The SandSnap Project: 2020 -- 2021 sieved grain-size data and associated sediment imagery

Overview

SandSnap is a collaborative project engaging citizen scientists to amass a sand beach grain size database and educating the next generation about coastal processes. See the following webpage for more details: https://sandsnap-erdcchl.hub.arcgis.com/  

SandSnap is funded by the US Army Corps of Engineers through the Coastal Inlets Research Program and the Regional Sediment Management Program.

SandSnap allows anyone with a cell phone to take an image of the sand with a US coin and measure the sand’s grain size using a deep learning neural network (Buscombe, 2020; McFall et. all, 2020). This model is trained using data obtained from sieved physical samples of sand. The purpose of this data release is to document the data sets that went into the SandSnap model, trained in Aug 2021, and used between August 2021 ongoing on this date October 26 2022.

Data formats and fields

usace_1024_aug_dry_set1_2_3_4_5_aug2021.csv

This is a spreadsheet that contains inputs for training the SandSnap SediNet model. SediNet is a deep-learning-based grain size predictor, by Dr Daniel Buscombe, Marda Science, LLC (https://github.com/DigitalGrainSize/SediNet). The SediNet model behind SandSnap v1 (August, 2021) is configured to estimate the grain size in pixels. A separate model is used to detect and size the coin, to estimate image scaling for grain size estimates in millimeters.

• File: name of image
• Latitude: WGS84 coordinate
• Longitude: WGS84 coordinate
• Population: an integer, identifying the site that the image came from. For internal model validation purposes (grouping error by site)
• dry: 0= visibly wet sand, 1= visibly dry sand    
• mm_px: millimeter per pixel scaling, computed from digitizing a coin in each image, as the diameter of the coin in millimeters, divided by the number of pixels across the diameter of the coin
• d10: 10th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters
• d16: 16th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters  
• d25: 25th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters
• d50: 50th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters    
• d65: 65th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters    
• d75: 75th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters    
• d84: 84th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters    
• d90: 90th percentile of the cumulative grain size distribution, obtained by sieve analysis, in millimeters    
• mean: mean grain size, obtained by sieve analysis, in millimeters       
• P10: 10th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• P16: 16th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• P25: 25th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• P50: 50th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• P65: 65th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• P75: 75th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• P84: 84th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• P90: 90th percentile of the cumulative grain size distribution, obtained by sieve analysis, in pixels
• Pmean: mean grain size, obtained by sieve analysis, in pixels

GrainSizeAdditionalImagesTraining_Aug2021_latlong.xlsx assigns a coordinate to imagery and contains the following fields:

*.zip format files

Zipped folders contain original images, as well as augmented and tiled images for analysis. Tiled images are patches of original images with no coin scale. Patches are 1024 x 1024 x 3 pixels. Augmented images are tiles that have been flipped in both horizontal dimensions.

*.py format files

Python code for creating tiled and augmented images

References

• Buscombe, D., 2020. SediNet: A configurable deep learning model for mixed qualitative and quantitative optical granulometry. Earth Surface Processes and Landforms, 45(3), pp.638-651.
• McFall, B.C., Young, D.L., Fall, K.A., Krafft, D.R., Whitmeyer, S.J., Melendez, A.E. and Buscombe, D., 2020. Technical Feasibility of Creating a Beach Grain Size Database with Citizen Scientists. ERDC Coastal and Hydraulics Laboratory.