological models are one of the common snow measurement approaches based on numerical models. The only drawback of these models is the lack of reliable meteorological observations as model inputs [17
] and the complex physics of snow are still not adequately understood which hampers the development of snowpack models. Further approaches are snow pits and manual probing [18
]. These traditional methods are both times consuming and delivering point-scale accuracies. To obtain basin-scale snow cover observations, conventional non-invasive remote sensing techniques have been used for monitoring snow dynamics which succeeds the potential of undersampling of point-scale snow measurements, however, are affected by detection limitations in the lower spatial and temporal resolution that characterize the measurement of snow cover [8
]. Other recently existing methods include GPS-reflectometry [19
] and airborne Light Detection and Ranging (LiDAR) mapping [22
], Terrestrial Laser Scanner (TLS) [23
], digital photogrammetry [27
], tachymetry [29
], Ground Penetrating Radar (GPR) [30
], time-lapse photography [2
], or satellite-based sensors [33
]. These techniques are of high costs (e.g., airborne or helicopter flights) or only useful supplements to weather stations and manual measurements (e.g., time-lapse photography) [3
Due to the miniaturization of navigation sensors, it is now possible to choose between different camera systems, spanning from heavy-weight LiDAR mapping sensors to light-weight mini-multispectral systems [16
] making it possible to explore new, unforeseen research directions. Especially attractive is the low-cost technique integration of high-resolution unmanned aerial vehicle (UAV)-based digital photogrammetry, which captures small-scale spatial variabilities of snow cover. Since the development of the Structure from Motion (SfM) algorithm [34
], it is possible to reconstruct georeferenced point clouds, high precision digital surface models (DSMs), and orthomosaics with a high spatial resolution, down to the centimeter level, by automatic matching of common features from a set of overlapping images [34
]. Using the SfM method for snow cover image matching is still considered to be problematic, due to the homogeneity of the snow cover surface, but images acquired in the visual (VIS, λ = 400–700 nm) part of the electromagnetic spectrum are delivering promising results. Recent investigations of this technique of snow depth estimation used imagery taken by UAV for large to small catchments [18
]. These examples have shown vertical accuracies in retrieving snow depth values from the UAV, reaching accuracies in the range of 0.1–0.3 m. This new technique provides relatively accurate solutions that may be used, not only to obtain the snow depth at high resolution in open areas, as in the studies mentioned above but can also provide information on snow accumulation and melting in forests.
The amount of forest cover is often described by estimating canopy closure (CC) or gap fraction. Also, the leaf area index (LAI) [42
] is a variable to specify the forest cover by calculating the leaf area per leaf area per unit ground area. However, none of the three variables would directly represent the canopy interception through the amount of forest cover, but it can be an input variable in snow interception models [10
]. Since UAV platforms achieve image quality equal to traditional airborne photogrammetry, downward-looking UAV imagery can be used as an indirect method for LAI estimates, which is comparable to digital hemispherical photograph (DHP) analysis of the sky and canopy [44
]. The only difference is the background, which is snow-covered terrain instead of the sky [45
]. Measurements and modeling, however, become more challenging with forest which is changing due to natural disturbances (e.g., bark beetle and windstorms). Although the application of UAV to measure either snow depth or LAI is not new, there is still an only limited number of published studies focusing on the use of these new sensing methods for such purpose. Linking snowpack distribution and forest LAI measurements by way of downward-looking UAV-based photogrammetry have not been done yet in forested environments. There are significant interactions between open field and forest snow processes and the impacts on snowpack evolution under forest canopy [46
] which are not fully understood increasing the need for further investigations.
The objectives of our study were to introduce a workflow for efficient UAV-based snowpack monitoring, in particular (i) the monitoring of snow accumulation and ablation processes, (ii) the assessment of canopy characteristics as the effective LAI of crowns using downward-looking UAV images (iii) to analyze the relationship between the dynamic snow depth and LAI distribution in diverse Norway spruce stand. Over a complex winter season, spanning from snow accumulation to ablation, we determined snow depth and LAI distribution in a mid-latitude montane forest small-scale environment, featuring different structure and health status, including forest-free area, healthy and disturbed forest resulting from bark beetle infestation. The UAV imaging campaigns for snow depth and LAI retrieval were accompanied with repeated manual snow depth monitoring, as well as by ground-based LAI measurements by the LAI-2200 plant canopy analyzer and DHP, used as a benchmark for testing the accuracy of our approach.
This study has presented the potential of small footprint UAV data for deriving metrics such as snow depth (snow metric) and LAI (forest metric) using single flights. As was found in previous studies, we could show that it is possible to estimate snow depth and indirect LAI from UAV data by using RGB-based UAV imagery. Snow depth was computed by subtracting snow-free from snow-covered DSMs during different snow conditions and compared to fewer and with more manual ground measurements. Differences in measured values obtained by these snow samplings strategies draw distinctions. Effective winter LAI estimates conducted as well from downward-looking UAV images were compared to conventional optical methods (e.g., DHP and LAI-2200 canopy analyzer) producing reasonable but underestimated LAIeff values. With increased flight height of the UAV, lower estimated LAI values were determined for all forest structures.
The relation between LAI and snow depth was analyzed for different snow conditions. As expected, LAIeff values negatively correlated to snow depth data during both snow conditions at the forest site. For snowmelt, the LAI as one of the predictors for snow distribution in forested areas represented weaker correlations. The reason is most likely that LAI is a much stronger predictor for snow accumulation, because it better describes the snow depth distribution during the accumulation period, while for snowmelt, more processes influence the snow variability. Elevation and topography have a marginal effect here, as the study plot in the forest zone is situated in a flat area. Thus, for snowmelt, factors that are more closely related to radiation and topography might give better results when describing the snow depth variability at a smaller scale.
To this end, the results demonstrated that, with this combined method, snow depth and LAI retrieval could be recorded simultaneously with one UAV survey. The conjoint survey can significantly improve the efficiency of both aerial and field mapping of snow cover distribution in areas featuring variable structure and health status of the forest canopy. In such areas, it proved reliability in the determination of the snow depth variation along with the LAI, enabling the interpretation of the effects of canopy coverage on snow accumulation and ablation.
However, the largest advantage of the UAV technology here is the high resolution and the high level of detail regarding the snow-covered terrain and the quick monitoring of indirect LAI compared to simple manual measurements of snow depth and point measurements of DHP or using LiCor-2200. By improving this method both parameters snow depth and indirect LAI might be used as input parameters for snow models captured in reduced time.
Further work is required to examine the performance of snow depth change mapping in terms of the precision of snow depth change estimates under denser canopies where the non-vegetated surface is mainly covered and the use of infrared photography improves identification of different snow features such as windblown snow into openings in the forest, ice crusts, and wet snow surfaces. Attributes that need improvement as they affected the thresholding of binary images for the indirect LAI analysis were shadows and snow on tree crowns.