|
Using Google Earth Imagery to Target Assessments of Ephemeral Gully Erosion
David A. Reece1, John A. Lory2,*, Timothy L. Haithcoat3, Brian K. Gelder4, Richard Cruse5
Published in Journal of the ASABE 66(1): 155-166 (doi: 10.13031/ja.15254). Copyright 2023 American Society of Agricultural and Biological Engineers.
1Division of Plant Sciences and Technology, University of Missouri, Columbia, Missouri, USA.
2Division of Plant Sciences, University of Missouri, Columbia, Missouri, USA.
3Institute for Data Science and Informatics, University of Missouri, Columbia, Missouri, USA.
4Agricultural Engineering, Iowa State University, Ames, Iowa, USA.
5Agronomy, Iowa State University, USA.
*Correspondence: loryj@missouri.edu
Submitted for review on 30 June 2022 as manuscript number NRES 15254; approved for publication as a Research Article and as part of the Soil Erosion Research Symposium Collection by Associate Editor Dr. Prasad Daggupati and Community Editor Dr. Kyle Mankin of the Natural Resources & Environmental Systems Community of ASABE on 15 December 2022.
Highlights
Abstract. Sustaining civilizations requires preventing the loss of agricultural topsoil through processes such as water and wind erosion. Ephemeral gully erosion contributes an estimated 40% of the total water-erosion soil loss from row-crop fields. The identification and tracking of gullies require monitoring fields over time; Google Earth provides high-quality imagery that can potentially meet both the temporal and spatial criteria for ephemeral gully monitoring. Our primary objective was to determine the probability that an ephemeral gully erosion feature could be reliably identified as an area of concern based on Google Earth imagery and/or other publicly available remotely sensed imagery. To develop the ground truth, we visited 72 fields in seven Missouri counties between mid-April and mid-June in 2018 (n = 26), 2019 (n = 21), and 2020 (n = 25) to verify the presence of erosion features. An unmanned aerial vehicle (UAV) was used to collect aerial imagery with an estimated ground sampling distance of 2.1 to 2.7 cm pixel-1 from all locations. From this imagery, ephemeral gullies were observed in 24 of the fields, and all ephemeral gullies in those fields were delineated. We then reviewed all imagery available in Google Earth from 2010 to 2020 for the 24 fields where ephemeral gullies were observed, delineating ephemeral gully features using a “definitive” and a “less stringent” criterion; we also evaluated 2008 and 2015 imagery from a second public source. In the first analysis, one random erosion feature was chosen from each field, and the percentage overlap of lines derived from publicly available information with the ground truth was determined. Combining all imagery sources, using the less stringent method of delineation, and applying a 15-m buffer resulted in a mean overlap rate of 91%. These results were superior to the definitive approach, as well as using a 3-m buffer. In a second analysis, we tested definitive and less stringent criteria in the field for simple intersection with the ground truth. The less stringent strategy, coupled with using a 15-m buffer, had a true positive rate of 81% and identified 100% of the ephemeral gullies at 63% of the locations. There were false positives in 38% of the fields, with a mean rate across locations of 15%. Adding public data from other sources improved the true positive rate while also increasing the false negative rate. At one location, all publicly available image sources failed to identify the single ephemeral gully in the field. This research represents a proof of concept that Google Earth and other publicly available imagery of sufficient quality can be used to target in-field ephemeral gully assessment in row crop fields in the humid regions of the US. Validation work is needed before this approach can be broadly adopted with confidence, given the many uncontrollable factors that can affect the efficacy of this approach.
Soil erosion rates exceed rates of soil formation in many row-cropped fields in the US and globally (Montgomery, 2007; Cruse et al., 2013; Pennock, 2019). Sustaining agriculture and the societies it supports requires protecting topsoil from both wind and water erosion (Montgomery, 2012; Pennock, 2019; Weltz et al., 2020). Water erosion processes are typically divided into four categories: sheet, rill, ephemeral gully, and classic gully (Foster, 1986; Bernard et al., 2010; USDA-NRCS, 2011; Grigar et al., 2020). Estimates of ephemeral gully erosion range from 10% to 95% of total water-caused erosion, with a median value near 40% (Bernard et al., 2010). Ephemeral gullies are complex structures derived from concentrated water flow attributed to human disturbance, but factors such as catchment size, soil erodibility, slope, and subsurface properties can exacerbate the process (Castillo and Gomez, 2016; Vanmaercke et al., 2021; Wang et al., 2021). Ephemeral gullies can be filled by tillage but, in contrast to rill erosion, tend to subsequently form again in the same locations (Foster, 1986; USDA-NRCS, 2011).
The Food Security Act of 1985 (U.S.C., 1985) requires US farmers to maintain “sustainable erosion rates on cropland, hayland, and pasture” to protect the nation’s long-term capability to produce food and fiber. The act requires farmers to implement conservation practices on fields designated as “highly erodible land” (HEL), an assessment based on soil type attributes within the field. The Natural Resource Conservation Service (NRCS) is the primary agency documenting farmer compliance with HEL regulations; NRCS personnel assess thousands of agricultural fields annually. A 2010 report (Bernard et al., 2010) stated “the Agency does not have a tool to predict and quantify ephemeral gully erosion, including the potential for nonstructural practices to control the erosion.” To ensure that current management is preventing the formation of ephemeral gullies, NRCS personnel currently walk the field perimeters of all assessed tracts (R. Miller, former Missouri NRCS State Agronomist, personal communication).
The identification and medium- to long-term monitoring of gully development requires observing fields across time (Vanmaercke et al., 2021). Google Earth provides high-quality imagery that can potentially meet both the temporal and spatial criteria for ephemeral gully monitoring. The capacity of Google Earth to study gully erosion has received attention in recent years but remains relatively undocumented. Gilad et al. (2012) used Google Earth imagery to map gullies that contributed to soil erosion near the Great Barrier Reef. Boardman (2016) explored the usefulness of Google Earth for mapping erosion features, including ephemeral gully erosion in agricultural fields in southern England, and focused on areas with known erosion. Karydas and Panagos (2020) expanded the use of Google Earth to a country-level analysis. Their study focused on 100 randomly selected agricultural sites in Greece. They used Google Earth to digitize ephemeral gully features that were visible in historical imagery, mostly from the last decade. This data was then used to estimate the potential for ephemeral gully erosion by land use type. A number of other studies have utilized Google Earth for erosion mapping and modeling (Frankl et al., 2013; Liu et al., 2018; Sheshukov et al., 2018). These studies used Google Earth as a starting point to identify locations where there is evidence of erosion, which was then used for further analysis and modeling.
The studies above began with using Google Earth to identify ephemeral gully occurrences and then used a variety of methods to confirm the presence of ephemeral gullies on the ground at these locations. This process has two assumptions: (1) that Google Earth can be used to accurately identify ephemeral gullies, and (2) that delineators consistently and accurately use their judgment when identifying ephemeral gullies. Maugnard et al. (2014) detailed many of the issues with the latter assumption, clearly demonstrating the need for careful and deliberate criteria to limit errors in judgment when delineating ephemeral gullies from aerial imagery. There remains a need, however, for research that explores how accurately Google Earth, potentially coupled with other sources of historical imagery, identifies ephemeral gullies when tested against pre-existing ground truth data. A successful methodology using these sources has the advantage of simplicity and utilizes a widely available and low-cost option for monitoring. To successfully implement such a system, assessment criteria need to be defined, and their success and failure rates in identifying ephemeral gullies need to be evaluated. A successful targeting system must reliably identify a high percentage of ephemeral gullies that exist in a field. However, a successful Google Earth protocol must also have sufficient discriminatory capabilities to substantially reduce the effort needed to confidently assess ephemeral gully erosion.
This project had two primary objectives. The first was to determine the probability that the location of an ephemeral gully erosion feature identified in ground truth imagery can be reliably identified using Google Earth and other publicly available historical imagery. The second objective was to determine the success and failure rates of our assessment criteria at the field scale in locations where ground truth had identified at least one ephemeral gully. From this assessment, we suggest strategies and situations where Google Earth imagery is likely to be an effective tool for targeting ephemeral gully erosion in row crop agriculture fields.
Materials and Methods
Field Site Descriptions and Data Collection
NRCS personnel identified and obtained permission for the project team to access 72 fields in seven Missouri counties: Audrain, Boone, Callaway, Cooper, Howard, Moniteau, and Monroe. All fields were planted with (or were to be planted with) corn (Zea maize L.) or soybean (Glycine max (L.) Merr.). Typically, fields were assessed prior to corn reaching the V4 stage and soybean reaching the V3 stage. All fields were classified as HEL by the NRCS. Data were collected between mid-April and mid-June in 2018 (n = 26), 2019 (n = 21), and 2020 (n = 25). Aerial imagery was obtained from all locations with a DJI Phantom 4 PRO unmanned aerial vehicle (UAV; DJI, Shenzhen, China). The stock camera had a 24-mm lens and 20-megapixel resolution, which generated an image of 5,472 x 3,648 pixels. The imagery was typically obtained at 100 m above ground level (AGL), though at some sites, 75 m AGL was used. The estimated ground sampling distance (GSD) was 2.1 cm pixel-1 and 2.7 cm pixel-1 for the 75- and 100-m images, respectively. All images were captured looking directly downward (90° to flat ground, or nadir perspective). The flight control software was Map Pilot Pro (Drones Made Easy, San Diego, CA), with the image overlap set at 80% within and between passes. A single orthomosaic was created from the collected images using Pix4D Mapper (Prilly, Switzerland). Additionally, imagery was obtained at each location to estimate residue cover using a bullseye grid method (Lory et al., 2021). Residue estimates and methods for 2018 and 2019 were previously reported (Lory et al., 2021; Upadhyay et al., 2022). For the 2020 locations only, the residue was estimated from images obtained at 2 m AGL (estimated GSD = 0.06 cm pixel-1) using the same UAV that obtained images at 75 m or 100 m. Ten images were analyzed using the bullseye grid method (average of two readers evaluating a unique 100-point grid; a total of 200 points). The location’s residue estimate was the mean of these 10 estimates.
The UAV imagery from the 72 locations was reviewed to select the final locations to use as the ground truth data set. Five locations were eliminated due to problems with the quality of the collected imagery. The remaining 67 images were examined for clear evidence of active, ephemeral gully erosion features. Four of these were not delineated because the evidence of erosion was only seen amidst terraced slopes. Thirty-nine locations had no evidence of ephemeral gully erosion. The remaining 24 locations had at least one ephemeral gully. Site details for the 24 locations used are reported in table 1.
| Table 1. Location, assessment date, and crop and ground cover for 24 field sites. For privacy reasons, GPS coordinates are approximate. Listed by assessment date. | ||||||||
| Assessment Date | ID | Missouri County | Latitude[a] | Longitude[a] | Crop[b] | Dominant Residue[c] | Residue (%) | Field Size (ha) |
| 4-May-18 | 19 | Audrain | 39.26 | -92.26 | NE | SB, C | 81 | 20.4 |
| 14-May-18 | 20 | Audrain | 39.26 | -92.21 | C (V3) | SB | 7 | 17.6 |
| 24-May-18 | 22 | Boone | 39.21 | -92.21 | C (V5) | SB | 19 | 7.5 |
| 24-May-18 | 23 | Boone | 39.21 | -92.21 | C (V4) | SB | 11 | 21.9 |
| 29-May-18 | 21 | Audrain | 39.26 | -92.14 | SB (V2) | C | 23 | 28.2 |
| 17-May-19 | 14 | Howard | 39.08 | -92.62 | C (V2) | SB | 9 | 5.9 |
| 29-May-19 | 15 | Howard | 39.01 | -92.57 | SB (V1) | SB | 64 | 19.7 |
| 11-Jun-19 | 18 | Audrain | 39.33 | -91.75 | NE | NE | 64 | 17.5 |
| 5-May-20 | 1 | Boone | 38.88 | -92.21 | NE | SB | 61 | 6.6 |
| 5-May-20 | 6 | Callaway | 38.94 | -91.81 | NE | C, GW | 73 | 8.4 |
| 12-May-20 | 2 | Boone | 38.88 | -92.21 | C (V1) | SB | 43 | 2.1 |
| 12-May-20 | 3 | Boone | 38.91 | -92.21 | C (V1) | SB | 43[d] | 7.7 |
| 12-May-20 | 4 | Boone | 38.91 | -92.21 | C (V1) | SB | 43 | 2.5 |
| 20-May-20 | 16 | Callaway | 38.94 | -91.81 | SB (V1) | C, GW | 74 | 92.7 |
| 8-Jun-20 | 24 | Boone | 39.11 | -92.24 | SB (V) | WSG | 73 | 31.2 |
| 11-Jun-20 | 5 | Boone | 39.19 | -92.12 | SB (V1) | C | 66 | 27.0 |
| 12-Jun-20 | 17 | Audrain | 39.17 | -92.07 | C (V3) | SB | 6 | 9.9 |
| 13-Jun-20 | 7 | Audrain | 39.26 | -92.18 | NE | C | 19 | 19.5 |
| 14-Jun-20 | 8 | Audrain | 39.25 | -92.18 | SB (V1) | C | 18 | 13.1 |
| 15-Jun-20 | 9 | Audrain | 39.25 | -92.18 | SB (V1) | WSG | 71 | 18.4 |
| 16-Jun-20 | 10 | Boone | 39.22 | -92.18 | C (V2) | SB, GW | 4 | 8.4 |
| 17-Jun-20 | 12 | Boone | 39.17 | -92.29 | SB (V1) | WSG, GW | 77 | 7.6 |
| 18-Jun-20 | 11 | Boone | 39.21 | -92.18 | C (V2) | SB | 2 | 9.9 |
| 19-Jun-20 | 13 | Audrain | 39.21 | -91.62 | C (V3) | WSG, GW | 42 | 10.8 |
[a] Approximate location in decimal degrees. [b] Planted crop (and growth stage): C=corn, SB=soybean; NE=not emerged; V=vegetative stage. [c] Dominant residue types: C=corn, SB=soybean; WSG=winter small grain; GW=green weeds. [d] Residue estimate from neighboring field under same management, location 4. | ||||||||
Ephemeral Gully Delineation Process from Ground Truth Imagery
To delineate a data set from the UAV-derived ground truth images, we developed a protocol for identifying the presence of ephemeral gully erosion. Our goal was to delineate erosion features from the point that rill erosion transitioned to ephemeral gully erosion (the start) to the point where the feature exited the field, or the flow path was no longer discernible within the field (the end). We defined ephemeral gully erosion features as having clear evidence that surface water flow created a path that cut through normal surface patterns in the field. Such evidence included: (1) displacement and clearing of surface residue along a path, (2) a trail cutting across or eliminating surface patterns such as the direction of tillage, wheel tracks, and planted rows, and (3) evidence of soil removal, such as a knick point at the initiation of the feature where water has cut through the surface soil and/or the sharpness of a channel edge associated with depth below the soil surface. When significant erosion occurred along a wheel track or other similar farming-induced water channeling feature, it typically was not considered ephemeral gully erosion unless it changed direction in response to topography. When a feature exhibited a pattern analogous to braided stream flow, the dominant path was followed, or, if this was not clear, the central pathway of the feature was delineated. Field boundary polygons were created to define the extent of the field that was visible within the image. Within each of these field boundaries, all observed ephemeral gully features were digitized using the georeferenced UAV imagery displayed in ArcGIS (ESRI, Redlands, CA). A single dendritic pattern was delineated by merging multiple lines from all the initiation points for the ephemeral gully within the field (fig. 1). Figures of the resulting ephemeral gully delineations (defined as the ground truth data set [GT]) overlain on the UAV imagery for the 24 fields are reported in the Supplemental Information.
Ephemeral Gully Delineation Process from Google Earth Imagery
The project also delineated ephemeral gully erosion using publicly available imagery. The primary source tested was the imagery available in Google Earth Pro (Google Earth LLC, Singapore). Google Earth uses a wide variety of imagery from different sources and provides little to no metadata directly (Google Earth Help, 2021). Sources include aerial imagery provided by state and federal agencies (e.g., USDA and FSA), public satellite data (mostly Landsat and Sentinel), and commercial satellite data (e.g., DigitalGlobe and Maxar Technologies). The average satellite GSD from the commercial sources was approximately 0.5 m, and 10 to 15 m from the publicly available satellite sources. For this project, we needed imagery that was typically <1.0 m GSD. An analyst decided which images had a sufficient resolution for the assessment. For most of the study area, Google Earth imagery from prior to 2010 was not used because it typically had an insufficient resolution for the project objectives. Ultimately, we assessed Google Earth imagery dated between 2010 and 2020 for its project suitability. All imagery was viewed in ‘natural color,’ with the viewing angle set to a nadir perspective.
 |
| Figure 1. Example of ground truth line delineating dendritic ephemeral gully from UAV imagery at location 6. |
For each of the 24 fields identified in the GT data as having ephemeral gully erosion, we developed a Google Earth line delineation protocol to identify all potential ephemeral gullies in the field. First, for each location, the date of all available imagery during the assessment period (2010 to 2020) was listed in a spreadsheet. The area defined by each field boundary was then assessed for the presence or absence of ephemeral gully erosion for the date of each image. Then every potential erosion feature observed in the target field was delineated in each available image from Google Earth and the strength of the evidence was classified into three categories: definitive, less stringent, and/or suggestive evidence of ephemeral gully erosion (fig. 2). Definitive evidence paralleled ground truth evidence, requiring clear indication of (1) displacement and clearing of surface residue along a path, (2) a trail cutting across or eliminating surface patterns such as the direction of tillage, wheel tracks, and planted rows, and (3) evidence of soil removal, such as a knick point at the initiation of the feature where water has cut through the surface soil and/or sharpness of a channel edge associated with depth below the soil surface (figs. 2 and 3). Less stringent evidence typically consisted of a path defined by residue displacement (1) and/or a path cutting across field patterns (2), but the sharpness of the evidence of soil movement was less clear (figs. 2 and 3). Less stringent evidence also included images taken later in the growing season when a crop on the field masked a clear view of the soil, but there was a clear disruption of the cropping pattern associated with the potential ephemeral gully. Suggestive evidence included recent tillage targeted at the area of a suspected erosion feature or patterns in the field such as wetness and color differences consistent with previous erosion along the pathway that supported the case that an ephemeral feature might reform in the same area (fig. 2). The same image could have more than one category of evidence; parts of one or more potential ephemeral gully paths might be categorized as definitive, while other sections or other lines might be defined as less stringent.
After all the available imagery dates were classified for each site, the next step was to delineate the full extent of all features in the field where there was unequivocal evidence of ephemeral gully erosion in Google Earth, defined as the “Google Earth definitive” data set (GED; fig. 3). Reviewing all images designated with definitive evidence of ephemeral gullies for a location, we delineated a line that represented the maximum extent of definitive evidence as observed across all temporally relevant images. This could lead to gaps in the field along a flow path where definitive criteria were not met. Lines were also terminated when they reached the edge of the target field boundary.
After the delineation of the GED data set, a second “Google Earth less stringent” set of lines (GELS) was created (fig. 3). Using the GED lines as the starting point, feature lines were extended, when justified, based on the maximum extent of the less stringent criteria. As with the GED lines, the GELS lines represented the maximum extent derived from all images based on the less stringent criteria. Again, any features extending beyond a field boundary were terminated when they reached the edge of the target field. There were locations where no ephemeral gully feature met the definitive criteria but did meet the less stringent conditions. For record keeping purposes, if multiple converging segments of a water flow path in the field were delineated, they were considered one ephemeral gully; counts were based on complete dendritic pathways, not the number of segments that met a specific criterion.
All lines were digitized using the “path” tool in Google Earth Pro. For both the definitive (GED) and less stringent (GELS) features, the pathways were combined to represent the most consistent and/or definitive evidence if there were deviations in the exact path among images. The result for each field location was two sets of dendritic lines, one for GED and one for GELS. After delineation, both sets of lines were exported from Google Earth Pro as KML files, imported into ArcGIS, and converted to feature classes.
Ephemeral Gully Delineation Process from Other Publicly Available Imagery
Two other sets of publicly available imagery were used as alternative sources to Google Earth and are available through the Missouri Spatial Data Information Service (MSDIS). Private contractors collected images in the spring of 2007, 2008, and 2009 (designated as the 2008 dataset), and then again in spring of 2015 and 2016 (designated as the 2015 dataset). Images were captured by aircraft at altitudes ranging from 7,900 to 8,800 meters above the ground. The 2008 images are 3-band (RGB) natural color, while the 2015 images include a fourth near-infrared band (RGB-IR). Spatial resolution is mostly 0.6 m GSD for 2008 and 0.46 m GSD for 2015, and both were collected in the local Universal Transverse Mercator (UTM) projection with the North American Datum of 1983 (NAD83).
 |
| Figure 2. Examples of definitive, less stringent, suggestive, and no evidence of ephemeral gully erosion from Google Earth imagery identified between 2011 and 2018. This ephemeral gully was in location 14. |
 |
| Figure 3. Examples of definitive (GED) and less stringent (GELS) evidence of ephemeral gully erosion from Google Earth imagery identified from 2012 and 2014. Lower image shows extent of GED and GELS line delineation for these images and others. This ephemeral gully was in location 22. |
From this source, two additional sets of lines were created, one derived from the MSDIS 2008 imagery (M08) and one from the MSDIS 2015 imagery (M15). The apparent lower resolution of these image sources dictated using “less stringent” criteria to delineate lines. Similar to the GT lines, the dendritic line sets were digitized using ArcGIS. Only features within the ground truth visual field boundary were delineated.
Delineations Quality Control
Before delineating lines using these methods, there was extensive interaction between the three people who digitized lines and/or reviewed imagery to calibrate and align the approach. All delineated lines were reviewed by a second individual. When there were discrepancies between the digitizer and the reviewer, the lines were referred for further evaluation by the team to resolve the differences. The end result was five sets of lines for each of the 24 fields where ephemeral gullies were observed in the ground truth imagery: GT, GED, GELS, M08, and M15.
Analysis One: Likelihood That Publicly Available Imagery Identified a Known Ephemeral Gully
Data Development for Delineated Line Overlap Comparisons
Our first objective was to estimate the likelihood that a known ephemeral gully would have been identified based on publicly available imagery. We did this by determining the degree of overlap between known ephemeral gullies (GT) and the delineated lines from the various sources of publicly available information (GED, GELS, M08, and M15).
From the 24 fields where we observed at least one ephemeral gully, one GT gully was randomly selected from each field (n = 24). An area of interest (AOI) was then established by creating a 40-m buffer around the selected GT line (fig. 4a). The area of interest had the sole purpose of limiting the distance evaluated for comparison with GT using erosion lines derived from publicly available imagery.
The AOI was clipped to remove portions outside of the field boundary. A 2 x 2 meter grid was established and oriented north/south in each field. Next, the GT line was converted to a raster using a basic binary classification, where any pixel touching the line was reclassified from a value of zero to a value of one (fig. 4b). The next step calculated the Euclidean distance from the centroid of each pixel in the AOI to the centroid of the nearest pixel in the GT line; pixels of the GT line were assigned a Euclidean distance of zero. Finally, the Euclidean distance raster data was converted to point data associated with the spatial coordinates of the center of the pixel (fig. 4c).
 |
| Figure 4. (a) Digitized GT line with 40-m AOI, (b) GT vector line converted to 2 x 2-m raster, and (c) Rasterized GT line with Euclidean distance raster, showing pixels extracted using the AOI. This ephemeral gully was in location 12. |
A similar process was repeated for each of the other four line sources (GED, GELS, M08, and M15). The main difference was that the evaluation process was based on the ground truth-defined AOI. Each delineated target line type (GED, GELS, M08, and M15) for the field was overlain with the AOI. If two or more separate lines were selected by this process, the line nearest to the GT line was retained. From that point on, the process was identical to that used for the ground truth. Each publicly available line selected was converted to a raster and reclassified. A raster with the Euclidean distance from the centroid of every pixel in the AOI to the centroid of the nearest pixel of that line type was created in alignment with the 2 x 2 meter grid. If the line type was not present in the AOI, all points for that line type were assigned a value of 9999. Finally, each of the four rasters was converted to point data associated with the spatial coordinates of the center of the pixel.
The data from the 24 AOIs were then combined into a single table, with one row of data for each point in the grid, representing each pixel in all of the line type rasters. The data reported for each pixel included the location ID, the grid point ID, the spatial coordinates of the pixel centroid, and the Euclidian distance of that grid point to each of the five line types: GT, GED, GELS, MO8, and M15. All steps were done using geoprocessing tools in ArcGIS and exported as a CSV file.
Evaluating Overlap of Publicly Available Line Sources with Ground Truth
We tested multiple strategies based on the GED, GELS, M08, and M15 lines to predict the location of GT lines. The initial raster lines were one pixel (2 x 2 m) based on the delineated line touching that pixel. Using the Euclidean distance information, we added a one-pixel buffer around the initial raster lines, expanding the base line width to three pixels, referred to henceforth as the 3-m buffer. Using binary classification, all pixels with a Euclidean distance of <2.83 m (the diagonal distance across one 2-m square pixel) within the AOI were assigned a value of one. This was completed separately for the ground truth and the four initial line types derived from publicly available information (See visualization in fig. 5; grey (GT) and white [GED, GELS, M08, and M15]). When a given line type did not exist within the AOI, all cells were set to zero. The buffered lines ensured that small differences in location accuracy and precision in the different sources of imagery did not interfere with line comparisons.
The remaining data development steps were applied only to the publicly available imagery data sets (M08, M15, GED, and GELS). First, two additional line categories derived from publicly available imagery were created by combining information from more than one line type. We created a line set of the combined data derived from the MSDIS datasets (MSD; M08 plus M15). We also created a dataset of all publicly available sources combined (ALL; MSD plus GELS). To combine line types, pixels in the AOI for each line type were added together, and any pixel with a sum >1 was set to one, and the other pixels were set to zero.
For each of these six line sets derived from publicly available imagery (M08, M15, GED, GELS, MSD, and ALL), we calculated a second “fuzzy” line buffer of 15 m from the center of the initial line raster pixels (fig. 5; orange pixels). Using binary classification, all pixels with a Euclidean distance of <14.15 m (the diagonal distance across seven 2-m square pixels) within the AOI were assigned a value of one. The “fuzzy” buffer was chosen as a reasonable distance within which a feature might be clearly visible when scouting a field for ephemeral gullies, either on foot or flying a drone at a low altitude of 20 to 30 m AGL.
We evaluated the success of each strategy (GED, GELS, M08, M15, MSD, and ALL) by assessing the overlap of each of the two (3-m and 15-m) buffered publicly available lines with the 3-m buffered GT lines. For each of the line sets, all pixels were multiplied with the GT pixel values, resulting in a value of one when the lines overlapped with the GT line and a value of zero for the rest of the pixels in the AOI where there was no overlap (see visualization of overlaps in fig. 5). For each assessment, the number of pixels equal to one was determined for each AOI and then divided by the number of pixels in the GT 3-m buffer and multiplied by 100 to convert to a percentage of overlap in the AOI. All calculations were done using SAS (ver. 9.4, Cary, NC).
 |
| Figure 5. Visualization of overlap percentages of GT line with two buffered types of publicly available lines. Rasterized GT line (black) with one-pixel (3 m) buffer (grey) shown overlaying Euclidean distance raster for each publicly available line type (GED, GELS, M08, and M15 in rows a, b, and c, respectively). For each publicly available line type, white is the 3-m buffer, orange is 15-m buffer, and blue is all remaining distances within AOI. (a) top: location 3, (b) middle: location 12, and (c) bottom: location 19. |
Analysis Two: Assessment of Field-based Success Rate of Identifying Ephemeral Gullies
Our second objective was to assess the percentage of true positives, false negatives, and false positives within a field. First, for each field, we recorded the total number of ephemeral gullies observed in the ground truth in each field. Second, we applied the 15-m buffer to all of the lines identified by each of the six publicly available line types (GED, GELS, M08, M15, MSD, and ALL) within the field boundary. We then recorded, for each line type, the total number of ephemeral gullies in the ground truth that intersected with any part of the buffered line. This value, divided by the total number of GT lines in the field and multiplied by 100, was the true positive success rate for that line type. The false negative rate was the inverse of the true positive rate, where existing GT lines failed to intersect with the other line type being evaluated, based on a 15-m buffer. We then recorded the number of times a 15-m buffered publicly available line failed to intersect any GT line. This value, divided by the total number of that line type and multiplied by 100, was the false-positive failure rate. This assessment was done for each of the 24 fields where at least one ephemeral gully was observed. A true negative count could not be assessed by our field-by-field methodology. All assessments were done using ArcGIS.
Results and Discussion
Assessment of Imagery Availability
In the 24 fields where we observed at least one ephemeral gully, the number of Google Earth images available ranged from three to nine (mean = 6.2; median = 6.0; STD = 1.5) per field. Figure 6 details the percentage of the locations in a county that had imagery with suitable image quality for ephemeral gully evaluation, grouped by seasonal 3-month quarters for the 11-year timeframe. Of the 44 quarters, 26 had no imagery at any location. At the county level, images were frequently available for only a subset of the county's locations. In Audrain (n = 9) and Boone (n = 11) counties, there was only one-quarter of the time that all the locations in those counties all had available imagery at the same time. This confirms that Google Earth imagery provided multiple images of each location but was not equally or uniformly represented within or among locations. This variation suggested the chances of observing a potential ephemeral gully using Google Earth varied with location.
 |
| Figure 6. Percentages of locations within county with Google Earth imagery by quarter. |
The number of images having GED and/or GELS evidence of ephemeral gullies at a location ranged from zero to seven (mean = 4.5; STD = 2.1) or 0 to 100% of the available images (mean = 72%; STD = 28). One of the 24 locations had no evidence of ephemeral gullies in any of the five Google Earth images. Seven locations (29%) showed evidence of an ephemeral gully in all available images. Twenty-two of 24 locations (92%) had evidence of ephemeral gullies in more than one image. For the two Missouri image sets, there was evidence of ephemeral gullies in 22 of the 24 fields for the 2008 image set and in 21 of the 24 fields for the 2015 data set.
A total of 147 ephemeral gullies were identified in the ground truth images of the 24 fields. The average number per field was 6 (STD = 7.4) but two locations (16 and 24) had over one-third of the ephemeral gullies between them. Five locations had one ephemeral gully. The two largest fields had the most ephemeral gullies, otherwise, there was little relationship between field size and the number of ephemeral gullies when those two locations were removed from the analysis.
Analysis One Results: Likelihood That Publicly Available Imagery Identified a Known Ephemeral Gully
Our first analysis considered how frequently publicly available information failed to identify an existing ephemeral gully (false negatives). A complete failure resulted in 0% overlap for the line type being evaluated (table 2). The GED (definitive) standard failed to identify 21% of the ephemeral gullies (n = 5). Using less stringent criteria (GELS) reduced the failure rate to 8%. The MSDIS Missouri imagery (less stringent criteria) had failure rates of 13% and 17% for the 2008 and 2015 imagery, respectively. This failure rate dropped to 4% (one location) when data from both years were combined (MSD). Combining the GELS and the MSD (ALL) also resulted in only one failure to identify an ephemeral gully. One location, 18, was not identified by either Google Earth or MSDIS imagery. This location had one ephemeral gully of a limited extent (~20 m). The historic record of images (n = 5) at this location documented high residue with suggestive evidence of erosion in 2010 only, and all but one image came from 2014 or earlier. We measured 81% residue at this location when we assessed the field in 2019 (table 1). A subsequent image from Google Earth dated May 2022 showed no evidence of ephemeral gullies. This location may represent an unusual and small breakdown in conservation practices, perhaps due to an extreme event, in a field where history suggests this was unexpected.
The most effective stand-alone publicly available imagery data set, assessed based on percentage line overlap, was the GELS approach (table 2). Compared to the GED strategy, the less stringent criteria increased the percentage coverage of ground truth in 19 of 24 locations, including reducing the number of locations where Google Earth imagery failed from five to two locations. When the MSDIS imagery was analyzed individually, it performed the worst, but when coupled together (MSD; table 2), it had better success than GED. The best results were obtained by combining data from all sources (ALL = GELS plus MSD). Increasing the buffer distance from 3 m to 15 m increased the average percentage overlap by nearly 20 percentage points when averaged across all six line types (table 2). When the 15-m buffer was used on the ALL line set, the average overlap was 91%, with 15 of the 24 locations (63%) fully overlapping the ground truth. Using GELS alone resulted in the same number of fully identified ground truth locations (63%), but overall coverage was reduced to 86%. This decrease was due to the increasing number of ephemeral gullies missed from one to two and the reduced coverage percentage in four locations (table 2). These results suggest that more sources of data, achieved through combining Google Earth and MSDIS data, and less stringent criteria, achieved through using the GELS standard and the 15-m buffer, improved the likelihood that publicly available imagery identified the location of existing ephemeral gullies.
| Table 2. Percentage of overlap of ground truth with erosion lines derived from publicly available imagery. Sources included Google Earth (definitive, GED; less stringent, GELS) and MSDIS (2008, M08; 2018, M15), plus the combinations of M08 and M15 (MSD) and GELS and MSD (ALL). Overlap reported for line buffer of 3 m and 15 m for the publicly obtained line. | |||||||||||||||||
| GED | GELS | M08 | M15 | MSD | ALL | ||||||||||||
| 3m | 15m | 3m | 15m | 3m | 15m | 3m | 15m | 3m | 15m | 3m | 15m | ||||||
| ID | Percent (%) | ||||||||||||||||
| 1 | 91 | 100 | 95 | 100 | 82 | 100 | 80 | 100 | 88 | 100 | 98 | 100 | |||||
| 2 | 0 | 0 | 85 | 100 | 74 | 100 | 73 | 100 | 83 | 100 | 92 | 100 | |||||
| 3 | 55 | 78 | 58 | 78 | 15 | 31 | 20 | 34 | 21 | 34 | 59 | 78 | |||||
| 4 | 37 | 66 | 57 | 79 | 58 | 94 | 53 | 80 | 63 | 94 | 70 | 94 | |||||
| 5 | 0 | 0 | 0 | 0 | 43 | 74 | 0 | 0 | 43 | 74 | 43 | 74 | |||||
| 6 | 78 | 100 | 81 | 100 | 59 | 85 | 54 | 89 | 75 | 99 | 92 | 100 | |||||
| 7 | 45 | 63 | 69 | 98 | 49 | 75 | 53 | 75 | 66 | 86 | 82 | 98 | |||||
| 8 | 81 | 100 | 82 | 100 | 21 | 62 | 0 | 0 | 21 | 62 | 87 | 100 | |||||
| 9 | 48 | 71 | 52 | 73 | 0 | 0 | 66 | 84 | 66 | 84 | 73 | 84 | |||||
| 10 | 55 | 100 | 55 | 100 | 25 | 72 | 56 | 100 | 58 | 100 | 58 | 100 | |||||
| 11 | 0 | 0 | 33 | 60 | 4 | 7 | 0 | 0 | 4 | 7 | 34 | 60 | |||||
| 12 | 41 | 72 | 79 | 100 | 43 | 93 | 91 | 100 | 93 | 100 | 95 | 100 | |||||
| 13 | 91 | 100 | 95 | 100 | 80 | 98 | 80 | 100 | 88 | 100 | 98 | 100 | |||||
| 14 | 69 | 100 | 77 | 100 | 67 | 100 | 81 | 100 | 82 | 100 | 83 | 100 | |||||
| 15 | 15 | 43 | 87 | 100 | 0 | 0 | 32 | 66 | 32 | 66 | 87 | 100 | |||||
| 16 | 55 | 83 | 62 | 91 | 27 | 49 | 26 | 50 | 33 | 52 | 70 | 92 | |||||
| 17 | 0 | 0 | 49 | 74 | 57 | 94 | 40 | 59 | 70 | 94 | 75 | 94 | |||||
| 18 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | |||||
| 19 | 80 | 100 | 84 | 100 | 33 | 82 | 23 | 50 | 40 | 82 | 89 | 100 | |||||
| 20 | 91 | 100 | 91 | 100 | 62 | 89 | 84 | 100 | 93 | 100 | 98 | 100 | |||||
| 21 | 87 | 100 | 89 | 100 | 79 | 100 | 98 | 100 | 100 | 100 | 100 | 100 | |||||
| 22 | 86 | 100 | 91 | 100 | 41 | 95 | 72 | 100 | 81 | 100 | 95 | 100 | |||||
| 23 | 61 | 100 | 71 | 100 | 24 | 77 | 58 | 100 | 60 | 100 | 74 | 100 | |||||
| 24 | 86 | 100 | 86 | 100 | 39 | 48 | 82 | 99 | 87 | 99 | 89 | 100 | |||||
| Mean | 52 | 70 | 68 | 86 | 41 | 68 | 51 | 70 | 60 | 81 | 77 | 91 | |||||
| Zeros | 21 | 21 | 8 | 8 | 13 | 13 | 17 | 17 | 4 | 4 | 4 | 4 | |||||
Analysis Two Results: Assessment of Field-based Success Rate of Identifying Ephemeral Gullies
Our second objective was to assess the percentage of true positives and false negatives within a field. For this analysis, we compared the success or failure based on all the observed ephemeral gullies within a location with all lines derived from the publicly available data. Based on the results in the previous section, we assessed within-field success using the 15-m buffers, and the standard for success was a simpler binary one: did the feature intersect at any point with a GT feature or not?
The GED true positive rate (56%) was lower than the GELS true positive rate (81%; table 3). The GED did have a lower false positive rate (7% versus 19%). However, the stringent GED criteria more frequently failed to identify ephemeral gullies across locations, and it was more likely to miss gullies within a location. Clearly, the suggestive criteria of the GELS approach were more suited to a targeting tool where the greater penalty should be for missing existing gullies. The GELS approach identified 100% of the gullies in 15 locations while failing to identify a single gully at one location (location 18 discussed above). The GELS approach could be successful when there were many ephemeral gullies in a field; the six locations with eight or more ephemeral gullies had a mean true positive rate of 88% with a false positive rate of 12%.
Adding the MSD lines to the GELS lines increased the true positive rate by almost 10% (table 3). The true positive percentage increased in seven of the nine locations that had less than 100% identification with GELS. The cost was that the false positive rate increased at ten locations, while only going down at two; the mean false positive rate increased from 15% to 25%. When we evaluated M08 and M15 separately, M08 generated false positives at three more locations, and the false positive rate was 16% compared to 12% for M15. This suggested that the older imagery (2008) may not have been as representative as the more recent imagery (2015).
The most damaging error is a false negative, where an existing ephemeral gully fails to be identified by a targeting system. Relying on the GELS or ALL strategies produced false negative rates of 19% and 11%, respectively. Notably, using either strategy, one field had no ephemeral gullies, a result that would suggest that field did not need to be monitored. In all other cases, while the system did not always identify all ephemeral gullies in the field, it targeted at least one ephemeral gully in a field, and typically identified all the present ephemeral gullies. In some fields, there were also false positives (38% of locations for GELS; 63% of locations for ALL), though in most cases, this was one or two false positives at a location. False positives represent extra effort expended for features that were not present, a less problematic error than false negatives, given their scale. In our approach, we were able to assess true positives, false positives, and false negatives; the methodology could not determine true negatives. Consequently, overall accuracy and other statistics associated with confusion matrices could not be determined.
| Table 3. Whole-field line counts, true positives, and false positives for two strategies for using publicly available data to assess ephemeral gully erosion (GELS, Google Earth less stringent; ALL, GELS plus MSDIS data). Line counts also reported for the ground truth. | |||||||||
| GT | GELS | ALL | |||||||
| ID | Line Count | Line Count | True Positives | False Positives | Line Count | True Positives | False Positives | ||
| % | % | % | % | ||||||
| 1 | 4 | 2 | 50 | 0 | 3 | 75 | 0 | ||
| 2 | 8 | 10 | 100 | 20 | 10 | 100 | 20 | ||
| 3 | 3 | 3 | 100 | 0 | 3 | 100 | 0 | ||
| 4 | 3 | 3 | 100 | 0 | 3 | 100 | 0 | ||
| 5 | 6 | 1 | 17 | 0 | 9 | 100 | 33 | ||
| 6 | 4 | 4 | 100 | 0 | 5 | 100 | 20 | ||
| 7 | 2 | 4 | 100 | 50 | 5 | 100 | 60 | ||
| 8 | 7 | 2 | 14 | 50 | 4 | 43 | 25 | ||
| 9 | 8 | 8 | 100 | 0 | 8 | 100 | 0 | ||
| 10 | 1 | 3 | 100 | 67 | 3 | 100 | 67 | ||
| 11 | 1 | 1 | 100 | 0 | 1 | 100 | 0 | ||
| 15 | 2 | 3 | 100 | 33 | 4 | 100 | 50 | ||
| 16 | 3 | 2 | 67 | 0 | 4 | 100 | 25 | ||
| 14 | 1 | 1 | 100 | 0 | 3 | 100 | 67 | ||
| 15 | 12 | 12 | 100 | 0 | 12 | 100 | 0 | ||
| 16 | 31 | 21 | 68 | 0 | 23 | 71 | 4 | ||
| 17 | 2 | 2 | 100 | 0 | 2 | 100 | 0 | ||
| 18 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | ||
| 19 | 12 | 16 | 67 | 50 | 20 | 92 | 45 | ||
| 20 | 1 | 2 | 100 | 50 | 2 | 100 | 50 | ||
| 21 | 3 | 3 | 100 | 0 | 5 | 100 | 40 | ||
| 22 | 3 | 3 | 67 | 33 | 4 | 67 | 50 | ||
| 23 | 5 | 6 | 100 | 17 | 11 | 100 | 55 | ||
| 24 | 24 | 22 | 92 | 0 | 23 | 96 | 0 | ||
| Count | 147 | 134 | - | - | 167 | - | - | ||
| Mean | 6.1 | 5.6 | 81 | 15 | 7.0 | 89 | 25 | ||
| Median | 3 | 3 | 100 | 0 | 4 | 100 | 23 | ||
This research was initiated to assess the success of Google Earth as a tool to help target efforts evaluating ephemeral gully erosion. With a successful targeting tool, time-consuming activities such as obtaining low elevation/high-resolution drone imagery or a “boots on the ground” assessment of a field can be substantially reduced. Instead of obtaining imagery for a whole field or walking the full field perimeter, the assessment can be targeted to the specific areas highlighted by this methodology. An important outcome of this research is that if a person visited or obtained images only in the targeted areas of the field, they will have high confidence that if they find no ephemeral gullies, there are likely none in the field. If they encountered an ephemeral gully, there may be more ephemeral gullies at points in the field not targeted by this strategy. That happened in 38% of the GELS assessments and 29% of the ALL assessments.
This study suggested that Google Earth imagery alone (using the GELS strategy) was successful as a targeting tool in this test. While adding the MSD lines improved the true positive rate and reduced false negatives, it did so while increasing false positives. Our results also suggested that relying only on more recent data may have improved that outcome. Intuitively, more data should improve outcomes if it is representative of current field management. As more sources of imagery with sub-meter GSD become available, it will likely help supplement Google Earth imagery.
Implications
This was a preliminary study that documented the feasibility of an approach in a small geographic area. More work is needed before this approach can be confidently applied to other locations. Our documentation of the frequency of Google Earth imagery (fig. 6) showed substantial gaps in the annual record over the 11-year period we evaluated. For an ephemeral gully to be observed through Google Earth or other imagery, the imagery must be obtained after the feature is formed but before the farmer takes steps to ameliorate the gully. Consequently, it is more likely to observe gullies where there is more frequent imagery when weather and management conspire to have gullies form more frequently and/or farmers do not rapidly fill gullies after they form. Success in central Missouri suggests that the strategy has the potential to succeed in other humid regions of the US.
In this project, we included 2020 data from Google Earth in our analysis. For the 2020 field season, this data was not available in “real time”, though in most cases it was taken before we visited the fields (typically 02 Mar 2020). Additionally, 2020 imagery would not be available for 2018 and 2019 locations. This was true at seven of the 2020 locations, one 2019 location, and two 2018 locations. In three of these locations, the only GED evidence came from the 2020 imagery. However, when including GELS criteria, there was only one location that relied on only 2020 imagery to delineate ephemeral gullies (location 8 in 2020). Including 2020 data had a limited effect on the delineation of GELS lines but artificially increased the number of images available at the time the field was assessed, particularly for GED criteria. Our conclusion that the GELS criteria were the better criteria was not affected by the inclusion of the 2020 imagery.
Other studies have used Google Earth imagery to delineate ephemeral gullies (e.g., Boardman, 2016; Karydas and Panagos, 2020). A contribution of this research was to develop more detailed criteria coupled with specific examples (e.g., figs. 2 and 3) to better calibrate the assessment method. Our goal was to quantify as much as possible and limit errors that are inevitable in a process reliant on human judgment and visual interpretation (Maugnard et al., 2014).
This work is a proof of concept that Google Earth and other publicly available imagery of sufficient quality can be used to target ephemeral gully assessments in row crop fields in an example located in a humid region of the US. Validation work is needed before this approach could be widely adopted with confidence, given the many uncontrollable factors that might affect the efficacy of this approach. The success of this strategy in this project done on farmer’s fields in Missouri suggests that such an effort is worthwhile.
Conclusions
We conclude that Google Earth imagery is useful for targeting the survey of fields for ephemeral gullies if the goal is to prioritize fields requiring additional “boots on the ground” assessment. When all lines identified by the GELS standard (using a 15-m buffer) were assessed, only one field would have been falsely identified as free of ephemeral gullies (4%). The system was successful at identifying all existing ephemeral gullies in 63% of the fields. The fact that this value was less than 100% may require visiting all targeted areas in a field to ensure a high likelihood that no field gets falsely labeled as ephemeral gully free. The mean number of false positives in fields with ephemeral gully erosion was 15%. This suggests there was a significant discriminatory capability for the tool to primarily identify features where there was a reasonable expectation of ephemeral gully erosion. Adding data from a second data source (MSDIS) did not alter these conclusions, though it did increase the false positive rate, perhaps because the first date was the oldest imagery used (2008).
There are significant limitations to expanding these conclusions to other locations. The frequency and intensity of rainfall events that cause ephemeral gully erosion will affect the success of this strategy. The strategy may be less effective in regions with fewer large precipitation events. The availability of imagery may also be an issue. However, the frequency of imagery was not an issue in this area because we identified ephemeral gullies in at least two images in all but three locations (88%). A region with fewer Google Earth images coupled with fewer large rainfall events could make the system less reliable. Additionally, going too far back in time may not be useful as our oldest imagery (over a decade old) produced the most false positives.
The success of this test suggests that this strategy will be successful in a large portion of the humid United States. Further efforts to validate this strategy are needed before it can be confidently adopted.
Acknowledgments
Thank you to the Missouri NRCS for their financial and tactical support for this project. Particular thanks to Glen Davis, Ron Miller, and Pat Turman for their insightful discussion on NRCS practices. The authors also extend thanks to the project staff and graduate and undergraduate workers who contributed to the project, including David Kleinsorge, Krystal Burkett-Tysdal, Theresa Musket, Isaac Shee, Parth Upadhyay, and Timotius Patrick Lagaunne. Particular thanks to Isaac Shee, who did much of the earlier GIS work.
References
Bernard, J., Bingner, R. L., Dabney, S. M., Langendoen, E. J., Lemunyon, J., Merkel, W.,... Wilson, G. V. (2010). Ephemeral gully erosion: A natural resource concern. Laboratory Publication. Report No. 69. US Department of Agriculture, Agricultural Research Service, National Sedimentation Laboratory.
Boardman, J. (2016). The value of Google Earth™ for erosion mapping. CATENA, 143, 123-127. https://doi.org/10.1016/j.catena.2016.03.031
Castillo, C., & Gómez, J. A. (2016). A century of gully erosion research: Urgency, complexity and study approaches. Earth Sci. Rev., 160, 300-319. https://doi.org/10.1016/j.earscirev.2016.07.009
Cruse, R. M., Lee, S., Fenton, T. E., Wang, E., & Laflen, J. (2013). Soil renewal and sustainability. In R. Lal, & B. A. Stewart (Eds.), Principles of sustainable soil management in agroecosystems (pp. 477-500). Boca Raton, LA: CRC Press. https://doi.org/10.1201/b14972-21
Foster, G. R. (1986). Understanding ephemeral gully erosion. In Soil Conservation: An assessment of the national resources inventory (Vol. 2, pp. 90-125). Washington, DC: National Academies Press. https://doi.org/10.17226/648
Frankl, A., Zwertvaegher, A., Poesen, J., & Nyssen, J. (2013). Transferring Google Earth observations to GIS-software: Example from gully erosion study. Int. J. Digital Earth, 6(2), 196-201. https://doi.org/10.1080/17538947.2012.744777
Gilad, U., Denham, R., & Tindall, D. (2012). Gullies, Google Earth and the Great Barrier Reef: A remote sensing methodology for mapping gullies over extensive areas. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XXXIX-B8, 469-473. https://doi.org/10.5194/isprsarchives-XXXIX-B8-469-2012
Google Earth Help. (2021). How images are collected. Retrieved from https://support.google.com/earth/answer/6327779?hl=en&ref_topic=7184180#zippy=%2Csatellite-aerial-images
Grigar, J., Duiker, S. W., & Flanagan, D. C. (2020). Understanding soil erosion by water to improve soil conservation. Crops & Soils, 53(3), 47-55. https://doi.org/10.1002/crso.20030
Karydas, C., & Panagos, P. (2020). Towards an assessment of the ephemeral gully erosion potential in Greece using Google Earth. Water, 12(2), 603. https://doi.org/10.3390/w12020603
Liu, K., Ding, H., Tang, G., Song, C., Liu, Y., Jiang, L.,... Ma, R. (2018). Large-scale mapping of gully-affected areas: An approach integrating Google Earth images and terrain skeleton information. Geomorphology, 314, 13-26. https://doi.org/10.1016/j.geomorph.2018.04.011
Lory, J. A., Upadhyay, P., Lagaunne, T. A. P., Spinka, C., Miller, R., Davis, G., & DeSouza, G. N. (2021). Capability of high-resolution RGB imagery to accurately document residue in row-crop fields. J. Soil Water Conserv., 76(5), 403-413. https://doi.org/10.2489/jswc.2021.00193
Maugnard, A., Cordonnier, H., Degre, A., Demarcin, P., Pineux, N., & Bielders, C. L. (2014). Uncertainty assessment of ephemeral gully identification, characteristics and topographic threshold when using aerial photographs in agricultural settings. Earth Surf. Process. Landforms, 39(10), 1319-1330. https://doi.org/10.1002/esp.3526
Montgomery, D. R. (2007). Soil erosion and agricultural sustainability. Proc. Natl. Acad. Sci., 104(33), 13268-13272. https://doi.org/10.1073/pnas.0611508104
Montgomery, D. R. (2012). Dirt: The erosion of civilizations. Berkeley: University of California Press. https://doi.org/10.1525/9780520952119
Pennock, D. (2019). Soil erosion: The greatest challenge for sustainable soil management. Rome, Italy: United Nations FAO. Retrieved from https://policycommons.net/artifacts/2232089/?soil-erosion/2989501/
Sheshukov, A. Y., Sekaluvu, L., & Hutchinson, S. L. (2018). Accuracy of topographic index models at identifying ephemeral gully trajectories on agricultural fields. Geomorphology, 306, 224-234. https://doi.org/10.1016/j.geomorph.2018.01.026
United States Code (U.S.C.). (1985). The food security act of 1985 (16 U.S.C. 3801-3862).
Upadhyay, P. C., Lory, J. A., DeSouza, G. N., Lagaunne, T. A., & Spinka, C. M. (2022). Classification of crop residue cover in high-resolution RGB images using machine learning. J. ASABE, 65(1), 75-86. https://doi.org/10.13031/ja.14572
USDA-NRCS. (2011). Subpart 501B: Water erosion. In National Agronomy Manual. Natural Resources Conservation Service.
Vanmaercke, M., Panagos, P., Vanwalleghem, T., Hayas, A., Foerster, S., Borrelli, P.,... Poesen, J. (2021). Measuring, modelling and managing gully erosion at large scales: A state of the art. Earth Sci. Rev., 218, 103637. https://doi.org/10.1016/j.earscirev.2021.103637
Wang, C., Cruse, R. M., Gelder, B., James, D., & Liu, X. (2021). Grid order prediction of ephemeral gully head cut position: Regional scale application. CATENA, 200, 105158. https://doi.org/10.1016/j.catena.2021.105158
Weltz, M. A., Huang, C.-H., Newingham, B. A., Tatarko, J., Nouwakpo, S. K., & Tsegaye, T. (2020). A strategic plan for future USDA agricultural research service erosion research and model development. J. Soil Water Conserv., 75(6), 137A-143A. https://doi.org/10.2489/jswc.2020.0805A