100 200 300 400 500
0.2
100 200 300 400 500
0.2 0.4 0.6 0.8
R[−] Figure 4.10:Reflectance R vs. snow density ρ
for 4 snow OEDs. Circles denote FRED simula-tion results in steps of ∼ 45 kg m−3, connecting lines are shown for a clearer visualization of the trend.
If snow density is fully neglected and snow OED is used as the only input parameter to analyze InfraSnow reflectance measurements, the resulting variability is much larger than the variability introduced by the natural snow inhomogeneity. For the plotted density range in Figure 4.10 reflectance shows a maximum relative variability of 25 % for OED = 200 µm up to 50 % for OED = 1400 µm. For an OED measurement method this means that only OED range can be determined within certain reflectance intervals. For natural snow, R > 0.7 measured by the InfraSnow then means that OED . 400 µm and R < 0.4 leads to an OED estimate of OED & 700 µm. The range 0.4 < R < 0.7 can be measured for a wide range of OEDs, which are not extreme values for fine- or very coarse-grained snow. Unfortunately, within this snow type range differences in OED can already lead to significantly different snow characteristics. Thus, snow density has to be accounted for to obtain a quantitative and useful OED measurement by the InfraSnow.
Currently, snow density can be determined with an accuracy of ∼ 50 kg m−3 by care-fully weighing a predefined snow volume or by electrical conductivity measurements. This corresponds to the resolution of the reflectance simulations plotted in Figure 4.10. The resulting absolute variability in reflectance is between 2 % and 6 % which translates to a relative reflectance variability of up to 18 % for dense snow of OED = 1400 µm. For all other snow types where higher reflectance values are recorded the relative reflectance variability caused by the density uncertainty is generally ≤ 5 %. These values are com-parable to the spatial variability caused by the natural snow inhomogeneity of the four snow blocks characterized in Table 4.1 and analyzed in Figure 4.6.
4.4 Conclusions and outlook
It has been shown that fast diffuse NIR reflectance measurements by a compact integrating-sphere unit, called InfraSnow, can be explained within the natural snow variability by the FRED model. The only required model input parameters are snow density and OED, derived from the true 3D snow microstructure.
The crucial point in the analysis is accounting for the exact geometry of the InfraSnow reflectance measurements. Due to the large sampling volume compared to the
integrating-sphere opening, radiation losses to the side cannot be neglected. Thus, DISORT and AFE approaches to radiative transfer do not fit the InfraSnow measurements.
In contrast to the results presented in theoretical studies by Xie et al. (2006) and Picard et al. (2009), no significant effect of grain shape is observed on the reflectance measurements in this thesis. Only a minor effect within the overall natural snow variability is imaginable. However, this possible dependance cannot be resolved in the presented analysis. This agrees with the findings of all previous experimental works on NIR snow reflectance (Matzl and Schneebeli , 2006; Painter et al., 2007; Gallet et al., 2009; Arnaud et al., 2011).
Both snow OED and density determine the sampling volume and radiation loss beyond the measurement unit. Gallet et al. (2009) also found a significant dependance on both snow OED and density for their reflectance measurements of small sampling volumes under laser-light illumination. To eliminate the impact of density on their measurements, they chose a light source at a longer NIR wavelength to reduce the penetration depth and thus minimize snow-sample boundary effects. This solution to establish a unique correlation between reflectance and OED cannot be realized for the InfraSnow as the entire integrating-sphere opening serves as diffuse illumination source. Hence, light losses beyond the integrating-sphere opening always remain and are especially crucial if a compact design with a small measuring footprint is retained.
The analysis also reveals that the main limitations to the resolution of an eventual snow OED measurement by the InfraSnow are the natural spatial snow inhomogeneity and the currently possible resolution to measure snow density.
The spatial reflectance variability across a surface area of 10 × 10 cm2is found to be up to 18 % due to the natural inhomogeneity of the snow microstructure. Tests on surface-reflecting materials, i.e. materials where no volume scattering occurs, have shown that the InfraSnow is also susceptible to specular highlights, which falsify the measurement interpretation as purely diffuse reflectance. This can explain the orientational variability of the InfraSnow reflectance measurements on snow of up to 4 %.
Knowing snow density with an accuracy of ∼ 50 kg m−3 before performing InfraSnow reflectance measurements, measured reflectance can be predicted by the FRED model within the spatial variability of snow. This demonstrates the possibility to measure snow OED within the natural spatial variability by the InfraSnow once density is determined in advance. Complementing the methods of Matzl and Schneebeli (2006); Painter et al.
(2007); Gallet et al. (2009) and Arnaud et al. (2011), the InfraSnow permits fast and simple snow-OED measurements in the field across a large snow-covered area by a compact portable device.
For the present InfraSnow design, a higher measurement accuracy can only be achieved if snow density is known more accurately. However, the current measurement accuracy found in this analysis is sufficient for use across large areas in the field, as the natural inhomogeneity of the snow microstructure across a small area of 10 × 10 cm2 already leads to a comparable or even larger reflectance variability. So, an InfraSnow reflectance measurement can be regarded as a representative reflectance value for an area of ∼ 100 cm2 within the spatial variability.
To reduce the influence of specular reflections from the snow surface, a baffle could be placed in the direct path between snow surface and detector. While this aims at reducing
4.4 Conclusions and outlook
the orientational variability, it cannot improve the practical measurement accuracy due to the greater impact of the two main limiting factors snow inhomogeneity and density measurement accuracy.
Replacing the current LED at 950 nm peak wavelength with an LED at a longer NIR wavelength reduces the sampling volume of the InfraSnow. This can lead to a smaller effect caused by the snow inhomogeneity as a more confined snow volume is probed. However, this design change also renders the InfraSnow more sensitive to isolated inhomogeneities (like a stray snow crystal on the snow surface) inside the new smaller sampling volume.
The lack of a unique correlation between measured reflectance and snow OED can also be seen as an additional opportunity: As an extension to the presented reflectance anal-ysis, a simultaneous measurement of snow OED and density is imaginable if 2 LEDs at different wavelengths are used inside the integrating sphere. If the measurement signals at both LED wavelengths offer a sufficient dynamic range to obtain a suitable measure-ment resolution, both OED and density could then be determined by the InfraSnow in combination with the FRED model.