Timeseries of Arctic-Boreal Lake Area Derived from CubeSat Imagery, 2017

This dataset provides near-daily lake area timeseries for 85,358 lakes across four study areas in Northern Canada and Alaska, USA, between May 1 and October 1, 2017. These lake area estimates were produced using digital images from newly developed Planet Labs CubeSats, small satellites with a 4-band (blue, green, red, near-infrared) camera payload. In constellation, CubeSats collected imagery at very high spatial (3-5m) and temporal (near-daily) resolution. From the imagery, each lake's mean, minimum, and maximum areas and seasonal dynamism were derived. The dataset covers four Arctic-Boreal regions: the Yukon Flats Basin (YFB) in eastern interior Alaska, and the Mackenzie River Valley (MRV), Canadian Shield Transect (CST), and Hudson Bay Lowland (HBL) in Canada.

The primary data product is a lake area timeseries for all lakes with a maximum area >0.01 km2. These estimates were created by (1) classifying all images into water/land, (2) tracking changes in lake area by calculating the total water contained within each lake object in a 60 m buffered mask, (3) using a supervised machine learning classification algorithm to remove poor quality observations, and (4) applying a 10-day median filter.

There are 473 total data files of three file types: four lake area timeseries files in .csv format, four .shp files of lake locations and area metrics, and 465 .tif files of buffered lake masks used to track changes in lake area. A companion .kmz file shows the image tiles at the four study areas.

This dataset provides near-daily lake area timeseries for 85,358 lakes across four study areas in Northern Canada and Alaska, USA, between May 1 and October 1, 2017. These lake area estimates were produced using digital images from newly developed Planet Labs CubeSats, small satellites with a 4-band (blue, green, red, near-infrared) camera payload. In constellation, CubeSats collected imagery at very high spatial (3-5m) and temporal (near-daily) resolution. From the imagery, each lake's mean, minimum, and maximum areas and seasonal dynamism were derived. The dataset covers four Arctic-Boreal regions: the Yukon Flats Basin (YFB) in eastern interior Alaska, and the Mackenzie River Valley (MRV), Canadian Shield Transect (CST), and Hudson Bay Lowland (HBL) in Canada.

The primary data product is a lake area timeseries for all lakes with a maximum area >0.01 km2. These estimates were created by (1) classifying all images into water/land, (2) tracking changes in lake area by calculating the total water contained within each lake object in a 60 m buffered mask, (3) using a supervised machine learning classification algorithm to remove poor quality observations, and (4) applying a 10-day median filter.

Project: Arctic-Boreal Vulnerability Experiment

The Arctic-Boreal Vulnerability Experiment (ABoVE) is a NASA Terrestrial Ecology Program field campaign based in Alaska and western Canada between 2016 and 2021. Research for ABoVE links field-based, process-level studies with geospatial data products derived from airborne and satellite sensors, providing a foundation for improving the analysis and modeling capabilities needed to understand and predict ecosystem responses and societal implications

Related Publication:

Cooley, S.W., Smith, L.C., Ryan, J.C., Pitcher, L.H. and Pavelsky, T.M. 2019. Arctic-Boreal lake dynamics revealed using CubeSat imagery. Geophysical Research Letters, 46. https://doi.org/10.1029/2018GL081584

Acknowledgements:

This work was funded by NASA ABoVE grant NNX17AC60A.

Spatial Coverage: Study areas across Northern Canada and Alaska, USA

ABoVE Reference Locations:

          Domain: Core and Extended regions

          State/territory: Alaska, USA; Canada

          Grid cell(s): ABoVE grid level C tile IDs for each of the four study areas are provided in a companion file. There are 701 level C grid tiles.

Spatial Resolution: 3.125 m

Temporal Coverage: May - October, 2017

Temporal Resolution: daily

Study Area: (all latitudes and longitudes given in decimal degrees)

Site	Westernmost Longitude	Easternmost Longitude	Northernmost Longitude	Southernmost Longitude
Yukon Flats Basin (YFB)	-149.1208528	-142.6388861	67.49692778	65.31843611
Mackenzie River Valley (MRV)	-135.59365	-129.6447639	70.3105	65.74845556
Canadian Shield Transect (CST)	-114.6404083	-110.4991444	65.57464444	62.29700556
Hudson Bay Lowland (HBL)	-97.53031944	-90.24840833	57.37728611	52.81486111

 

Data File Information:

There are 473 total data files of three file types: four lake area timeseries files in .csv format, four .shp files of lake locations and area metrics, and 465 *.tif files of buffered lake masks used to track changes in lake area. A companion .kmz file shows the image tiles in the four study areas.

Table 1.  Data file descriptions

File name	Description
YFB_ Lake_Time_Series.csv	Daily timeseries of lake areas in Yukon Flats Basin (YFB) study area. 8,854 lakes total.
MRV_ Lake_Time_Series.csv	Daily timeseries of lake areas in Mackenzie River Valley (MRV) study area. 37,957 lakes total.
CST_ Lake_Time_Series.csv	Daily timeseries of lake areas in Canadian Shield Transect (CST) study area. 21,560 lakes total.
HBL_ Lake_Time_Series.csv	Daily timeseries of lake areas in Hudson Bay Lowland (HBL) study area. 16,987 lakes total.
Yukon_Flats_Basin_Lakes.zip	Shapefile with lake centroids and lake metrics calculated from the YFB lake timeseries.
Mackenzie_River_Valley_Lakes.zip	Shapefile with lake centroids and lake metrics calculated from the MRV lake timeseries.
Canadian_Shield_Transect_Lakes.zip	Shapefile with lake centroids and lake metrics calculated from the CST lake timeseries.
Hudson_Bay_Lowland_Lakes.zip	Shapefile with lake centroids and lake metrics calculated from the HBL lake timeseries.
site-buffered_mask_tilename.tif Example:  Yukon_Flats_Basin-buffered_mask_0769609.tif	Buffered lake masks are used to track changes in lake area. Lake masks were created through a multi-step process involving classification of imagery. There is a GeoTIFF mask file for each 25 km x 25 km Planet Labs image tile – 465 total. File naming convention, where: site is the spelled-out study area and tilename is a concatenation of the 2-digit UTM Zone (06, 07, 08, 09, 11, 12, 14, 15) and the 5-digit number of the 25 km x 25 km Planet Labs image tile. See Table 4 for respective study area UTM Zones.
Companion Files
CubeSat_Buffered_Mask_Tiles.kmz	The .kmz file shows the outline of each 25 km x 25 km Planet Labs image tile – 465 total, at the 4 study areas. Each tile is identified by number along with the corresponding site-buffered_mask_tilename.tif file name.
ABoVE_grid_level_C_tile_IDs.csv	ABoVE grid level C tile IDs for each of the four study areas are provided in a companion file. There are 701 level C grid tiles.

 

Data File Details & Properties:

SITE_Lake_Time_Series.csv. User Note: Most daily values for a lake are “NaN”, indicating that no image of the lake was captured on that day so no area could be derived.

Table 2. SITE_Lake_Time_Series.csv File structure

Column name	Units/format	Description
Date	yyyy-mm-dd	Image date. May 1 – October 1, 2017. There are 154 rows.
Lake_1 Lake_2 Lake_3 ... Lake_xxxxx	Lake area in m2	Column names are the lake IDs and the values are the respective lakes' area. The total number of lakes/columns for each study area is provided in a table below.

 

The total number of lakes/columns for each study area and mean number of observations per lake.

Study area	Number of lakes/columns	Mean number of valid area observations per lake.
Yukon Flats Basin	8,854	37
Mackenzie River Valley	37,957	21
Canadian Shield Transect	21,560	32
Hudson Bay Lowland	16,987	31

 

Shapefiles: SITE_Lakes.zip

Four shapefiles (.zip), one for each study area, provide the location of the centroid of each lake and also provide seven metrics calculated from the lake area timeseries data.

User Notes: Depending on the shape of the lake, a lake’s centroid may not be contained within the lake itself. Lake IDs were assigned based on a 60 m buffered lake mask for each image tile. In some areas multiple lakes may thus be contained within the same lake ID if they are less than 60 m apart. To clarify exactly which areas are included in the lake area timeseries, use the lake masks *.tif files.

Table 3. Shapefile Attributes

Variable	Description
X	Longitude of lake centroid (in WGS84)
Y	Latitude of lake centroid (in WGS84)
Tile	Number of the Planet Labs image grid tile containing that lake
ID	Lake ID number
Number_Observations	Number of valid area observations received by that lake between May 1 and October 1, 2017
Mean_Area	Mean area of the lake (in m2) over the seasonal timeseries
Maximum_Area	Seasonal maximum observed lake area (m2)
Minimum_Area	Seasonal minimum observed lake area (m2)
Dynamism	The lake’s seasonal dynamism (m2), calculated as seasonal maximum minus seasonal minimum lake area
Percent_Dynamism	Seasonal dynamism divided by its maximum lake area (%)
Rchange	Rchange (stylized as RΔ in the manuscript) refers to the mean lateral distance (in m) between the lake’s maximum and minimum lake boundaries.

 

GeoTIFF Files: site-buffered_mask_tilename.tif

Each file is 8000 x 8000 pixels (25 km x 25 km) with a resolution of 3.125 m per pixel

Each lake within the mask is assigned a unique integer that corresponds to the ID number in the SITE_Lake_Time_Series.csv file and lake centroid .shp file.

Table 4. GeoTIFF file details for each Study Area

Study area	Number of mask files/image tiles	Projection details	EPSG
Yukon Flats Basin Lakes	79	WGS84_UTM_Zone_6N; WGS84_UTM_Zone_7N	32606; 32607
Mackenzie River Valley Lakes	145	WGS84_UTM_Zone_8N; WGS84_UTM_Zone_9N	32608; 32609
Canadian Shield Transect Lakes	75	WGS84_UTM_Zone_11N; WGS84_UTM_Zone_12N	32611; 32612
Hudson Bay Lowland Lakes	166	WGS84_UTM_Zone_14N; WGS84_UTM_Zone_15N	32614; 32615

 

Companion File Details

The CubeSat_Buffered_Mask_Tiles.kmz file shows the outline of each 25 km x 25 km Planet Labs image tile – 465 total tiles – within the 4 study areas. Each tile is identified by number along with the corresponding site-buffered_mask_tilename.tif file name.

The file ABoVE_grid_level_C_tile_IDs.csv provides the ABoVE grid level C tile IDs for each of the four study areas.

The Arctic-Boreal region contains the highest density of lakes on Earth, and variability in lake extent serves as a useful indicator of diverse climatic and physiographic processes. Observing fine-scale changes has traditionally been difficult owing to the coarse spatial and temporal resolution of available satellite imagery, but CubeSats (mini satellites) allow for daily monitoring of lake area at 3-m resolution.

The high temporal resolution of this dataset enables mapping the full seasonal progression of lake area, and the high spatial resolution enables observation of fine-scale (<10 m) fluctuations along lake margins that were previously difficult to quantify with coarser resolution satellites. In total, this dataset contains > 2.2 million individual observations of lake area. Fluctuations in Arctic-Boreal lakes reflect climate change and regulate freshwater methane and CO2 emissions. This dataset will allow one to explore temporal dynamics of Arctic-Boreal lakes and improve estimates of trace gas flux.

The CubeSat-derived lake tracking method used to produce this dataset was specifically developed for large-scale applications. Therefore, the intended use of this dataset is for assessment of large-scale patterns in lake area dynamics rather than examination of individual lake timeseries. While it may still be useful for this purpose, the authors suggest caution when interpreting individual lake timeseries (Cooley et al. 2019).

Uncertainty values for individual lake area observation are not available as quantifying individual error over such a large dataset is very difficult. However, a summary of uncertainty values is provided. The machine learning observation filtering method used on images to remove poor quality lake area observations performs at 96-99% accuracy (depending on the study area), therefore inclusion of poor quality data is likely to be minimal. An uncertainty analysis was performed on lake metrics calculated from each timeseries. In brief, the normalized RMSE values for the lake metrics reported (namely mean lake area, maximum lake area, minimum lake area and seasonal dynamism) range from 1,248 m2 to 23,135 m2, or 0.6-3.6% of the maximum lake areas. A full discussion of sources of uncertainty can be found in the corresponding manuscript (Cooley et al., 2019).

This dataset covers four Arctic-Boreal regions: the Yukon Flats Basin (YFB) in eastern interior Alaska, and the Mackenzie River Valley (MRV), Canadian Shield Transect (CST), and Hudson Bay Lowland (HBL) in Canada. These sites were chosen to cover diverse physiographic and climatic terrains and are co-located with targeted areas of NASA’s Arctic-Boreal Vulnerability Experiment (ABoVE).

 

Figure 2. Location of four study areas within the ABoVE study domain.

The primary products of the dataset are the lake area timeseries created from near-daily CubeSat (Planet Team, 2018) imagery for all lakes with a maximum area >0.01 km2. These timeseries are created by (1) classifying all images into water/land based on a local Normalized Difference Water Index (NDWI) histogram threshold, (2) tracking changes in lake area calculating the total water contained within each lake object in a 60 m buffered mask, (3) using a supervised machine learning classification algorithm to remove poor quality observation, and (4) applying a 10 day median filter. A complete description of the methods can be found in Cooley et al. (2019).

 

Figure 3. Example of an individual lake area timeseries between May 1 and October 1, 2017.

 

Figure 4. Lake centroid shapefile for the Yukon Flats Basin, Alaska. Each lake is represented by an individual circle (not to scale) and is colored by the lake’s percent dynamism (maximum – minimum seasonal lake area, divided by the maximum lake area). The Yukon River and its major tributaries are shown as blue dashed lines.

 

Changes in lake area extent were measured by applying a 60 m buffered lake mask to the classified images. This object-based method of lake tracking reduces the impact of geolocation error (Planet CubeSat’s stated geolocation accuracy is ~10m) and enables efficient detection given the large data volumes. Buffered lake masks were created through a multi-step process involving initial classification of all imagery between June 1 and July 15, 2017. The final mask is defined as all areas classified as water in at least 30% of images and is then buffered by 60 m. Each mask was manually checked to ensure that all lakes were included and all rivers were removed. Given the large size of the dataset, it is possible that some lakes have been missed or some small rivers reaches accidentally included. Lake area masks are not intended to represent the actual maximum boundary of a given lake; rather they represent each lake’s early season (typically maximum) extent plus an additional 60 m buffer. Additionally, the mask tiles overlap, so overlapping areas were removed to ensure that water bodies are not included in multiple masks.

Figure 5. Example of a buffered lake mask, overlaid on an Normalized Difference Water Index (NDWI) image where lakes are white. Notice the 60 m buffered extending out from the lake margins.

The CubeSat-derived lake tracking method used to produce this dataset was specifically developed for large-scale applications. Therefore, the intended use of this dataset is for assessment of large-scale patterns in lake area dynamics rather than examination of individual lake timeseries. While it may still be useful for this purpose, the authors suggest caution when interpreting individual lake timeseries.

Cooley, S.W., Smith, L.C., Ryan, J.C., Pitcher, L.H. and Pavelsky, T.M. 2019. Arctic-Boreal lake dynamics revealed using CubeSat imagery. Geophysical Research Letters, 46. https://doi.org/10.1029/2018GL081584

Planet Team. (2018). Planet application program interface: In space for life on Earth. San Francisco, CA. Retrieved from https://www.planet.com