Report details
Bibliographic Data
Title:
Snow Distribution Statistical Modelling and UAV-borne Remote Sensing of Snow Reflectance in the Arctic
Authors:
Year:
2015
Report no.:
15-07
Topics:
Geoteknik,
Energi
Keywords:
PDF file:
Supervisors:
Abstract:
The present study is part of a project that aims at mapping snow depth and snow optical
properties using Unmanned Aerial Vehicle (UAV) -borne digital photogrammetry. It is organized
in two parts, the rst part aims at estimating snow distribution from snow depth point
measurem....
The present study is part of a project that aims at mapping snow depth and snow optical
properties using Unmanned Aerial Vehicle (UAV) -borne digital photogrammetry. It is organized
in two parts, the rst part aims at estimating snow distribution from snow depth point
measurements by means of regression tree models, and the second at extracting Hemispherical
Directional Re
ectance Factor of snow from images taken by a digital re
ex camera xed to a
UAV.
The objective of the rst part is to assess the capacity of regression tree models and of a set
of predictors to distribute snow over an area from point measurements and at testing whether
they could be used in the future for high resolution snow depth data acquired by UAV-borne
photogrammetry. In four sites around Longyearbyen in Svalbard, Norway, snow depth was
measured by manual probing or Ground Penetrating Radar. Then a regression tree model was
built for each site and for two Digital Elevation Models (5 m and 50 m resolution) to predict
snow depth from elevation slope, two wind exposure indexes Sx and Sb, and a radiation index.
For all sites except one the 5 m models showed higher R2 than the 50 m ones indicating that
regression trees successfully model snow depth at a small scale. The 5 m snow distribution
models give R2 ranging from 0.39 (4 snow classes) to 0.67 (2 snow classes) which is satisfactory
compared to previous studies using regression trees. Most snow distribution models were found
to be site speci c and could not be applied on other sites. It was addressed to di erences of
snow distribution processes or to the e ect of di erent snow depth sampling methods. The snow
distribution model built in Longyearbyen and snow density data were used to estimate total Snow
Water Equivalent (SWE) stored in the snow cover of Longyearbreen glacier. The regression tree
model estimated a total SWE value of 8:13 2:90 105 m3 and showed an improved precision
when compared to the results of a constant snow depth model and of an elevation gradient
model. Errors and limitations of the present modelling framework are discussed and suggestions
are given for further studies.
In the second part, a method to estimate snow re
ectance from UAV-borne digital images
is investigated. The approach presented, uses the digital camera to measure luminance in a
scene and then estimates the re
ectance of a surface by comparing its luminance to the one
of a surface of known re
ectance in the picture. The calibration of the camera simply relates
the camera response to variations in a target brightness, nding a linear relationship between
luminance and pixel intensity. It has the advantage of requiring no speci c equipment and to
avoid the complex calculation of camera spectral sensitivity. Vignetting e ect was corrected
The camera accuracy is assessed by investigating shot noise and dark current. In the eldwork
conditions, shot noise is found to be the most relevant source of noise and sucient exposure
of the pictures is recommended to reduce its e ect. The method was tested during a eldwork
campaign nearby Sisimiut, West Greenland. The UAV is
own vertically up to 7.6 meters and
17 pictures are used to estimate the the re
ectance of a snow patch and of a 10% re
ective
validation surface. The estimated re
ectance values of the snow patch range from 0.97 to 0.98
with an uncertainty of 0.04 to 0.08. This data could not be compared to other eld measurement
because of human mistake in the spectroradiometer measurement of the snow patch re
ectance.
The estimation of the validation surface re
ectance showed that a minimum number of pixels
are necessary to determine accurately the re
ectance of a surface. Based on a rst estimation
of this threshold, the estimation of the validation surface re
ectance gives a Root Mean Square
Error of 0.05 which is relatively high compared to the surface real re
ectance. It is attributed to
the low brightness of the validation surface which increases noise. The necessity of a reference
surface in the picture is discussed and suggestions for future work are given.
Appendices:
No appendices available.
Features & Map
GEOPHYSICAL-DATA
FIELD-MEASUREMENT
RESOURCE
PHOTO
SAMPLE
OTHER
BOREHOLE
LAB MEASUREMENT
| ID | Name | Type | Date | Description | |
|---|---|---|---|---|---|
| No features available. | |||||