Bringing Down Costs with Code: Measuring Snow Depth from Photographs Using R
Disciplines
Data Science | Environmental Monitoring
Abstract (300 words maximum)
The purpose of this project was to obtain the snow depth from repeat photographs taken from low-cost trail cameras, a potentially more cost-effective way to measure snow depth compared to using expensive equipment or by manually reading snow sticks. Our project attempts to use color segmentation to identify snow and track the changes to it over time. The images were taken at the Chimney Park AmeriFlux site in southeastern Wyoming between 2018 and 2020. The data obtained from the images were compared against the data from a high-precision snow depth sensor. A two-step algorithm was written in R for to obtain the snow depth from the images: The first step was to break down the photograph into two clusters of color and store the sizes of the clusters as a total percent of the photographs. The cluster that had higher levels of blue was detected as snow and had its total percent stored to be compared against the data obtained from the snow sensor. When compared to sensor data, our approach performed reasonably well with an R-squared for the full year of 2019 of 0.44. The algorithm performed the best during the snow melt in Spring of 2019 with an R-squared of 0.60. Ongoing challenges of our approach include the shifting of the trail camera throughout the year which results image misalignments, seasonal and daily changes in lighting from the sun because it changes how the camera perceives the colors, and the detection of snow when there is no snow. Nevertheless, our results so far show promise in an alternative, cost-effective method in obtaining snow depth.
Academic department under which the project should be listed
CSM - Ecology, Evolution, and Organismal Biology
Primary Investigator (PI) Name
Mario Bretfeld
Bringing Down Costs with Code: Measuring Snow Depth from Photographs Using R
The purpose of this project was to obtain the snow depth from repeat photographs taken from low-cost trail cameras, a potentially more cost-effective way to measure snow depth compared to using expensive equipment or by manually reading snow sticks. Our project attempts to use color segmentation to identify snow and track the changes to it over time. The images were taken at the Chimney Park AmeriFlux site in southeastern Wyoming between 2018 and 2020. The data obtained from the images were compared against the data from a high-precision snow depth sensor. A two-step algorithm was written in R for to obtain the snow depth from the images: The first step was to break down the photograph into two clusters of color and store the sizes of the clusters as a total percent of the photographs. The cluster that had higher levels of blue was detected as snow and had its total percent stored to be compared against the data obtained from the snow sensor. When compared to sensor data, our approach performed reasonably well with an R-squared for the full year of 2019 of 0.44. The algorithm performed the best during the snow melt in Spring of 2019 with an R-squared of 0.60. Ongoing challenges of our approach include the shifting of the trail camera throughout the year which results image misalignments, seasonal and daily changes in lighting from the sun because it changes how the camera perceives the colors, and the detection of snow when there is no snow. Nevertheless, our results so far show promise in an alternative, cost-effective method in obtaining snow depth.