Skip to main content
NCBI home page
EPA Author Manuscripts logoLink to EPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Jan 5.
Published in final edited form as: Int J Remote Sens. 2019 May 1;40(15):5768–5798. doi: 10.1080/01431161.2019.1582112

The potential role of very high-resolution imagery to characterise lake, wetland and stream systems across the Prairie Pothole Region, United States

Melanie K Vanderhoof a, Charles R Lane b
PMCID: PMC7784670  NIHMSID: NIHMS1528607  PMID: 33408426

Abstract

Aquatic features critical to watershed hydrology range widely in size from narrow, shallow streams to large, deep lakes. In this study we evaluated wetland, lake, and river systems across the Prairie Pothole Region to explore where pan-sharpened high-resolution (PSHR) imagery, relative to Landsat imagery, could pro-vide additional data on surface water distribution and movement, missed by Landsat. We used the monthly Global Surface Water (GSW) Landsat product as well as surface water derived from Landsat imagery using a matched filtering algorithm (MF Landsat) to help consider how including partially inundated Landsat pixels as water influenced our findings. The PSHR outputs (and MF Landsat) were able to identify ~60–90% more surface water interactions between waterbodies, relative to the GSW Landsat product. However, regardless of Landsat source, by doc-umenting many smaller (<0.2 ha), inundated wetlands, the PSHR outputs modified our interpretation of wetland size distribution across the Prairie Pothole Region.

1. Introduction

Surface-water spatial and temporal dynamics, such as the expansion and contraction of lake systems, the filling, spilling, and merging of wetlands (Leibowitz, Mushet, and Newton 2016), or the lateral and longitudinal evolution of stream channels (Fritz et al. 2018) have critical effects on watershed hydrology (Ward 1989), biogeochemical processes (Covino 2017), and habitat functions (Schofield et al. 2018). Satellite imagery provides a key tool to characterise these interactions over spatial scales and time (Alsdorf, Rodriguez, and Lettenmaier 2007). For example, satellite imagery is commonly used to document varia-bility in wetland or lake surface-water extent (Sheng et al. 2016; Jin et al. 2017) and flood stage along large rivers (Schumann et al. 2010; Ogilvie et al. 2015). The appropriate choice in image spatial resolution is generally agreed as the resolution at which local variance peaks (Woodcock and Strahler 1987). However, because hydrologic features that play critical roles informing watershed hydrology occur across such a wide range of object sizes (e.g. small wetlands and narrow streams to large lakes), it is more likely that a single appropriate image resolution for water resource applications does not exist. For instance, whereas Landsat imagery can be expected to monitor large features with ease (e.g. lakes and large wetlands), it can also be expected to miss smaller features less than its minimum mapping unit, previously shown to be approximately 9 to 11 pixels (<1 ha) (e.g. most rivers and streams, small wetlands) (Ozesmi and Bauer 2002; Knight and Lunetta 2003; Lechner et al. 2009). Consequently, a reliance on Landsat imagery may provide only a partial understanding of surface water distribution and interactions. As commercial high-resolution imagery becomes increasingly available we can explore how the spatial resolution of imagery influences our interpretation of surface water characteristics such as surface water distribution and movement.

Surface-water extent has been remotely monitored using imagery from a wide range of satellites including the moderate-resolution imaging spectroradiometer (MODIS, 1 km resolution, Schumann et al. 2010; Chen et al. 2013), Landsat (30 m resolution, Huang et al. 2011; Pekel et al. 2016), Sentinel-1 (Huang et al. 2018), commercial high resolution (2–5 m resolution, White and Lewis 2011; Whiteside and Bartolo 2015; Cooley et al. 2017), lidar (1–3 m resolution, Lang and McCarty 2009; Wu and Lane 2017), and synthetic aperture radar (SAR) (3–60 m resolution, Schmitt and Brisco 2013; Bolanos et al. 2016). Landsat surface-water products at national and global extents are becoming increasingly available. These efforts include the Global Surface Water (GSW) Landsat product produced using Google Earth Engine (Pekel et al. 2016) and USGS’s Dynamic Surface Water Extent (DSWE) product (Jones 2015) both of which aim to use the entire Landsat archive to map surface water on a global and continental United States (CONUS) scale, respectively.

High accuracy statistics are commonly reported for such water products (e.g. Du et al. 2014; Mueller et al. 2016; Pekel et al. 2016). This is likely because, although most users understand the detectable limits of Landsat (e.g. Ozesmi and Bauer 2002; Knight and Lunetta 2003; Lechner et al. 2009), accuracy statistics generated using random sampling can be expected to reflect the size distribution of the waterbodies, which can potentially obscure the spatial distribution of error. For example, if a large lake comprises 90% of the surface water across the watershed, we can expect that approximately 90% of the validation points will occur within the lake. However, if the lake flows into a major river via a narrow conveyance, this single narrow channel may have a disproportionate impact on the hydrology of the watershed (e.g. Shaw, Pietroniro, and Martz 2013; Spence and Phillips 2014), but the ability of a remote sensing product to monitor such a narrow conveyance is unlikely to be reflected in commonly reported accuracy statistics. Although the effect of object size on accuracy in remote sensing is intuitive, efforts to quantify this phenomenon across different remote sensing applications are very limited in the literature (e.g. Castilla et al. 2014), and its effect on our interpretation of surface water patterns has not been characterised to date.

In this study we compared how our interpretation of surface water distribution and surface water movement between waterbodies were informed by using pan-sharpened high-resolution imagery (0.5 m resolution) relative to Landsat imagery (30 m resolution) across landscapes dominated by (1) wetlands, (2) lakes, and (3) rivers (seven sites total). We selected the Prairie Pothole Region as our area of interest as (1) the region exhibits substantial interannual variability in surface water extent necessitating the need for routine surface water monitoring (Huang et al. 2011; Beeri and Phillips 2007), (2) other barriers to remotely evaluating surface-water (e.g. trees, topographical shade) are limited, enabling us to isolate the interaction between imagery and aquatic system configurations, and (3) the median size of wetlands in the region is only 0.16 ha (Wu and Lane 2017), less than two Landsat pixels, meaning moderate resolution satellites may be inadequate. Whereas the PSHR images (0.5 m resolution) can still be expected to underestimate surface water, especially narrow features (e.g. swales, ephemeral channels), it currently represents the ‘cutting edge’ of our ability to remotely detect water at a landscape scale (Amro et al. 2011) (Figure 1). We explored the research question, where can high-resolution imagery, relative to Landsat imagery, enhance our understanding of surface water distribution and movement across the Prairie Pothole Region?

Figure 1.

Figure 1.

Open in a new tab

Comparisons of a river using (a) Landsat TM imagery (30 m) versus (b) Worldview-2 imagery (2 m), and a zoomed-in portion of the river using (c) Worldview-2 versus (d) pan-sharpened Worldview-2 imagery (0.5 m). The river is a tributary to the Niobrara River in northern Nebraska. Comparisons of a wetland using (e) Landsat TM imagery versus (f) Worldview-2 imagery, and a zoomed-in portion of the wetland using (g) Worldview-2 versus (h) pan-sharpened Worldview-2. The wetland is adjacent to the Waubay National Wildlife Refuge in South Dakota. Each comparison is shown at identical scales. Copyright 2006 and 2012 Digital Globe, Next View License.

2. Methods

2.1. Study area

The Prairie Pothole Region (PPR) extends from Saskatchewan and Alberta, Canada, south into the upper Midwestern United States. Within the United States, the PPR extends across parts of North Dakota, South Dakota, Montana, Minnesota, Iowa and Nebraska (Figure 2). The PPR is part of the temperate prairie and west-central semi-arid prairie ecoregions (Level II, Omernik and Griffith 2014). The region is characterised by landscape features formed during its recent glacial history, including drift plains, large glacial lake basins and shallow river valleys. These portions of the PPR support row crop agriculture (59%). Areas where glaciers left deposits of uneven glacial till are dominated by grasslands (18%) and livestock grazing (10%) (Homer et al. 2015). Mean annual precipitation (1981–2010) across the selected sites noted below ranged from 490 to 635 mm yr−1 (PRISM Climate Group 2012). Mean temperature maximum and minimum (1981–2010) across the selected sites range from 9.6 to 15.8°C and −1.9 to 2.6°C, respectively (PRISM Climate Group 2012).

Figure 2.

Figure 2.

Open in a new tab

Distribution of sites representing wetlands, lakes, and rivers across which surface-water interactions and mapped surface-water extent were characterised using locally trained Landsat and globally trained Landsat, with high-resolution imagery used as the reference dataset. Copyright 2006 and 2012 Digital Globe, Next View License.

2.2. Image and site selection

Sites (n = 7) were selected across the PPR to represent different lentic and lotic systems including (1) wetlands, (2) lakes, and (3) rivers (Figure 2). Site selection was guided in part by the representativeness of a site and in part by the availability of high resolution imagery acquired from DigitalGlobe (Westminster, CO). A total of 19 high-resolution images (Worldview-2, QuickBird-2, GeoEye-1) and 13 Landsat images (TM, ETM+, OLI) were processed (Table 1). Images were restricted to snow-free conditions (May-October). When possible, we processed two high-resolution images per site representing different dates to account for the potential influence of climatic and antecedent conditions. A corresponding Landsat image selected by nearest date and cloud-free (<10% cloud for the portion overlapping the high-resolution image extent) was compared with the high-resolution images. The date gap between the high-resolution and Landsat images averaged 16 days (ranged 4 to 40 days) with no average bias in the relative timing of the high-resolution images versus the Landsat images (<1 day).

Table 1.

Images utilised in the analysis by site and wetness index for the month preceding the image dates. A total of 19 high-resolution images were processed relative to 13 Landsat images and 13 Global Surface Water (GSW) monthly product rasters. TM: Thematic Mapper, ETM+: Enhanced Thematic Mapper plus, OLI: Operational Land Imager, PHDI: Palmer Hydrological Drought Index, SP24: 24 month standardized precipitation index.

Site High-Resolution Image
Date		 Image
Count		 Satellite Landsat Image Date Satellite Landsat path/
row		 Gap
(Days)		 GSW
(month)		 PHDI SP24
Cottonwood Lake study area wetlands 29 August 2011 1 Worldview-2 7 September 2011 TM p31r27 9 Sept 2011 9.93 2.51
24 July 2015 1 Worldview-2 August 2015 OLI p31r27 7 Aug 2015 3.76 1.47
Waubay National Wildlife Refuge wetlands 10 October 2004 1 QuickBird-2 29 September 2004 TM p29r29 −11 Sept 2004 3.53 0.27
30 April 2012 1 Worldview-2 3 April 2012 ETM+ p29r29 −27 Apr 2012 2.35 1.15
Waubay National Wildlife Refuge lakes 7 June 2012 3 Worldview-2 1 July 2012 ETM+ p29r29 24 Jun 2012 1.75 0.99
8 June 2015 3 Worldview-2 18 July 2015 OLI p29r29 40 Jul 2015 1.76 0.25
Barnes Lake 30 July 2011 2 Worldview-2 5 July 2011 TM p31r27 −25 Jul 2011 8.91 2.20
30 May 2015 2 GeoEye-1 14 June 2015 OLI p31r27 15 Jun 2015 3.62 1.24
Horseshoe Lakes 15 April 2015 1 Worldview-2 11 April 2015 OLI p31r27 −4 Apr 2015 −1.37 0.20
Sheyenne River 21 April 2003 1 QuickBird-2 26 April 2003 TM p31r27 5 Apr 2003 −2.02 −0.56
15 April 2015 1 Worldview-2 11 April 2015 OLI p31r27 −4 Apr 2015 −1.37 0.20
Niobrara River 15 August 2006 1 QuickBird-2 16 July 2006 TM p30r30 −30 Jul 2006 −3.86 −0.07
30 June 2010 1 Worldview-2 25 June 2010 TM p30r30 −5 Jun 2006 7.59 2.26
Open in a new tab

Sites with a high density of wetlands were represented by the USGS Cottonwood Lake study area (and nearby wetlands) (47°7’43” N, 99°8’46” W), near Jamestown, North Dakota and wetlands adjacent to the Waubay National Wildlife Refuge (Waubay NWR wetlands) (45°27’35” N, 97°31’59” W) (Figure 2). When wetlands were defined using the National Wetlands Inventory (NWI) dataset (USFSW 2010), the Cottonwood Lake site showed a high density of very small (<0.2 ha) wetlands (7.8 to 9.2 wetlands per km2 for the 2011 and 2015 image extents, respectively), as well as a high density of small (0.2 to 1.0 ha) wetlands (3.7 and 4.7 wetlands per km2 for the 2011 and 2015 image extents respectively). The Waubay National Wildlife Refuge wetland site showed a density of 4.5 to 4.8 wetlands per km2 of very small (<0.2 ha) wetlands, for the 2004 and 2012 image extents, respectively, and a density of 2.3 and 2.5 wetlands per km2 of small (0.2 to 1.0 ha) wetlands for the 2004 and 2012 image extents, respectively. The Cottonwood Lake study area was also included because of the availability of field-collected water level data that was used to validate both the Landsat and high-resolution imagery outputs (see section 2.6). For lakes, we were interested in how the sources of imagery differentially mapped lake expansion, so lakes that exhibited an expansion of extent as conditions became wetter were intentionally selected (Vanderhoof, Alexander, and Todd 2016). The Waubay NWR lakes were selected as the area contains multiple large lakes including Waubay Lake, Spring Lake, Rush Lake, Blue Dog Lake, and Bitter Lake (Table 2). Additional lake sites included Barnes Lake in North Dakota, west of Arrowwood National Wildlife Refuge, Horseshoe Lake, and a third lake (unnamed) just southwest of Horseshoe Lake (Table 2) (to be referred to as the Horseshoe Lakes). The latter two lakes are located just north of the Sheyenne River in North Dakota. River sites include a section of the Niobrara River in northern Nebraska (25 km reach, 42°52’13” N, 98°55’29” W) and the Sheyenne River in North Dakota (39 km reach, 47°50’43” N, 98°56’4” W), as well as the tributaries present within the image extent were used to represent our river sites (see Figure 2). All visibly wet (as observed using the raw pan-sharpened high-resolution images) rivers and streams, as defined by the High Resolution National Hydrography Dataset (NHD) (USGS 2013), within the image extents were included in the river analysis. Across the study areas, static data layers (e.g. the NWI and NHD) identified a total of 8911 wetlands, totaling 8843 ha, 8 lakes totaling 4294 ha, and 23 streams and rivers with a total length of 142 km.

Table 2.

The name, size and location of the lakes analysed per site in the analysis. Lake sizes were calculated using the National Wetlands Inventory lacustrine polygons (USFWS, 2010).

Site and Lake Size (ha) Location
Waubay National Wildlife Refuge lakes
Waubay Lake 1803 45°25’27” N, 97°23’8” W
Spring Lake 361 45°24’45” N, 97°19’30” W
Rush Lake 436 45°20’32” N, 97°21 ‘56” W
Blue Dog Lake 479 45°21’5” N, 97°17’47” W
Bitter Lake 737 45°16’39” N, 97°19’4” W
Barnes Lake
Barnes Lake 210 47°15’56” N, 99°16’29” W
Horseshoe Lake
Horseshoe Lake 219 47°51’14” N, 98°47’31” W
unnamed lake 49 47°49’3” N, 98°49’56” W
Open in a new tab

2.3. High-resolution image processing

The high-resolution images were classified as water and non-water. Climate condi-tions at the time of the image collection were characterised using a monthly Palmer Hydrological Drought Index (PHDI) and the Standardized Precipitation index over 24 months (SP24) (NOAA NCDC 2014). Wetter images were selected for inclusion as available so that surface-water expansion would be potentially present on the land-scape. A list of the high-resolution images (Worldview-2, QuickBird-2, GeoEye-1) used in the analysis is shown in Table 1. The images were converted to surface reflectance using Fast Line-of-sight Atmospheric Analysis of Hypercubes (FLAASH) in ENVI (Boulder, CO) (Adler-Golden et al. 1998, 1999). The high-resolution images (Worldview-2, QuickBird-2, GeoEye-1) were delivered at 2 m resolution. We used the Gram-Schmidt pan-sharpening algorithm (Laben and Brower 2000) to pan-sharpen each spectral band to 0.5 m resolution. Surface water was identified using the Maximum Likelihood (ML) supervised classification, a popular statistical classifier (e.g. Lin et al. 2015). We tested multiple classification approaches including the Matched Filtering algorithm but found that ML out-performed other techniques. If water was spectrally diverse across the image (i.e. clear, dark water, shallow, turbid water, water with high chlorophyll content), then multiple categories of water were classified per image (for example (1) dark water, (2) algae-covered water, and (3) turbid water) in addition to classifying other non-water classes including (4) photo-synthetic vegetation, (5) non-photosynthetic vegetation, and (6) bare soil. This approach reduced the spectral heterogeneity within each of the water classes whilst enabling the spectral signature of water to be image specific. A Frost Filter with a window size of 3 pixels by 3 pixels was applied to the classified image to reduce speckling both within and between waterbodies. The classes were then reclassified to a binary classification (water, not water). Roads were a common source of surface water extent disruption captured by the high-resolution imagery. To account for this, following image binary classification, if the water body was continuous on both sides of the road we assumed a culvert was present and manually edited the surface-water extent to extend across the road. This manual step was taken to improve the accuracy of our efforts to quantify surface water movement between water bodies. This output will be referred to as the PSHR imagery.

2.4. Landsat image processing

The purpose of this analysis was not to generate novel image processing approaches, but instead to characterise the relative impact of products on our interpretation of surface water distribution and interactions. To this end, we used a Landsat image processing approach previously applied across the PPR (Vanderhoof, Alexander, and Todd 2016; Vanderhoof et al. 2018). The images were delivered from the U.S. Geological Survey having been converted to surface reflectance using the Landsat Ecosystem Disturbance Adaptive Processing System (Masek et al. 2006). Landsat ETM+ images with the scan-line corrector off were gap-filled using a Gapfill plug-in tool (triangulation method) in ENVI (version 5.3, Harris, Melbourne, Florida). A minimum noise fraction transformation tool was used to reduce random noise across the image (Green et al. 1988). The per pixel fraction water was identified using the Matched Filtering algorithm in ENVI (Turin 1960; Vanderhoof, Alexander, and Todd 2016). This algorithm is ideal for cases in which the end user is only interested in one or a few target endmembers (e.g. water). The algorithm quantifies the presence of the target endmember whilst ignoring the unknown composite background (Frohn et al. 2012). This algorithm is locally trained on water spectral signatures derived from open-water polygons manually selected within each path/row, resulting in a water signature specific to each image. Three to four polygons (minimum size of 1 ha per polygon) per path/row were selected, with approximately 20 ha total training area per path/row combination. The output values were linearly stretched to maximize the spread of pixel values, enhancing our ability to distinguish water from non-water pixels. A per-pixel threshold value of 0.3 was selected to discern and bin water (≥0.3) and non-water (<0.3) pixels, and has been previously shown to allow wetlands larger than 0.2 ha to be reliably (defined as >70% of the features of that size class) detected in the Prairie Pothole Region (Vanderhoof et al. 2018). The National Land Cover Database (NLCD)(2011) was used to mask out impervious surfaces, defined as low, medium and high-density development (Homer et al. 2015), which can show spectral confusion with surface water. This output will be referred to as the MF (matched filtering) Landsat.

2.5. Global surface water Landsat product

In addition to the MF Landsat, the monthly GSW Landsat product (Pekel et al. 2016) was also included to provide a sense of the amount of variability that can occur between Landsat products. This product was generated using a machine learning approach and applied to the entire Landsat archive (1984–2015) and has reported errors of omission of <5% and errors of commission <1% when validated at a global scale (Pekel et al. 2016). As described by Pekel et al. (2016), water was classified by first generating class cluster hulls that described the distribution of water spectral signatures from Landsat. The visual analytic clusters were then used to guide a procedural sequential decision tree using Landsat spectral bands and ancillary data layers. For pixels that were not unambiguously assigned, evidential reasoning was used to guide class assignment using consideration of geographic location and temporal trajectory (Pekel et al. 2016). Additional details in the image processing for the GSW Landsat product are provided in Pekel et al. (2016). We downloaded the monthly rasters from Google Earth Engine (image collection: JRC Monthly Water History v1.0 (1984–2015)). Monthly rasters were selected to ensure that the same Landsat images processed by us (MF Landsat) were also processed by the GSW Landsat product (Table 1). Because the rasters are delivered from Google Earth Engine as binary water/not water, we were unable to gap-fill months comprised only of Landsat-7 ETM+ SLC-off images (n = 2). Where possible, we visually interpolated these data gaps (e.g. correcting where large lake features were continuous on either side of data gaps). Because the Landsat 7 ETM+ (scan-line corrector off) represents a substantial portion of the Landsat archive we felt it was important to consider its influence on our ability to characterise surface water.

2.6. Surface-water extent validation

The binary (water and non-water) images were validated using field-collected water level data. Water levels (one per wetland) are regularly measured at a series of tempor-ary and permanent wetlands located at the Cottonwood Lake study area near Jamestown, North Dakota (Mushet, Rosenberry, and Solensky 2016). The water levels were converted to surface-water extent using a lidar digital elevation model (DEM) (1 m resolution) that included bathymetry of waterbodies for which lidar did not detect the feature bottom (Mushet, Roth, and Scherff 2017). Water levels were identified for the date closest to the date of image collection (Table 3). At each water level measurement point, the water depth (e.g. 2 m) was added to the bathymetry DEM to create a water elevation raster. The elevation at the point of measurement was then used to reclassify the raster where elevations lower than the water depth were classified as water. Because the elevation of the water surface is likely to be inconsistent across space, the surface-water extent was generated individually for each wetland (and water-level measuring point). Water validation points were randomly selected across the generated surface-water extent, and non-water validation points were randomly selected from within a buffer (between 3 m and 30 m from the water edge for the PSHR imagery, between 30 m and 100 m from the water edge for Landsat (MF and GSW)) around the wetland features (Figure 3). Validation points (100 water, 100 non-water) were independently generated for each imagery resolution and for each of two dates (29 August 2011 and 24 July 2015), providing a total of 200 water points and 200 non-water points for validation of each classified image (PSHR images, MF Landsat, GSW Landsat product). Validation points were separated by a minimum of 2 m for the PSHR imagery and 30 m for the Landsat products. Metrics presented included overall accuracy, omission error, commission error, dice coefficient, and relative bias. Omission and commission errors were calculated for the category ‘water’. The dice coefficient is the conditional probability that if the reference dataset identifies a pixel as water, the output product will as well, integrating errors of omission and commission (Fleiss 1981). The relative bias is the proportion that surface-water extent is under (negative) or overestimated (positive) relative to the total surface-water extent in the reference data (Bainbridge 1985).

Table 3.

The accuracy of the pan-sharpened high-resolution surface-water outputs validated using surface-water level data collected at wetlands within the Cottonwood Lake study area, North Dakota.

Sensor Image Date Field Collected
Water Level Date		 Omission
Error (%)		 Commission
Error (%)		 Overall
Accuracy
(%)		 Dice
Coefficient
(%)		 Relative
Bias (%)		 Water Level
Wetland Count		 Average
wetland size
(ha)		 Wetland size
range (ha)		 Wetlands
Mapped
Count
Worldview-2 29 August 2011 26 August 2011 11 0.0 94.5 94.2 −11 8 4.2 0.3 to 17.6 7
Worldview-2 24 July 2015 27 July 2015 13 1.1 93.0 92.6 −12 8 3.8 0.3 to 16.3 7
Worldview-2 Averaged Averaged 12 0.6 93.8 93.4 −12 ~ ~ ~ ~
Open in a new tab

Figure 3.

Figure 3.

Open in a new tab

An example (27 July 2015) of the surface-water extent derived from the Cottonwood Lake study area water level measurements and the corresponding randomly selected validation points used to validate the high-resolution surface-water extent. P wetland codes refer to historically permanent wetlands and T wetland codes refer to historically temporary wetlands.

Using the PSHR surface-water extent as our reference dataset, additional accuracy statistics were also generated for the Landsat products (MF Landsat and GSW Landsat) by randomly generating points, (1) across all sites (6000 water points, 6000 non-water points), as well as (2) independently generated points for each aquatic system, with 200 water points and 200 non-water points from each image extent and aquatic system (n = 1600 for wetlands, n = 2000 for lakes, n = 1600 for rivers). To generate the validation points across all sites, the PSHR surface-water extent was used for one date per site (Cottonwood – 2011, Wildlife Refuge wetlands – 2004, Wildlife Refuge lakes – 2015, Barnes Lake – 2011, Horseshoe Lake – 2015, Sheyenne River – 2015, Niobrara River – 2006, Table 1).

2.7. Surface-water extent comparison

Surface-water extent mapped by the PSHR imagery was compared to surface-water extent mapped by the MF Landsat and the GSW Landsat product in each aquatic system. The NWI wetland dataset, which is meant to represent wetland extent under median climate conditions (U.S. Fish and Wildlife Service (USFWS) 2010), was used to define individual wetlands in reference to the mapped surface-water extent. The wetland dataset has only been partially updated so that according to the metadata, the wetland extent was generated between 1977–2016. Wetland boundaries, internal to a single, continuous NWI wetland, were dissolved. The co-occurrence of more than one dissolved NWI wetland or lacustrine feature within a single, continuous surface-water polygon was used to quantify the total number of surface water interactions and the creation of wetland clusters. For example, the merging of three wetlands into a single continuous surface-water area creates one wetland cluster and three wetland-to-wetland surface water interactions. Each Landsat path/row output was clipped to the extent of its corresponding high-resolution image to ensure that the same area was being compared. This meant that for each site, an image date represented a unique extent across that site. Consideration of surface-water features was restricted to the type of aquatic system being represented in that image. For example, we excluded lake and river features when assessing wetland hydrologic characteristics. Similarly, in images representing lakes, only surface water continuous with NWI lacustrine features were considered, whereas in images representing river systems, only streams and river channels were considered.

For wetland systems, we compared the (1) total surface-water extent, (2) number of inundated NWI wetlands, (3) total count of NWI wetland-to-NWI wetland surface-water interactions, and (4) the number of clusters created by NWI wetlands merging. For lakes we compared the number of NWI features (wetlands and lakes) detected within continuous lake perimeters as well as the total continuous lake surface-water extent. In river systems we compared the percent of the river mapped by river width and data source. The High Resolution NHD (1:24,000) was used to define streams (USGS 2013). All high-resolution NHD lines overlapping a high-resolution image extent were considered. The pan-sharpened natural color image was used to eliminate high-resolution NHD lines that were visually identified to be dry at the time of image collection. We note that a river could be observed visually to contain water, and thus included in the analysis, but inadequately mapped as wet by the PSHR imagery because of spectral confusion during image processing and classification. For rivers with standing or flowing water, river width was calculated as the average of 20 randomly selected wetted-width measure-ments taken along the length of each river using the pan-sharpened natural-color image. For each ‘wet river’ (a subset of the NHD high resolution stream lines), 100 points were randomly generated along the center line. This totaled 23 streams (n = 2300 points) across the sites, dates and different streams/rivers within each site and date. The percent of the river points classified as wet by each imagery was recorded.

3. Results

3.1. Accuracy of surface water outputs

The Cottonwood Lake study area water level field data indicated that water was present in eight of the wetlands measured. For the two dates used to validate the images, the validation wetlands ranged from 0.2 to 17.6 ha in size and averaged 3.8 to 4.2 ha in size (Table 3). The PSHR imagery showed an overall accuracy of 94% and a dice coefficient of 93% with a 12% omission and 0.6% commission error in mapping surface-water extent (Table 3). This can be compared to the 22 and 42% error of omission observed when validating the MF Landsat and GSW Landsat products, respectively, with the water level field data (Table 4, overall accuracy 87 and 79%, respectively). Relative bias was negative for all three data sources indicating that surface-water extent tended to be underestimated (Tables 3 and 4).

Table 4.

A comparison of the accuracy statistics generated when the validation points were randomly sampled across (1) all image extents, (2) lakes only, (3) wetlands only, (4) rivers only, and (5) using field-based water level data from the Cottonwood Lake Study Area. Validation points were generated independently for each of the five separate validation efforts. MF: matched filtered, GSW: Global Surface Water monthly product, n: number of validation points.

Sensor Validation n Omission
Error (%)		 Commission
Error (%)		 Overall
Accuracy (%)		 Dice
Coefficient
(%)		 Relative
Bias (%)
MF Landsat All aquatic systems 12,000 7 1.9 95.6 95.5 −5.2
Lakes 2,000 3.6 1.7 97.4 97.3 −1.9
Wetlands 1,600 7.5 4.8 93.9 93.8 −2.9
Rivers 1,600 64.3 2.4 67.4 52.3 −63.4
Field water levels 400 22 4.4 87.3 85.9 −18.5
GSW Landsat All aquatic systems 12,000 15.7 1.4 91.6 90.9 −14.5
Lakes 2,000 6.4 1.9 95.9 95.8 −4.6
Wetlands 1,600 14.4 2.4 91.8 91.2 −12.3
Rivers 1,600 84.3 0.8 57.8 27.2 −84.1
Field water levels 400 42 0 79.2 73.1 −41.8
Open in a new tab

3.2. Total surface-water extent

We found that in mapping total surface-water extent across the PPR study sites, the PSHR imagery documented approximately 10 to 15% more surface water than Landsat outputs. When points were randomly selected across all PPR study area extents (wetlands, lakes, and river scene extents, n = 12,000), using the PSHR inunda-tion extent to define the reference condition, the MF and GSW Landsat showed a 7 and 16% error of omission, a 2 and 1% error of commission, and a relative bias of −5% and −15%, respectively, for surface water. Accuracy statistics by aquatic system can help us understand the distribution of the surface water missed by Landsat, relative to using PSHR imagery. Errors of commission for surface water were <5% across all systems and Landsat products, whereas errors of omission for surface water ranged widely and were highest in river systems. The MF Landsat had errors of omission of 4%, 8%, and 64% for lake, wetland, and river systems, respectively. Similarly, errors of omission for the GSW Landsat were 6%, 14%, and 84% for lake, wetland, and river systems, respectively. In wetland systems, we found that the two Landsat products showed similar errors of omission (8% vs 14%), whereas the PSHR imagery mapped 1% less and 26% more wetland surface water than the MF Landsat and GSW Landsat products, respectively.

3.3. Characterizing the distribution of surface water

Having an accurate understanding of the spatial distribution of surface water, in addition to the total quantity of surface-water, is often of interest for applications such as the manage-ment of nutrient loading and wildlife habitat. Not surprisingly, the PSHR imagery provided a substantial advantage in detecting small, inundated wetlands, as well as river and streams. The Prairie Pothole Region (PPR) contains a higher density of small (<1 ha) and in particular very small (<0.2 ha) NWI wetlands relative to regions west, south and east of the PPR (Figure 4). According to the NWI dataset, the density of very small wetlands (7.7 and 5.0 wetlands per km2) and small (2.9 and 2.8 wetlands per km2) wetlands is greatest across parts of North and South Dakota, respectively (Figure 4). The large number of small wetlands that are mapped by the NWI dataset often raises concern regarding the appropriateness of using Landsat imagery in a region dominated by such small wetlands. The wetland image dates used in the analysis ranged from April through September, therefore more accurately repre-senting the growing season and not the spring peak in wetness. Only 42% of the NWI wetlands across our wetland sites were mapped as inundated by the PSHR imagery. The remaining 58% of the NWI wetlands is likely a mix of ephemeral wetlands dry at the time of image collection, as well as wetlands that have been lost, or are no longer functioning as wetlands attributable to agriculture and associated tile drainage (Werner et al. 2016). If the number of wetlands detected by Landsat is compared instead to the wetlands mapped as inundated by the PSHR imagery, we can more adequately evaluate the advantage provided by the PSHR imagery outputs relative to Landsat to map the distribution of inundated wetlands. The MF Landsat mapped 28% of the NWI wetlands as inundated but found 65%of the wetlands identified as inundated by the PSHR imagery. The GSW Landsat, meanwhile, mapped 20% of the NWI wetlands as inundated, but found 46% of the wetlands identified as inundated by the PSHR imagery. An example of this is shown in Figure 5, where we see considerable variation in the ability to map the surface-water extent of smaller wetlands in the northern portion of the study area. We found that the differential ability to detect small wetlands can change our interpretation of the size distribution of wetlands (Figure 6(a)) but appears to have very little impact on our interpretation of the contribution of different water body sizes to the total surface water extent (Figure 6(b)). This suggests that using PSHR imagery we would interpret the landscape as dominated by small (<0.2 ha) wetlands, whereas using Landsat imagery we would interpret the landscape as dominated by moderate-sized wetlands (0.2–8.0 ha). In contrast, all sources agreed that most of the total surface water occurred in large (>8.0 ha) waterbodies.

Figure 4.

Figure 4.

Open in a new tab

Heat maps showing the point density of National Wetlands Inventory wetlands (a) smaller than 0.2 ha in size and (b) between 0.2 and 1.0 ha in size.

Figure 5.

Figure 5.

Open in a new tab

A comparison of a section of the Cottonwood Lake Study Area as (a) seen by high-resolution imagery, (b) mapped from pan-sharpened high-resolution (PSHR) imagery, (c) seen by Landsat, (d) mapped by locally trained (MF) Landsat, (e) mapped by the Global Surface Water (GSW) monthly Landsat product maximum. Copyright 2011 Digital Globe, Next View License.

Figure 6.

Figure 6.

Open in a new tab

The distribution of wetlands across the Cottonwood Lake Study Area and the Waubay National Wildlife Refuge area defined as (a) the frequency of wetlands in each size class as a percent of all wetlands identified, and (b) the summed area of wetlands in each size class as a percent of all wetlands identified. Using PSHR imagery we would interpret the landscape as dominated by small (<0.2 ha) wetlands, whereas using Landsat imagery we would interpret the landscape as dominated by moderate-sized wetlands (0.2–8.0 ha) (a), however all sources agree that most of the water occurs in large (>8.0 ha) waterbodies (b) Error bars indicate standard error estimates. NWI: National Wetlands Inventory wetlands, PSHR: pan-sharpened high-resolution, MF: locally trained, GSW: Global Surface Water monthly product.

Outside of very large rivers and large floods, monitoring rivers and streams is typically not possible due to the narrowness of the features. Across the region, for example, rivers potentially identifiable with Landsat (>30 m wide), as identified by the North American River Width Data Set (NARWidth) (Allen and Pavelsky 2015), represent, on average, 1% of total stream length, as mapped by the High Resolution NHD dataset. Across the two river sites, we mapped a total of 23 streams and rivers, where 14 showed a width of <10 m, 4 showed a width averaging 10 m to 30 m, and 5 showed a width of >30 m (Table 5). Although most streams and rivers are simply too narrow to be detected with Landsat imagery, the MF Landsat mapped approximately double the stream and river length relative to the GSW Landsat product (22 and 10% for rivers 10 to 30 m wide, and 71 and 37% for rivers 31 to 85 m wide, for MF Landsat and GSW Landsat product, respectively) (Table 5 and Figure 7). In contrast, the PSHR imagery mapped 37%, 72%, and 93% of rivers <10 m wide, 10 to 30 m wide and 31 to 85 m wide, respectively (Table 5).

Table 5.

Summary of the percent of stream length mapped by stream width and data source. Average stream/river width was determined from the pan-sharpened high-resolution images (PSHR). MF: matched filtering, GSW: Global Surface Water monthly product.

Average
wetted
width (m)		 PSHR (%
Detected)		 MF
Landsat
(%
Detected)		 GSW (%
Detected)		 MF
Landsat
relative to
PSHR (%)		 GSW
relative
to PSHR
(%)		 Site Month, Year
(PSHR)
2 7 0 0 −100 −100 Niobrara Tributary Aug 2006
3 22 0 0 −100 −100 Sheyenne River Tributary Apr 2003
3 7 0 0 −100 −100 Sheyenne River Tributary Apr 2003
3 2 0 0 −100 −100 Sheyenne River Tributary Apr 2003
3 24 5 0 −79 −100 Sheyenne River Tributary Apr 2003
3 12 1 0 −92 −100 Sheyenne River Tributary Apr 2003
3 27 0 0 −100 −100 Sheyenne River Tributary Apr 2015
4 2 0 0 −100 −100 Niobrara Tributary Jun 2010
4 75 0 0 −100 −100 Niobrara Tributary Aug 2006
6 54 5 0 −91 −100 Sheyenne River Tributary Apr 2003
7 56 0 0 −100 −100 Niobrara Tributary Aug 2006
7 83 0 0 −100 −100 Niobrara Tributary Jun 2010
8 95 5 0 −95 −100 Niobrara Tributary Jun 2010
9 55 4 0 −93 −100 Sheyenne River Tributary Apr 2015
12 66 0 0 −100 −100 Sheyenne River Tributary Apr 2003
15 (4 to 30) 58 38 4 −34 −93 Niobrara Tributary Jun 2010
16 92 4 0 −96 −100 Sheyenne River Tributary Apr 2015
29 (3 to 80) 72 45 36 −38 −50 Sheyenne River Tributary Apr 2003
33 94 28 5 −70 −95 Sheyenne River Apr 2015
41 90 56 5 −38 −94 Niobrara Tributary Jun 2010
45 96 81 29 −16 −70 Sheyenne River Apr 2003
47 94 97 70 3 −26 Niobrara River Aug 2006
84 90 93 78 3 −13 Niobrara River Jun 2010
<10 m 37 1 0 −96 −100 averaged ~
10 to 30 m 72 22 10 −67 −86 averaged ~
31 to 85 m 93 71 37 −23 −60 averaged ~
Open in a new tab

Figure 7.

Figure 7.

Open in a new tab

Mapped river extent across the braided Niobrara River, Nebraska (4 to 150 m wide) and a tributary (8 m wide) (top) and the Sheyenne River, North Dakota and a small tributary (2 to 3 m wide) (bottom). A comparison of stream extent using a Worldview-2 image (a, b) and QuickBird-2 image (e, f), matched filtering (MF) Landsat 8 (top, c) and MF Landsat 5 (bottom, g), and the Global Surface Water monthly product (d and h). The raw high-resolution images are used as the back-ground in all panels. Copyright 2003, 2015 Digital Globe, Next View License.

3.4. Characterizing surface water movement

As landscapes become wetter, they begin to hydrodynamically interact via surface flow through both channelised conveyances as well as fill-spill and fill-merge processes. Because water carries nutrients, pollutants, and biota, accurately characterizing its movements is critical to predicting water quality (Fritz et al. 2018; Lane et al. 2018). In wetland systems, the PSHR imagery found only 7% more wetland to wetland interactions relative to the MF Landsat, but 91% more wetland to wetland interactions than the GS Landsat (Table 6). Similarly, the PSHR imagery found an identical number of wetland clusters as the MF Landsat, created as wetlands merge together, but 62% more wetland clusters than found with the GSW Landsat (Table 6). Lakes are also abundant across parts of the PPR (Figure 8). Unexpectedly, the findings were very similar in lake systems. The PSHR imagery detected only 2% more NWI wetlands within the continuous lake extent than the MF Landsat, but 57% more NWI wetlands than the GSW Landsat. The relatively minor difference in the abundance of wetland-to-wetland interactions between the PSHR imagery and the MF Landsat can be largely attributed to Landsat missing some narrow interactions between wetlands, but also falsely detecting some interactions or overestimating interactions in other areas (Figure 9). Conversely, the PSHR imagery detected narrow interactions but also detected narrow disconnections, or situations where the surface-water extent contracted sufficiently that individual wetlands emerged (Figure 9). Although water in these cases, is still likely moving in the shallow-subsurface, the rate of movement can be expected to be impacted. The more conservative surface-water extent defined by the GSW Landsat meant that the product tended to miss narrower interactions and consequently showed a 15% smaller, continuous lake surface-water extent (Table 7) and fewer wetland interactions in both wetland and lake systems. Figure 10 shows an example of where MF Landsat also missed narrow interactions, for example in Figure 10 the PSHR image detected a narrow interaction between the lake and wetlands north of the lake that was missed by the MF Landsat. Also shown in Figure 10, the GSW Landsat product, missed a key surface-water interaction between Barnes Lake and the lake just north of it, producing a much smaller total lake extent. These results indicate that relatively narrow (<40 m wide) interactions or conveyances can be important in lake systems too, not just wetland systems. We found that as lake levels rise, wetlands are not always subsumed – instead topography can maintain relatively narrow (<40-m wide) interactions between lakes and nearby lakes or wetlands.

Table 6.

Wetland metrics across wetland mosaic sites and years, where the year refers to the year the satellite image was collected. Change is presented relative to the pan-sharpened high resolution (PSHR) imagery, as well as for the PSHR imagery relative to the Landsat products. NWI: National Wetlands Inventory, MF: matched filtering, GSW: Global Surface Water monthly product.

Metric Site
 Cottonwood
Lake wetlands
 Waubay
National
Wildlife Refuge
wetlands
 Change
relative to
PSHR (%)
 PSHR
relative to
Landsat(%)
Source 2011 2015 2004 2012 all all
Surface-water extent. (ha) PSHR (0.5 m) 3330.5 3274.7 3258.9 4009.8 ~ ~
MF Landsat (30 m) 3391.8 3089.4 3678.4 3826.5 0.8 −0.8
GSW Landsat (30 m) 2596.1 3014.8 2589.7 2775.2 −20.9 26.4
NWIa 3085 4567 1858 2139 ~ ~
NWI wetlands detected. (count) PSHR (0.5 m) 1803 1296 735 1106 ~ ~
MF Landsat (30 m) 844 1038 505 819 −35.1 54.1
GSW Landsat (30 m) 565 719 376 621 −53.8 116.6
NWI wetland/wetland interactions. (count) PSHR (0.5 m) 232 230 257 511 ~ ~
MF Landsat (30 m) 236 223 272 424 −6.1 6.5
GSW Landsat (30 m) 125 164 103 252 −47.6 91.0
Clusters of NWI wetlands. (count) PSHR (0.5 m) 68 76 80 143 ~ ~
MF Landsat (30 m) 70 93 84 120 0.0 0.0
GSW Landsat (30 m) 40 63 35 88 −38.4 62.4
Open in a new tab
a

As the NWI wetland dataset can be outdated (generated 1977–2016 depending on the location), wetlands may or may not still be present and may or may not contain water at the time of the image collection.

Figure 8.

Figure 8.

Open in a new tab

Point density of lakes where each point represents a National Wetlands Inventory polygon classified as lacustrine and >20 acres in size. Points were converted to density using a kernel function where each pixel was 100 km2 and the search radius was defined using a spatial variant of Silverman’s Rule of Thumb.

Figure 9.

Figure 9.

Open in a new tab

A comparison of surface-water extent as shown with (a) raw Worldview-2 imagery, and as mapped with (b) pan-sharpened high-resolution (PSHR) imagery, (c) matched filtering (MF) Landsat, and (d) the Global Surface Water (GSW) monthly Landsat product. The high-resolution imagery can map narrow interactions missed by Landsat (yellow circle), whereas the MF Landsat sometimes detected interactions missed by the GSW Landsat (green circle). MF Landsat can also falsely detect interactions, such as across roads, where a culvert may occur but is not visible (purple circle). Copyright 2012 Digital Globe, Next View License.

Table 7.

Continuous lake extent and wetlands detected within lake extent by site and year. Wetland/lake surface water interactions occur when a lake expands, subsuming a wetland, or a wetland fills and spills into a lake. Change is presented relative to the pan-sharpened high resolution (PSHR) outputs, as well as for the PSHR outputs relative to the Landsat products. NWI: National Wetlands Inventory, MF: matched filtering, GSW: Global Surface Water monthly product.

Metrics Site
 Waubay National
Wildlife Refuge
lakes
 Barnes Lake
 Horseshoe
Lakes
 Change
relative to
PSHR (%)
 PSHR
relative to
Landsat
(%)
Year 2012 2015 2011 2015 2015 all all
NWI wetlands detected within lake surface-water extent. (count) PSHR (0.5 m) 557 618 127 79 73 ~ ~
MF Landsat (30 m) 562 618 107 75 64 1.9 2.0
GSW Landsat (30 m) 263a 538 9 61 58 −36.1 56.5
Continuous lake surface-water. extent (ha) PSHR (0.5 m) 12,595.9 14,510.8 1171.8 1044.5 886.5 ~ ~
MF Landsat (30 m) 12,776.7 14,538.7 1139.0 1003.2 843.6 0.3 −0.3
GSW Landsat (30 m) 5565.0a 13,067.7 205.0 930.8 741.6 −32.1 (−15.2b) 47.3 (18.0b)
Open in a new tab
a

Product affected by the Scan-Line Corrector off (SLC-off) error.

b

Percent change in continuous lake surface-water extent relative to PSHR outputs, excluding the Wildlife Refuge lakes (2012).

Figure 10.

Figure 10.

Open in a new tab

A comparison of Barnes Lake, North Dakota as (a) seen by high-resolution imagery, (b) mapped by pan-sharpened high-resolution (PSHR, 0.5 m) imagery with National Wetlands Inventory (NWI) polygons shown within the continuous lake extent, (c) seen by Landsat, (d) mapped by matched filtering (MF, 30 m) Landsat and the Landsat Global Surface Water (GSW, 30 m) monthly product. Only surface water continuous with Barnes Lake is shown. The GSW missed the surface connection between Barnes Lake and the wetlands north of it. Key interactions detected or missed are shown in the yellow and green circles. Copyright 2011 Digital Globe, Next View License.

4. Discussion

Enhancing our understanding of water distribution and movement across regions such as the Prairie Pothole Region is of high importance. As the ‘duck factory’ of North America, smaller wetlands across the PPR are highly valued as waterfowl habitat (Ballard et al. 2014), however anthropogenic-related wetland losses across the region have occurred dispropor-tionately for smaller wetlands, losses attributable to agricultural development, including tile drainage, and climate-induced drying (Johnston 2013; Dahl 2014). In river systems, improv-ing our ability to monitor river stage has multiple applications including enhancing flood predictions, as well as informing the distribution of flow to downstream gaged points, and improving river management (Schumann et al. 2010; Demarchi, Bizzi, and Piegay 2016). In addition, under wetter conditions, surface-water extent expands, and movement of water can occur between wetlands via ‘fill-and-spill’ (Tromp-van Meerveld and McDonnell 2006) or ‘fill-and-merge’ (Vanderhoof and Alexander 2016; Leibowitz, Mushet, and Newton 2016). The movement of this water is highly relevant as it informs the movement of nutrients, sediment and pollutants across a watershed, profoundly effecting water quality (Leibowitz, Mushet, and Newton 2016) and wildlife habitat suitability (Krapu et al. 2004; McLean, Mushet, and Stockwell 2016).

In this study we found that PSHR imagery, relative to Landsat imagery, provided a substantial advantage in documenting small wetlands and rivers. Although this finding was not unexpected, what was more surprising was that the difference was substantial enough to change our interpretation of the wetland size distribution across the land-scape (Figure 6A), a finding relevant, for example, to predicting the movement of (or habitat availability for) aquatic species, such as insects, amphibians and reptiles that use multiple waterbodies (Schofield et al. 2018). Another surprising finding was that Landsat, relative to the PSHR imagery, can potentially provide similar estimates of wetland-to-wetland and wetland-to-lake water movement as well as the creation of wetland clusters, although with less positional accuracy. In addition, we found that relatively narrow (<60 m wide) conveyances were not restricted to wetland systems, but often still relevant in lake systems. However, the substantial underestimation of these metrics by the GSW product suggests that considering mixed water pixels is critical to not only mapping smaller wetlands (<1 ha), but also for detecting key surface-water interactions. The automation of estimating the sub-pixel water fraction from Landsat imagery over large areas and long-time periods could therefore be used to improve upon national to global Landsat products (e.g. DeVries et al. 2017).

The performance of a remote sensing product can be highly variable even after controlling for the source of imagery (e.g. Feyisa et al. 2014; Fisher, Flood, and Danaher 2016). Variable performance in water resource applications can occur, in part, because of the heterogeneity in the spectral characteristics of waterbodies, which can vary with chlorophyll concentration, turbidity, water depth, waterbody bottom material, and observation conditions (Ozesmi and Bauer 2002; Arst 2003; Zomer, Trabucco, and Ustin 2009). As most traditional classifiers depend on homogeneity across the region of interest (Foody, McCulloch, and Yates 1995; Brown, Gunn, and Lewis 1999), as feature heterogeneity increases, for example when both open water and vegetated water occur, misclassification errors can also increase (Ozesmi and Bauer 2002). Approaches that accommodate high heterogeneity, for example by classifying mixed (e.g. partially vegetated) pixels as water, may be able to better document surface water characteristics relative to an approach that primarily maps open water, such as the GSW Landsat product. It is also possible that machine learning approaches, which are increasing in popularity (e.g. Tuia et al. 2011; Hird et al. 2017), may out-perform more traditional classifiers such as ML or the Matched Filtering algorithm, particularly when applied at a regional scale.

Whereas the finer spatial resolution of commercial high-resolution imagery can enhance our ability to characterize surface water distribution and movement (e.g. Cooley et al. 2017), surface water patterns across regions such as the PPR, can be expected to vary not only interannually, but also seasonally (Beeri and Phillips 2007), peaking with snowmelt and precipitation events (Leibowitz, Mushet, and Newton 2016). DigitalGlobe’s satellites, however, do not collect data at regular intervals, and although other sources of high-resolution data, such as Planet’s constellations of CubeSats, are available at regular intervals, the cost of imagery (e.g. DigitalGlobe, Planet) may limit its role in operational monitoring of surface water distribution and movement. Considering these challenges, it might instead be preferable to utilize freely available sources of imagery such as Sentinel-1 and Sentinel-2 (10 to 20 m resolution) that improve upon Landsat spatial resolution whilst retaining regular collection intervals (e.g. Huang et al. 2018). Even given limited high-resolution imagery, however, useful data can still be extracted that could be potentially integrated into a dataset such as the NWI dataset. Patterns of surface water as mapped by high-resolution imagery could be used, for example, to help characterize minimum and maximum surface water extents, and corresponding changes in wetland size and fre-quency (influenced by wetlands wetting up as well as merging or emerging). This data would be useful to (1) identify wetlands that are no longer present or functional due to agricultural activities such as tile drainage (Werner et al. 2016), as well as (2) help predict how land use changes or shifts in climate patterns will impact the movement of nutrients and pollutants (Lane et al. 2018).

To explore how single collection PSHR images could be used to identify potential water movement, for example, we visually identified (using the PSHR raw images) dry con-veyances (e.g. swales, ephemeral channels) between wetlands and between lakes and wetlands that showed evidence of regular water presence but was either dry or non-continuous at the time of the image collection (Figure 11). In lake systems, where extent has been shown to reflect longer term trends in wetness relative to wetlands (e.g. Zhang, Schwartz, and Liu 2009), visible, potential conveyances were found to represent a relatively minor percent of wetland-to-lake interactions (6% of all identified wetland-to-lake interactions). However, across wetland sites the potential conveyances represented a potential 46% increase in the total number of wetland-to-wetland interactions, even when using images collected during wet years. These initial findings support the need to characterise potential episodic or short-duration interactions (e.g. Evenson et al. 2016; Hay et al. 2018), or move towards near-daily surface-water products that include multiple sensors such as both Landsat and Sentinel-2 (DeVries et al. 2017) or incorporate fine to moderate resolution SAR imagery (e.g. Radarsat-2, Sentinel-1) to map conditions through cloud cover (Miles et al. 2017; Pham-Duc, Prigent, and Aires 2017).

Figure 11.

Figure 11.

Open in a new tab

Examples of potential conveyances between wetlands that showed visual evidence of occasional to intermittent surface-water presence, but were dry or not continuously wet at the time of the image collection. Copyright 2004 and 2011 Digital Globe, Next View License.

We must also consider how applicable our findings are to other landscapes. The river, lake and wetland sites examined in this study were spread across multiple states (North Dakota, South Dakota, and Nebraska) but restricted to the Prairie Pothole Region. Sites were selected across this region to have a relatively high density of wetlands, large lakes, and surface-water interactions (Beeri and Phillips 2007; Huang et al. 2011; Vanderhoof, Alexander, and Todd 2016). The findings presented here can be expected to be relevant to waterbodies across other non-forested regions (e.g. systems dominated by grassland, shrub/scrub or savannah) and to be particularly relevant to floodplains, permafrost landscapes and other formerly glaciated landscapes that often exhibit low topographic gradients and low rates of infiltration that can lead to highly dynamic surface water storage (Hamilton, Sippel, and Melack 2004; Yao et al. 2007; Aragón, Jobbágy, and Viglizzo 2011; Kuppel et al. 2015). We might also expect the findings to be relevant in wetlands dominated by water-adapted vegetation, where products restricted to open water will likely underestimate spectrally mixed pixels. Examples of such areas include the Everglades in Florida (Jones 2015), as well as wetlands dominated by plant species such as cattails (Typha spp.) or sedges (Scirpus spp.) (e.g. palustrine emergent wetlands). However, the findings may be less applicable in heavily forested regions (including areas with forested wetlands) where both PSHR imagery as well as Landsat imagery may underestimate surface-water extent during the growing season when tree cover can mask water. In such regions, mapping surface water with multispectral imagery is typically restricted to leaf-off periods (e.g. Vanderhoof et al. 2017) or may require a reliance on SAR imagery (e.g. Hess et al. 2015; Schlaffer et al. 2016).

This study found that input-data granularity influences our interpretation of sur-face water distribution and movement that can potentially impact subsequent appli-cations of surface water monitoring such as resource management or modeling efforts. For example, hydrologic and biogeochemical model outputs may change markedly depending on the defined expanse of wetted areas, flowpath extent, active contributing areas, and processing length (e.g. Baker, Weller, and Jordan 2007; Nadeau and Rains 2007; Rains et al. 2015; Golden et al. 2017). Whereas Landsat can provide adequate estimates of total surface water extent, for applications in which knowledge of surface water distribution and movement through a watershed, between waterbodies, or into the stream network is important (e.g. Golden et al. 2016; Evenson et al. 2018), either PSHR imagery, a Landsat product that considers partially inundated pixels, or an alternative finer resolution source of imagery (e.g. Sentinel-2 at 10 to 20 m resolution) may be necessary.

5. Conclusion

This study was unique in contrasting high-resolution imagery with Landsat products not just across diverse aquatic systems, but also comparing their ability to characterise surface-water movement between waterbodies, which can have important consequences for watershed hydrology (Shaw, Pietroniro, and Martz 2013; Golden et al. 2016). Mixed pixel approaches to process Landsat, relative to PSHR imagery, documented similar estimates of total surface water extent and frequency of interactions between waterbodies, however, by documenting many more small, inundated wetlands, the PSHR imagery in the PPR, modified our interpretation of the landscape-scale wetland size distribution. Our findings also suggest that whilst global water products such as the GSW Landsat product can allow for an unprecedented improvement in our understanding of large-scale patterns in sur-face-water distribution, it may be less appropriate for regional or localised applications, particularly in regions dominated by a high density of small wetlands, or where the distribution of water in smaller waterbodies or the movement of surface-water as aquatic systems interact via conveyances (e.g. swales, rivers, and streams) is of interest. With the launch of an increasing number of satellites that collect SAR imagery (e.g. Sentinel-1) or higher-resolution imagery, including Sentinel-2 (10 to 20 m resolution) as well as networks of microsatellites (e.g. Dove, RapidEye (5 m resolution, Planet, San Francisco)), opportunities to monitor surface-water extent and interactions between aquatic systems at finer spatial resolutions and near-daily return intervals are rapidly becoming a possibility. The increased availability of such high-resolution imagery necessitates an improved understanding of where multi-spectral high-resolution imagery can enhance our understanding of surface water storage, distribution and movement.

Acknowledgments

We thank Hayley Distler and Marena Gilbert for their assistance in processing the imagery and ancillary datasets, and Jay Christensen, Nicole Fairaux and the anonymous reviewers for their helpful comments on earlier versions of the manuscript. The views expressed in this manuscript are solely those of the authors and do not necessarily reflect the views or policies of the U.S. EPA. High-resolution imagery was provided through the NextView License. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. Data generated in this study will be publicly available through ScienceBase (doi:10.5066/P9BVAURT) following publication.

Funding

This research was funded by the President’s Drought Resilience Initiative through an interagency agreement between the U.S. Environmental Protection Agency, Office of Research and Development and the U.S. Geological Survey [DW-014-92454401-0]. This research was also funded by the U.S. Geological Survey, Land Resources Mission Area, Land Change Science Program.

Footnotes

Disclosure statement

No potential conflict of interest was reported by the authors.

References

  1. Adler-Golden SM, Berk A, Bernstein LS, Richtsmeierl S, Acharyal PK, Matthew MW, Anderson GP, Allred CL, Jeong LS, and Chetwynd JH. 1998. “FLAASH, A MODTRAN4 Atmospheric Correction Package for Hyperspectral Data Retrievals and Simulations” AVIRIS 1998 Proceedings, 97 vols., 9–14, Pasadena, CA: JPL Publication, 1–6. [Google Scholar]
  2. Adler-Golden SM, Matthew MW, Bernstein LS, Levine RY, Berk A, Richtsmeier SC, Acharya PK, et al. 1999. “Atmospheric Correction for Shortwave Spectral Imagery Based on MODTRAN4.” SPIE Proceedings Imaging Spectrometry V 3753: 61–69. [Google Scholar]
  3. Allen GH, and Pavelsky TM. 2015. “Patterns of River Width and Surface Area Newly Revealed by the Satellite-Derived North American River Width Data Set.” Geophysical Research Letters 42 (2): 395–402. doi: 10.1002/2014GL062764. [DOI] [Google Scholar]
  4. Alsdorf DE, Rodriguez E, and Lettenmaier DP. 2007. “Measuring Surface Water from Space.” Reviews of Geophysics 45 (2): 1–24. doi: 10.1029/2006RG000197. [DOI] [Google Scholar]
  5. Amro I, Mateos J, Vega M, Molina R, and Katsaggelos AK. 2011. “A Survey of Classical Methods and New Trends in Pansharpening of Multispectral Images.” EURASIP Journal on Advances in Signal Processing 79: 1–22. [Google Scholar]
  6. Aragón R, Jobbágy EG, and Viglizzo EF. 2011. “Surface and Groundwater Dynamics in the Sedimentary Plains of the Western Pampas (Argentina).” Ecohydrology 4 (3): 433–447. doi: 10.1002/eco.149. [DOI] [Google Scholar]
  7. Arst H. 2003. Optical Properties and Remote Sensing of Multi-Componental Water Bodies Vol. XII of Marine Science and Coastal Management Ch. 1 Berlin, Germany: Springer Science Praxis. [Google Scholar]
  8. Bainbridge TR 1985. “The Committee on Standards: Precision and Bias.” ASTM Standardization News 13: 44–46. [Google Scholar]
  9. Baker ME, Weller DE, and Jordan TE. 2007. “Effects of Stream Map Resolution on Measures of Riparian Buffer Distribution and Nutrient Retention Potential.” Landscape Ecology 22 (7): 973–992. doi: 10.1007/s10980-007-9080-z. [DOI] [Google Scholar]
  10. Ballard T, Seager R, Smerdon JE, Cook BI, Ray AJ, Rajagopalan B, Kushnir Y, Nakamura J, and Henderson N. 2014. “Hydroclimate Variability and Change in the Prairie Pothole Region, the “Duck Factory” of North America.” Earth Interactions 18 (14): 1–28. doi: 10.1175/EI-D-14-0004.1. [DOI] [Google Scholar]
  11. Beeri O, and Phillips RL. 2007. “Tracking Palustrine Water Seasonal and Annual Variability in Agricultural Wetland Landscapes Using Landsat from 1997 to 2005.” Global Change Biology 13: 897–912. [Google Scholar]
  12. Bolanos S, Stiff D, Brisco B, and Pietroniro A. 2016. “Operational Surface Water Detection and Monitoring Using Radarsat 2.” Remote Sensing 8: 285, 1–18. doi: 10.3390/rs8040285. [DOI] [Google Scholar]
  13. Brown M, Gunn S, and Lewis H. 1999. “Support Vector Machines for Optimal Classification and Spectral Unmixing.” Ecological Modeling 120: 167–179. doi: 10.1016/S0304-3800(99)00100-3. [DOI] [Google Scholar]
  14. Castilla G, Hernando A, Zhang C, and McDermid J. 2014. “The Impact of Object Size on the Thematic Accuracy of Landcover Maps.” International Journal of Remote Sensing 35 (3): 1029–1037. doi: 10.1080/01431161.2013.875630. [DOI] [Google Scholar]
  15. Chen Y, Huang C, Ticchurst C, Merrin L, and Thew P. 2013. “An Evaluation of MODIS Daily and 8-Day Composite Products for Floodplain and Wetland Inundation Mapping.” Wetlands 33 (5): 823–835. doi: 10.1007/s13157-013-0439-4. [DOI] [Google Scholar]
  16. Cooley S, Smith L, Stepan L, and Mascara J. 2017. “Tracking Dynamic Northern Surface Water Changes with High-Frequency Planet CubeSat Imagery.” Remote Sensing 9: 1306. doi: 10.3390/rs9121306. [DOI] [Google Scholar]
  17. Covino T 2017. “Hydrologic Connectivity as a Framework for Understanding Biogeochemical Flux through Watersheds and along Fluvial Networks.” Geomorphology 277 (SupplementC): 133–144. doi: 10.1016/j.geomorph.2016.09.030. [DOI] [Google Scholar]
  18. Dahl TE 2014. Status and Trends of Prairie Wetlands of the United States 1997 to 2009, 67 Washington, DC: U.S. Department of the Interior, Fish and Wildlife Service, Ecological Services. [Google Scholar]
  19. Demarchi L, Bizzi S, and Piegay H. 2016. “Hierarchical Object-Based Mapping of Riverscape Units and In-Stream Mesohabitats Using LiDAR and VHR Imagery.” Remote Sensing 8 (2): 1–23. doi: 10.3390/rs8020097. [DOI] [Google Scholar]
  20. DeVries B, Huang C, Lang M, Jones J, Huang W, Creed I, and Carroll M. 2017. “Automated Quantification of Surface Water Inundation in Wetlands Using Optical Satellite Imagery.” Remote Sensing 9: 807. doi: 10.3390/rs9080807. [DOI] [Google Scholar]
  21. Du Z, Li W, Zhou D, Tian L, Ling F, Wang H, Gui Y, and Sun B. 2014. “Analysis of Landsat-8 OLI Imagery for Land Surface Water Mapping.” Remote Sensing Letters 5 (7): 672–681. doi: 10.1080/2150704X.2014.960606. [DOI] [Google Scholar]
  22. Evenson GR, Golden HE, Lane CR, and D’Amico E. 2016. “An Improved Representation of Geographically Isolated Wetlands in a Watershed-Scale Hydrologic Model.” Hydrological Processes 30 (22): 4168–4184. doi: 10.1002/hyp.10930. [DOI] [Google Scholar]
  23. Evenson GR, Golden HE, Lane CR, McLaughlin DL, and D’Amico E. 2018. “Depressional Wetlands Affect Watershed Hydrological, Biogeochemical, and Ecological Functions.” Ecological Applications 28 (4): 953–966. doi: 10.1002/eap.1701. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Feyisa GL, Meilby H, Fensholt R, and Proud SR. 2014. “Automated Water Extraction Index: A New Technique for Surface Water Mapping Using Landsat Imagery.” Remote Sensing of Environment 140: 23–35. doi: 10.1016/j.rse.2013.08.029. [DOI] [Google Scholar]
  25. Fisher A, Flood N, and Danaher T. 2016. “Comparing Landsat Water Index Methods for Automated Water Classification in Eastern Australia.” Remote Sensing of Environment 175: 167–182. doi: 10.1016/j.rse.2015.12.055. [DOI] [Google Scholar]
  26. Fleiss JL 1981. Statistical Methods for Rates and Proportions. 2nd ed. New York, NY: John Wiley & Sons. [Google Scholar]
  27. Foody G, McCulloch M, and Yates W. 1995. “Classification of Remotely Sensed Data by an Artificial Neural Network: Issues Related to Training Data Characteristics.” Photogrammetry Engineering & Remote Sensing 61 (4): 391–401. [Google Scholar]
  28. Fritz KM, Schofield KA, Alexander LC, McManus MG, Golden HE, Lane CR, Kepner WG, LeDuc SD, DeMeester JE, and Pollard AI. 2018. “Physical and Chemical Connectivity of Streams and Riparian Wetlands to Downstream Waters: A Synthesis.” Journal of the American Water Resources Association 54 (2): 323–345. doi: 10.1111/1752-1688.12632. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Frohn RC, D’Amico E, Lane C, Autrey B, Rhodus J, and Liu H. 2012. “Multi-Temporal Sub-Pixel Landsat ETM+ Classification of Isolated Wetlands in Cuyahoga County, Ohio, USA.” Wetlands 32: 289–299. doi: 10.1007/s13157-011-0254-8. [DOI] [Google Scholar]
  30. Golden HE, Creed IF, Ali G, Basu NB, Neff BP, Rains MC, McLaughlin DL, et al. 2017. “Integrating Geographically Isolated Wetlands into Land Management Decisions.” Frontiers in Ecology and the Environment 15 (6): 319–327. doi: 10.1002/fee.1504. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Golden HE, Sander HA, Lane CR, Zhao C, Price K, D’Amico E, and Christensen JR. 2016. “Relative Effects of Geographically Isolated Wetlands on Streamflow: A Watershed-Scale Analysis.” Ecohydrology 9 (1): 21–38. doi: 10.1002/eco.1608. [DOI] [Google Scholar]
  32. Green AA, Berman M, Switzer P, and Craig MD. 1988. “A Transformation for Ordering Multispectral Data in Terms of Image Quality with Implications for Noise Removal.” IEEE Transactions on Geoscience and Remote Sensing 26 (1): 65–74. doi: 10.1109/36.3001. [DOI] [Google Scholar]
  33. Hamilton S, Sippel S, and Melack J. 2004. “Seasonal Inundation Patterns in Two Large Savanna Floodplains of South America: The Llanos De Moxos (Bolivia) and the Llanos Del Orinoco (Venezuela and Colombia).” Hydrological Processes 18 (11): 2103–2116. doi: 10.1002/(ISSN)1099-1085. [DOI] [Google Scholar]
  34. Hay L, Norton P, Viger R, Markstrom S, Regan RS, and Vanderhoof M. 2018. “Modeling Surface-Water Depression Storage in a Prairie Pothole Region.” Hydrological Processes 32 (4): 462–479. doi: 10.1002/hyp.11416. [DOI] [Google Scholar]
  35. Hess LL, Melack JM, Affonso AG, Barbosa C, Gastil-Buhl M, and Novo EMLM. 2015. “Wetlands of the Lowland Amazon Basin: Extent, Vegetative Cover, and Dual-Season Inundated Area as Mapped with JERS-1 Synthetic Aperture Radar.” Wetlands 35 (4): 745–756. doi: 10.1007/s13157-015-0666-y. [DOI] [Google Scholar]
  36. Hird JN, DeLancey ER, McDermid GJ, and Kariyeva J. 2017. “Google Earth Engine, Open-Access Satellite Data, and Machine Learning in Support of Large-Area Probabilistic Wetland Mapping.” Remote Sensing 9: 1315, 1–27. doi: 10.3390/rs9121315. [DOI] [Google Scholar]
  37. Homer C, Dewitx J, Yang L, Jin S, Danielson P, Xian G, Coulston J, Herold N, Wickham J, and Megown K. 2015. “Completion of the 2011 National Land Cover Database for the Conterminous United States – Representing a Decade of Land Cover Change Information.” Photogrammetric Engineering and Remote Sensing 81: 345–354. [Google Scholar]
  38. Huang S, Dahal D, Young C, Chander G, and Liu S. 2011. “Integration of Palmer Drought Severity Index and Remote Sensing Data to Simulate Wetland Water Surface from 1910 to 2009 in Cottonwood Lake Area, North Dakota.” Remote Sensing of Environment 115: 3377–3389. doi: 10.1016/j.rse.2011.08.002. [DOI] [Google Scholar]
  39. Huang W, DeVries B, Huang C, Lang MW, Jones JW, Creed IF, and Carroll ML. 2018. “Automated Extraction of Surface Water Extent from Sentinel-1 Data.” Remote Sensing 10 (5): 797, 1–18. doi: 10.3390/rs10050797. [DOI] [Google Scholar]
  40. Jin H, Huang C, Lang MW, Yeo IY, and Stehman SV. 2017. “Monitoring of Wetland Inundation Dynamics in the Delmarva Peninsula Using Landsat Time-Series Imagery from 1985 to 2011.” Remote Sensing of Environment 190: 26–41. doi: 10.1016/j.rse.2016.12.001. [DOI] [Google Scholar]
  41. Johnston C 2013. “Wetland Losses Due to Row Crop Expansion in the Dakota Prairie Pothole Region.” Wetlands 33 (1): 175–182. doi: 10.1007/s13157-012-0365-x. [DOI] [Google Scholar]
  42. Jones JW 2015. “Efficient Wetland Surface Water Detection and Monitoring via Landsat: Comparison with in Situ Data from the Everglades Depth Estimation Network.” Remote Sensing 7 (9): 12503–12538. doi: 10.3390/rs70912503. [DOI] [Google Scholar]
  43. Knight JF, and Lunetta RS. 2003. “An Experimental Assessment of Minimum Mapping Unit Size.” IEEE Transactions on Geoscience and Remote Sensing 41 (9): 2132–2134. doi: 10.1109/TGRS.2003.816587. [DOI] [Google Scholar]
  44. Krapu GL, Pietz PJ, Brandt DA, and Cox RS Jr. 2004. “Does Presence of Permanent Freshwater Affect Recruitment in Prairie-Nesting Dabbling Ducks?” Journal of Wildlife Management 68 (2): 332–341. doi: 10.2193/0022-541X(2004)068[0332:DPOPFW]2.0.CO;2. [DOI] [Google Scholar]
  45. Kuppel S, Houspanossian J, Nosetto MD, and Jobbágy EG. 2015. “What Does It Take to Flood the Pampas? Lessons from a Decade of Strong Hydrological Fluctuations.” Water Resources Research 51 (4): 2937–2950. doi: 10.1002/2015WR016966. [DOI] [Google Scholar]
  46. Laben CA, and Brower BV. 2000. Process for enhancing the spatial resolution of multispectral imagery using pan-sharpening. US Patent 6011875.
  47. Lane CR, Leibowitz SG, Autrey BC, LeDuc SD, and Alexander LC. 2018. “Hydrological, Physical, and Chemical Functions and Connectivity of Non-Floodplain Wetlands to Downstream Waters: A Review.” Journal of American Water Resources Association 54 (2): 346–371. doi: 10.1111/1752-1688.12633. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Lang MW, and McCarty GW. 2009. “Lidar Intensity for Improved Detection of Inundation below the Forest Canopy.” Wetlands 29 (4): 1166–1178. doi: 10.1672/08-197.1. [DOI] [Google Scholar]
  49. Lechner AM, Stein A, Jones SD, and Ferwerda JG. 2009. “Remote Sensing of Small and Linear Features: Quantifying the Effects of Patch Size and Length, Grid Position and Detectability on Land Cover Mapping.” Remote Sensing of Environment 113 (10): 2194–2204. doi: 10.1016/j.rse.2009.06.002. [DOI] [Google Scholar]
  50. Leibowitz SG, Mushet DM, and Newton WE. 2016. “Intermittent Surface Water Connectivity: Fill and Spill Vs Fill and Merge Dynamics.” Wetlands 36 (S2): 323–342. doi: 10.1007/s13157-016-0830-z. [DOI] [Google Scholar]
  51. Lin C, Wu CC, Tsogt K, Ouyang YC, and Chang CI. 2015. “Effects of Atmospheric Correction and Pansharpening on LULC Classification Accuracy Using WorldView-2 Imagery.” Information Processing in Agriculture 2 (1): 25–36. doi: 10.1016/j.inpa.2015.01.003. [DOI] [Google Scholar]
  52. Masek JG, Vermote EF, Saleous N, Wolfe R, Hall EF, Huemmrich F, Gao F, Kutler J, and Teng-Kui L. 2006. “A Landsat Surface Reflectance Data Set for North America, 1990–2000.” Geoscience and Remote Sensing Letters 3: 68–72. [Google Scholar]
  53. McLean KI, Mushet DM, and Stockwell CA. 2016. “From “Duck Factory” to “Fish Factory”: Climate Induced Changes in Vertebrate Communities of Prairie Pothole Wetlands and Small Lakes.” Wetlands 36 (S2): 407–421. doi: 10.1007/s13157-016-0766-3. [DOI] [Google Scholar]
  54. Miles KE, Willis IC, Benedek CL, Williamson AG, and Tedesco M. 2017. “Toward Monitoring Surface and Subsurface Lakes on the Greenland Ice Sheet Using Sentinel-1 SAR and Landsat-8 OLI Imagery.” Frontiers in Earth Science 5. doi: 10.3389/feart.2017.00058. [DOI] [Google Scholar]
  55. Mueller N, Lewis A, Roberts D, Ring S, Melrose R, Sixsmith J, Lymburner L, et al. 2016. “Water Observations from Space: Mapping Surface Water from 25 Years of Landsat Imagery across Australia.” Remote Sensing of Environment 174: 341–352. doi: 10.1016/j.rse.2015.11.003. [DOI] [Google Scholar]
  56. Mushet DM, Rosenberry DO, and Solensky MJ. 2016. Cottonwood Lake Study Area - Water Surface Elevations: U.S. Geological Survey Data Release. doi: 10.5066/F7707ZJ6. [DOI] [Google Scholar]
  57. Mushet DM, Roth CL, and Scherff EJ. 2017. Cottonwood Lake Study Area – Digital Elevation Model with Topobathy: U.S. Geological Survey Data Release. doi: 10.5066/F7V69GTD. [DOI] [Google Scholar]
  58. Nadeau T-L, and Rains MC. 2007. “Hydrological Connectivity between Headwater Streams and Downstream Waters: How Science Can Inform Policy.” Journal of the American Water Resources Association 43 (1): 118–133. doi: 10.1111/j.1752-1688.2007.00010.x. [DOI] [Google Scholar]
  59. NOAA National Climatic Data Center. 2014. “Data Tools: 1981–2010 Normals.” Accessed June 1 2017 http://www.ncdc.noaa.gov/cdo-web/datatools/normals
  60. Ogilvie A, Belaud G, Delenne C, Bailly JS, Bader JC, Oleksiak A, Ferry L, and Martin D. 2015. “Decadal Monitoring of the Niger Inner Delta Flood Dynamics Using MODIS Optical Data.” Journal of Hydrology 523: 368–383. doi: 10.1016/j.jhydrol.2015.01.036. [DOI] [Google Scholar]
  61. Omernik JM, and Griffith GE. 2014. “Ecoregions of the Conterminous United States: Evolution of a Hierarchical Spatial Framework.” Environmental Management 54 (6): 1249–1266. doi: 10.1007/s00267-014-0364-1. [DOI] [PubMed] [Google Scholar]
  62. Ozesmi SL, and Bauer ME. 2002. “Satellite Remote Sensing of Wetlands.” Wetlands Ecology and Management 10: 381–402. doi: 10.1023/A:1020908432489. [DOI] [Google Scholar]
  63. Pekel JF, Cottam A, Gorelick N, and Belward AS. 2016. “High-Resolution Mapping of Global Surface Water and Its Long-Term Changes.” Nature 540: 418–422. doi: 10.1038/nature20584. [DOI] [PubMed] [Google Scholar]
  64. Pham-Duc B, Prigent C, and Aires F. 2017. “Surface Water Monitoring within Cambodia and the Vietnamese Mekong Delta over a Year, with Sentinel-1 SAR Observations.” Water 9 (6): 366. doi: 10.3390/w9060366. [DOI] [Google Scholar]
  65. PRISM Climate Group, Oregon State University. 2012. Accessed July 10 2012 http://prism.oregonstate.edu
  66. Rains MC, Leibowitz SG, Cohen MJ, Creed IF, Golden HE, Jawitz JW, Kalla P, Lane CR, Lang MW, and McLaughlin DL. 2015. “Geographically Isolated Wetlands are Part of the Hydrological Landscape.” Hydrological Processes 30 (1): 153–160. doi: 10.1002/hyp.10610. [DOI] [Google Scholar]
  67. Schlaffer S, Chini M, Dettmering D, and Wagner W. 2016. “Mapping Wetlands in Zambia Using Seasonal Backscatter Signatures Derived from ENVISaT ASaR Time Series.” Remote Sensing 8 (5): 402–424. doi: 10.3390/rs8050402. [DOI] [Google Scholar]
  68. Schmitt A, and Brisco B. 2013. “Wetland Monitoring Using the Curvelet-Based Change Detection Method on Polarimetric SAR Imagery.” Water 5: 1036–1051. doi: 10.3390/w5031036. [DOI] [Google Scholar]
  69. Schofield KA, Alexander LC, Ridley CE, Vanderhoof MK, Fritz KM, Autrey B, DeMeester J, et al. 2018. “Biota Connect Aquatic Habitats Throughout Freshwater Ecosystem Mosaics.” Journal of American Water Resources Association 54 (2): 372–399. doi: 10.1111/1752-1688.12634. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Schumann G, Di Baldassarre G, Alsdorf DE, and Bates PD. 2010. “Near Real-Time Flood Wave Approximation on Large Rivers from Space: Application to the River Po, Northern Italy.” Water Resources Research 46: W05601. doi: 10.1029/2008WR007672. [DOI] [Google Scholar]
  71. Shaw DA, Pietroniro A, and Martz LW. 2013. “Topographic Analysis for the Prairie Pothole Region of Western Canada.” Hydrological Processes 27: 3105–3114. [Google Scholar]
  72. Sheng Y, Song C, Wang J, Lyons EA, Knox BR, Cox JS, and Gao F. 2016. “Representative Lake Water Extent Mapping at Continental Scales Using Multi-Temporal Landsat-8 Imagery.” Remote Sensing of Environment 185: 129–141. doi: 10.1016/j.rse.2015.12.041. [DOI] [Google Scholar]
  73. Spence C, and Phillips RW. 2014. “Refining Understanding of Hydrological Connectivity in a Boreal Catchment.” Hydrological Processes 29 (16): 3491–3503. doi: 10.1002/hyp.10270. [DOI] [Google Scholar]
  74. Tromp-van Meerveld HJ, and McDonnell JJ. 2006. “Threshold Relations in Subsurface Stormflow: 2. The Fill and Spill Hypothesis.” Water Resources Research 42: W02411. doi: 10.1029/2004WR003800. [DOI] [Google Scholar]
  75. Tuia D, Volpi M, Copa L, Kanevski M, and Mufioz-Marl J. 2011. “A Survey of Active Learning Algorithms for Supervised Remote Sensing Image Classification.” IEEE Journal of Selected Topics in Signal Processing 5 (3): 606–617. doi: 10.1109/JSTSP.2011.2139193. [DOI] [Google Scholar]
  76. Turin G 1960. “An Introduction to Matched Filters.” IRE Transactions on Information Theory 6: 311–329. doi: 10.1109/TIT.1960.1057571. [DOI] [Google Scholar]
  77. U.S. Fish and Wildlife Service (USFWS). 2010. National Wetlands Inventory website. Washington, DC: U.S. Department of the Interior, Fish and Wildlife Service; http://www.fws.gov/wetlands/ [Google Scholar]
  78. U.S. Geological Survey (USGS). 2013. The National Hydrography Dataset (NHD). Reston, VA: U.S. Geological Survey; ftp://nhdftp.usgs.gov/DataSets/Staged/States/FileGDB/HighResolution/ [Google Scholar]
  79. Vanderhoof MK, and Alexander LC. 2016. “The Role of Lake Expansion in Altering the Wetland Landscape of the Prairie Pothole Region.” Wetlands 36 (S2): 309–321. doi: 10.1007/s13157-015-0728-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  80. Vanderhoof MK, Alexander LC, and Todd MJ. 2016. “Temporal and Spatial Patterns of Wetland Extent Influence Variability of Surface Water Connectivity in the Prairie Pothole Region, United States.” Landscape Ecology 31 (4): 805–824. doi: 10.1007/s10980-015-0290-5. [DOI] [Google Scholar]
  81. Vanderhoof MK, Distler HE, Mendiola DA, and Lang M. 2017. “Integrating Radarsat-2, Lidar and Worldview-3 Imagery to Maximize Detection of Forested Inundation Extent in the Delmarva Peninsula, USA.” Remote Sensing 9 (2): 105–125. doi: 10.3390/rs9020105. [DOI] [Google Scholar]
  82. Vanderhoof MK, Lane C, McManus M, Alexander L, and Christensen J. 2018. “Wetlands Inform How Climate Extremes Influence Surface Water Expansion and Contraction.” Hydrology and Earth System Science 22: 1851–1873. doi: 10.5194/hess-22-1851-2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  83. Ward JV 1989. “The Four-Dimensional Nature of Lotic Ecosystems.” Journal of the North American Benthological Society 8 (1): 2–8. doi: 10.2307/1467397. [DOI] [Google Scholar]
  84. Werner B, Tracy J, Johnson WC, Voldseth RA, Guntenspergen GR, and Millett B. 2016. “Modeling the Effects of Tile Drain Placement on the Hydrologic Function of Farmed Prairie Wetlands.” Journal of the American Water Resources Association 52 (6): 1482–1492. doi: 10.1111/jawr.2016.52.issue-6. [DOI] [Google Scholar]
  85. White DC, and Lewis MM. 2011. “A New Approach to Monitoring Spatial Distribution and Dynamics of Wetlands and Associated Flows of Australian Great Artesian Basin Springs Using QuickBird Satellite Imagery.” Journal of Hydrology 408 (1–2): 140–152. doi: 10.1016/j.jhydrol.2011.07.032. [DOI] [Google Scholar]
  86. Whiteside TG, and Bartolo RE. 2015. “Use of WorldView-2 Time Series to Establish a Wetland Monitoring Program for Potential Offsite Impacts of Mine Site Rehabilitation.” International Journal of Applied Earth Observation and Geoinformation 42: 24–37. doi: 10.1016/j.jag.2015.05.002. [DOI] [Google Scholar]
  87. Woodcock CE, and Strahler AH. 1987. “The Factor of Scale in Remote Sensing.” Remote Sensing of Environment 21: 311–332. doi: 10.1016/0034-4257(87)90015-0. [DOI] [Google Scholar]
  88. Wu Q, and Lane CR. 2017. “Delineating Wetland Catchments and Modeling Hydrologic Connectivity Using Lidar Data and Aerial Imagery.” Hydrology and Earth System Sciences 21: 3579–3595. doi: 10.5194/hess-21-3579-2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  89. Yao T, Pu J, Lu A, Wang Y, and Yu W. 2007. “Recent Glacial Retreat and Its Impact on Hydrological Processes on the Tibetan Plateau, China, and Surrounding Regions.” Arctic, Antarctic, and Alpine Research 39 (4): 642–650. doi: 10.1657/1523-0430(07-510)[YAO]2.0.CO;2. [DOI] [Google Scholar]
  90. Zhang B, Schwartz FW, and Liu G. 2009. “Systematics in the Size Structure of Prairie Pothole Lakes through Drought and Deluge.” Water Resources Research 45: W04421. doi: 10.1029/2008WR006878. [DOI] [Google Scholar]
  91. Zomer RJ, Trabucco A, and Ustin SL. 2009. “Building Spectral Libraries for Wetland Land Cover Classification and Hyperspectral Remote Sensing.” Journal of Environmental Management 90: 2170–2177. doi: 10.1016/j.jenvman.2007.06.028. [DOI] [PubMed] [Google Scholar]

RESOURCES