Using Sentinel-2 Imagery and Machine Learning Algorithms to Assess the Inundation Status of Nebraska Conservation Easements during 2018–2021

Abstract

Conservation easements (CEs) play an important role in the provision of ecological services. This paper aims to use the open-access Sentinel-2 satellites to advance existing conservation management capacity to a new level of near-real-time monitoring and assessment for the conservation easements in Nebraska. This research uses machine learning and Google Earth Engine to classify inundation status using Sentinel-2 imagery during 2018–2021 for all CE sites in Nebraska, USA. The proposed machine learning approach helps monitor the CE sites at the landscape scale in an efficient and low-cost manner. The results confirmed effective inundation performance in these floodplain or wetland-related CE sites. The CE sites under the Emergency Watershed Protection-Floodplain Easement (EWPP-FPE) had the highest inundated area rate of 18.72%, indicating active hydrological inundation in the floodplain areas. The CE sites under the Wetlands Reserve Program (WRP) reached a mean annual surface water cover rate area of 8.07%, indicating the core wetland areas were inundated periodically or regularly. Other types of CEs serving upland conservation purposes had a lower level of inundation while these uplands conservation provided critical needs in soil erosion control. The mean annual surface water cover rate is 0.96% for the CE sites under the Grassland Reserve Program (GRP). The conservation of the CEs on uplands is an important component to reduce soil erosion and improve downstream wetland hydrological inundation performance. The findings support that the sites with higher inundation frequencies can be considered for future wetland-related conservation practices. The four typical wetland-based CE sites suggested that conservation performance can be improved by implementing hydrological restoration and soil erosion reduction at the watershed scale. The findings provided robust evidence to discover the surface water inundation information on conservation assessment to achieve the long-term goals of conservation easements.

1. Introduction

Conservation easements (CEs) are private lands purchased to limit the use of the land in order to protect its conservation values, and such land plays an important role in the provision of ecological services. Private lands provide critical ecological services for sediment control, water quality improvement, open spaces, wildlife habitats, groundwater recharge, and cultural ecosystem services [1]. Conservation of private lands is extremely important because private lands serve as habitats for over 95% of federally listed species in the United States [2,3]. The multiple types of CE programs provide private landowners with a market-based tool that represents a voluntary but legally binding agreement to protect the environmental amenities and ecosystem services of their lands [4,5,6]. CEs impose permanent or long-term private land-use restrictions associated with property deeds [3,7]. Moreover, landowners who purchase private land with a CE can still use, sell, and bequeath the land, subject to easement restrictions.

Established by the Swampbuster provision in the Food Security Act of 1985 and administrated by the U.S. Department of Agriculture-Natural Resources Conservation Service (USDA-NRCS), the Wetland Reserve Program (WRP) began in 1990 to allow agricultural producers to restore or set aside wetlands for 30-year or permanent easements [8]. Conservation programs in the U.S. were further integrated into the Agricultural Act of 2014. The Agricultural Conservation Easement Program (ACEP) merged three programs: the Farm and Ranch Lands Protection Program (FRPP), the Grassland Reserve Program (GRP), and the WRP. The Regional Conservation Partnership Program (RCPP) combined the Agricultural Water Enhancement Program (AWEP) with other programs. The Emergency Watershed Protection Program (EWPP) was established in 1978 and amended in 1996 to add those lands damaged by flooding as floodplain easements. These programs provide critical support for environmental quality, such as water quality, wildlife and biodiversity, soil erosion control, groundwater recharging, and carbon reduction [8,9,10].

The effectiveness of conserved lands has been investigated in recent studies [11,12]. Conservation lands are often remote areas across the landscape, and on-site monitoring of such land is expensive [13]. Satellite-based Earth observation is a suitable method of monitoring conservation lands on a large scale across a specific state. Eichenwald et al. [1] used Google Earth Engine and 31 years of Landsat images to monitor the habitats of 24 vertebrates and compared the effectiveness of conserved lands. Their results showed that conservation on private lands is critical for long-term habitat protection. More research has been accomplished to detect the world’s surface water changes [14]. Pekel et al. [15] used Landsat satellite data to map surface water globally and its changes from 1984 to 2015 by deriving the normalized difference vegetation index (NDVI) and hue-saturation value (HSV). Donchyts et al. [16] analyzed the surface water changes by accessing data from the Aqua Monitor, a GEE and Landsat-based tool. Many automated algorithms or methods, such as support vector machines (SVM), decision trees, and random forests, have been widely used for land-cover classification [17,18,19].

The Sentinel-2 satellite was launched by the European Space Agency (ESA) in 2015 and includes a constellation of two polar-orbiting satellites to monitor land surface conditions over a 5-day period using high-resolution multispectral sensors. Sentinel-2 satellites have 13 spectral bands in the electromagnetic spectrum between 0.665 μm and 2.190 μm. High-quality multispectral images from the Sentinel-2 satellite can reach a maximum of 10 m spatial resolution. The Sentinel-2 Multi-Spectral Instrument (MSI) offers the necessary spatial and temporal resolution for wetland monitoring and assessment [20].

Owing to the high spatial heterogeneity and temporal dynamics of wetlands, the development of effective remote-sensing algorithms and indices to detect water under wetland vegetation coverage and wet soils has been challenging [21,22]. Because the water in wetlands changes daily and seasonally, the extent and spectral signature of wetlands can be highly dynamic. Many wetlands have shallow water areas that may also be mixed with vegetation or soil, and spectral reflectance may not be easily detected. By definition, mono-temporal classification approaches cannot fully describe temporal dynamics [20]. Near-infrared (NIR) and shortwave infrared (SWIR) spectroscopy can effectively reflect water and wet surface characteristics, and water indices have been used as a practical approach for mapping wetland inundation conditions from the spectral bands of satellite images [23,24,25,26,27,28]. Commonly applied indices include the normalized difference water index (NDWI), modified normalized difference water index (MNDWI), normalized difference moisture index (NDMI), and NDVI.

Machine learning algorithms have recently received attention in wetland imaging classification [29,30,31]. Kordelas et al. [32] developed an unsupervised approach for estimating the extent of inundation from radiometrically corrected Sentinel-2 data. Lefebvre et al. [33] used water in wetlands (WIW) to track spatiotemporal changes in inundation patterns of wetlands with variable heights and densities over time from Sentinel/Landsat images. Pena-Regueiro et al. [22] analyzed Sentinel-2 images using seven indices to detect small water bodies in wetlands with a high diversity of temporal and spatial flooding patterns, and a comparison of the indices showed that the NDWI had the highest performance for extracting water surfaces. Huang et al. [34] developed an automatic classification tree approach for classifying surface water extent from Sentinel-1 synthetic aperture radar (SAR) data in a Prairie Pothole Region site and achieved an overall accuracy of 79–93%. SVM models were also applied in this study as surface water classification models among the commonly used machine learning algorithms. SVM represents one of the most commonly used machine learning algorithms in remote sensing, land-cover classification, and mapping fields, and the SVM algorithm is typically used in remote sensing to address both classification and regression problems. SVM has also been extensively applied to land-use and land-cover (LULC) classification problems. Shao and Lunetta [35] collected MODIS time-series data and used an SVM to conduct LULC classification in North Carolina, and the results showed that SVM has a superior generalization capability, even with small training sample sizes. Machine learning algorithms have shown a significant advantage in LULC classification in recent years, thus making them an important part of the remote-sensing field, especially LULC classification [36,37]. A study by Thanh Noi and Kappas [38] also showed that SVM classifiers have the highest overall accuracy and lowest sensitivity to training sample size compared with RF classifiers. The difference in training size also affects the performance of RF and SVM classifiers. Ma et al. [27] compared SVM with RF classifiers and showed that SVM works better with small training set sizes. Sheykhmousa et al. [39] compared RF and SVM models based on an evaluation of 251 journal papers and found that the overall accuracy (OA) of SVM is generally higher and lower than that of RF when applied at spatial resolutions higher than 10 m and lower than 100 m, respectively. In this case, the spatial resolution of Sentinel-2 was 10 m, which fits most of the former cases. Zhang et al. [40] compared six main machine learning algorithms and confirmed that SVM is an optimal algorithm when using Google Earth Engine (GEE) to classify wetland land cover in Nebraska. Overall, an SVM with a linear kernel classifier presented advantages in surface water classification in wetlands within Nebraska. As a result, this study applies a linear kernel SVM classifier as a surface water classifier. Two types of classification approaches are generally applied in remote-sensing classification: pixel-based classification and objective-based classification [36,37,41]. The classification model used in this study is a pixel-based classification model, and it has been successfully applied in water body and wetland classification [36,42].

GEE is an open-source platform to access, process, and analyze various Earth observation data, including Sentinel-2 imagery. Several recent studies have used GEE to map and assess wetland conditions [43,44]. GEE provides a large amount of remote-sensing data and data processing approaches, including many built-in machine learning algorithms [45,46]. GEE can also provide different machine learning classifiers for pixel-based classification in this study. Satellite images from Sentinel-2 and machine learning models were imported into GEE and used to further analyze the classification data. GEE is an efficient tool for mapping all CE sites across Nebraska and determining the surface water changes.

To date, studies have not monitored all CEs within an area using remote-sensing methods. Therefore, this study has two research objectives: (1) to develop an efficient, sustainable, and low-cost approach to monitoring surface water inundation status in all CEs across the state of Nebraska; (2) to detect, map, and analyze the surface water changes in all CEs by applying machine learning models using the GEE platform and other GIS tools.

2. Materials and Methods

2.1. Study Area

Nebraska, a state in the U.S., is located on the Dissected Till Plains in the eastern part, and the Great Plains in the north, central, and northwest. The climate of the state is also a humid continental climate (Dfa) in the east and a humid subtropical climate (Cfa) in the west. About 90% of the land in Nebraska is agricultural land. According to GIS data from Nebraska’s CE database for 2021, Nebraska has a total of 663 CEs, including the WRP, GRP, FRPP, ACEP-Wetlands Reserve Easements (ACEP-WRE), and EWPP-Floodplain Easement (EWPP-FPE). Quantitative monitoring and assessment of the inundation dynamics, wildlife usage, and vegetation conditions can provide insightful information on the performance of CEs. Among the 663 CEs, 1 of them is physically outside the state boundary. Because CEs can cross state boundaries, CE sites that overlapped Nebraska’s state boundary were excluded to ensure the accuracy of the assessment. Thus, only CEs within the boundary of Nebraska were calculated in this study. Finally, this study mapped and analyzed 662 CE sites, and their locations are illustrated in Figure 1. The analysis of Nebraska’s four wetland complexes is contained in a case study. The wetland complexes in Nebraska are grouped into four types: playas, sandhills, saline/alkaline, and riverine [47,48].

2.2. Data Sources

This project used GEE to collect and preprocess remote-sensing data from the research area. Multispectral images from Sentinel-2 were collected for training and classification purposes. Sentinel-2 is a two-satellite system with a temporal resolution of 5 days. The high temporal resolution of Sentinel-2 allows the images of the research area to be collected weekly. The datasets were processed in the GEE platform and directly imported into the GEE platform using the built-in eeImageCollection(“COPERNICUS/S2_SR”) command. The Sentinel-2B satellite was launched in March 2017, and the Sentinel-2 system has been fully functional since 2018. Therefore, this study used data from 2018 to 2021. Considering the climate of Nebraska, this study only used data from March to November each year to avoid snow and ice cover conditions.

The Sentinel-2 bands used in this study are B2 (490 nm), B3 (560 nm), B4 (665 nm), B5 (705 nm), B6 (740 nm), B7 (783 nm), B8 (842 nm), B8a (865 nm), B11 (1610 nm), and B12 (2190 nm). Among these bands, the RGB (B2, B3, B4) and near-infrared (NIR: B8) bands have spatial resolutions of 10 m. The other six bands, including the vegetation red edge (B5, B6, B7), narrow NIR (B8a), and shortwave infrared (SWIR: B11, B12) bands, have spatial resolutions of 20 m. Remote-sensing data were collected monthly and filtered by the cloudy pixel percentage provided by Sentinel-2. All bands from Sentinel-2 used for classification in this study are listed in Table 1.

In this study, index bands were calculated and added for training to increase the training accuracy. NDWI (NDWI = (B3 − B8)/(B3 + B8)), NDVI (NDVI = (B8 − B4)/(B8 + B4)), MDNWI (MNDWI = (B3 − B11)/(B3 + B11)), and NDMI (NDMI = (B8 − B11)/(B8 + B11)) were added as index bands for training. The index data were calculated using the Image.normalizedDifference code in GEE and Sentinel-2 spectral data were directly imported into the GEE code editor.

Soil data from the 2021 Soil Survey Geographic Database (SSURGO) for soil condition analysis were downloaded from the USDA Geospatial Data Gateway (https://datagateway.nrcs.usda.gov/, accessed on 1 February 2022). For all CEs in Nebraska, the five most frequent soil types are Valentine-Els complex, moist, 0 to 9 percent slopes (11.44 km²); Els-Ipage complex, 0 to 3 percent slopes (9.10 km²); Luton silty clay, rarely flooded (5.90 km²); Fluvaquents, frequently flooded, wet (5.67 km²); Albaton silty clay, 0 to 2 percent slopes, occasionally flooded (5.35 km²). The shapefiles of the CE sites were obtained from the Nebraska Office of the USDA-NRCS. These shapefiles were preprocessed with QGIS and then uploaded to Google Drive, where GEE data can be directly imported and processed.

2.3. Data Analysis

The workflow of this study is illustrated in Figure 2. The data mentioned in the previous paragraphs are available on the GEE platform and can be directly collected using an image collection code. The data also include shapefile data, which contain the boundaries of the research area.

The classification model used in this study was a pixel-based model. The seven steps performed for the geospatial analysis of the 662 CE csites are listed below.

• (1). Selecting geospatial data: For surface water classification, the Sentinel-2 images are imported into GEE and filtered by date (March to November each year from 2018 to 2021). The QA-60 band was used to sort and select the least cloud-covered images. The shapefile of the CE sites is uploaded to GEE as the boundary.

• (2). Labeling data: The geometric tool in GEE is used to label different land-cover classes. The Rainwater Basin Annual Habitat Surveys data in 2020 and field surveys in 2021 were used as supportive information to label and verification which confirmed the designations of water and land classes in the training area. Images of RGB, NIR, NDWI, and NDSI created in GEE are used as references to make labels. All 662 CEs in the Nebraska area are separated into five zones (east, mid-east, central, mid-west, and west) to minimize bias and the effect of climate. CEs from each zone each month are used to make labels. About twenty thousand pixels are generated using the randomPoints function. All features input in the classifier are randomly sampled using a random column function. After collecting the sample by the random column function, 70% of the sample is used for training and 30% for testing.

• (3). Calculating indices: The built-in normalized difference function of GEE is used to calculate the NDVI, NDWI, NDMI, and MNDWI. Then, those index bands are added together with spectral bands as training bands.

• (4). Selecting machine learning classifiers for wetland classification: According to former research, linear kernel SVM is selected as the classifier for all the classifications. All eleven spectral bands and four index bands are applied as training bands.

• (5). Calculating surface water cover: The classification results are exported from GEE as TIFF files. QGIS is used to batch process and transfer all TIFF files into shapefiles. Finally, the surface water cover condition in every pixel within all 662 CE sites from 2018 to 2021 is determined.

• (6). Calculating the surface water inundation frequency: All water cover data are calculated in the last step. Water cover data for all CE sites are used to calculate the water cover rate. At least 25 images are used for the water cover rate calculation for each site.

• (7). Calculating and mapping inundation conditions: Water cover data in every pixel calculated in the last two steps are applied in this step to calculate the pixel inundation frequency in all 662 CE sites. The inundation frequency map is constructed using QGIS and ArcGIS.

2.4. Classification Accuracy Assessment

The accuracy assessment used a testing sample, which represented 30% of the input sample. In this case, the overall accuracy (OA), recall, precision, and F1 score were calculated to compare different classifiers with different parameters.

(1)$\mathrm{Overall}\mathrm{Accuracy}=\frac{\mathrm{TP}+\mathrm{TN}}{\mathrm{TP}+\mathrm{TN}+\mathrm{FP}+\mathrm{FN}}$

(2)$\mathrm{Recall}=\frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FN}}$

(3)$\mathrm{Precision}=\frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FP}}$

(4)$\mathrm{F}1\mathrm{score}=2\times \frac{\mathrm{Recision}\times \mathrm{Recall}}{\mathrm{Precision}+\mathrm{Recall}}$

Here, a true positive (TP) indicates that the detected condition is present, a true negative (TN) indicates that an undetected condition is absent, a false positive (FP) indicates that the detected condition is absent, and a false negative (FN) indicates that an undetected condition is present. For example, in this case, a TP for water means that pixels classified as water are indeed water in the real world. The F1 score is calculated using precision and recall, with values ranging from 0 to 100. These indicators reflect the accuracy of the model. Such an accuracy test can provide a deeper understanding of how to use machine learning methods to effectively classify multispectral land-cover image data in Nebraska’s natural conservation land settings.

3. Results

This study uses OA, recall, precision, and F1 score to assess classification accuracy. The accuracy assessment is calculated after each classification is performed. The mean OA for all classifications is 99.79%, ranging from 95.45% to 100%. The mean recall is 99.57%, ranging from 87.53% to 100%. The mean precision is 99.21%, ranging from 90.60% to 100%. The mean F1 score is 99.36%, ranging from 93.35% to 100%. The result of the accuracy assessment shows that the SVM with a linear kernel classifier is accurate and reliable in surface water classification in Nebraska during the growing season (March to November). Ground truth is also used to verify the classification results. We compared the Sentinel-2 classification results with the Rainwater Basin’s annual habitat survey data in 2020. There were 72 CEs with visible surface water from the Sentinel-2 classification results that were 100% verified in the annual habitat survey data. The total area of surface water in CEs within the Rainwater Basin is 2.80 km² counted by the annual habitat surveys, while this number is 2.46 km² our classification result, which covered 87.68% of the total surface water area.

The annual temporal patterns of hydrological performance for all CE sites are illustrated in Figure 3. The blue line shows the four-year average water cover rate for all 662 CE sites. Large variations were found among the months and the years. The highest inundated areas were found in April 2019. Approximately 13% of the total areas were inundated in April, and 10% in May, June, and July. A lower level of inundation was observed in August (with 5% inundation area), September (with 7% inundation area), October (with 6% inundation area), and November (with 7% inundation area).

The results illustrated in Figure 4 show a more prominent display of the surface water conditions at each site during the four-year study period. Among the 662 sites, nine sites did not have surface water during a four-year period, and they accounted for 1.36% of all sites. Of the 662 sites, 178 sites had surface water cover during the four-year period, and they accounted for 26.89% of all sites. The median of all sites for all four years is 3.14%. The year 2019 shows a relatively wetter condition with a median surface water cover rate of 4.21%, ranging from 0 to 99.9%. The median for the other three years (2018: 2.78%, 2020: 3.05%, and 2021: 2.73%) is less than the median of all sites for all four years.

The surface water cover rate by area in each category of CE sites is illustrated in Figure 5. Table 2 lists the percentage of inundated areas in each category of CE sites during 2018–2021. The results show that 7.85% of the CE land areas were under inundated conditions. The water cover condition results showed that significant inundation differences occurred over the years (mean surface water cover rates for 2018–2021 are 6.82%, 12.82%, 6.17%, and 5.61%, respectively). The year 2019 turned out to be a wetter year among the four years, with a surface water cover rate of 12.82%, which is much more than the mean value (7.85%). This study showed that 64.64% of the total area of all CEs in this study was not inundated, and 0.62% of the total area presented an over 90% (100% not included) chance of water cover from 2018 to 2021. Ultimately, 0.54% of the total area of all CE sites experienced 100% inundation during the past four years.

The CE sites under the EWPP-FPE had the highest inundated area rate of 18.72%, indicating active hydrological inundation in the floodplain areas. The CE sites under the WRP reached the mean annual surface water cover rate by area at 8.07%, indicating the core wetland areas were inundated periodically or regularly. Other types of CEs serving for upland conservation purposes had a lower level of inundation but provided critical conservation needs in soil erosion control. The GRP category is the one that shows significantly less water among the five categories. The surface water cover rate of GRP shows that GRP is the lowest CE type in each of the four years.

3.1. Soil Conditions with Inundation Frequency

Through analyzing the inundation maps with the SSURGO data, the results indicate that the most common soil types on uplands include Valentine-Els complex, moist (0 to 9 percent slopes); Els-Ipage complex (0 to 3 percent slopes); Luton silty clay (rarely flooded); Valentine fine sand (rolling, 9 to 24 percent slopes, moist). During the past four years, no inundation conditions were observed on these soils.

For the high frequency of inundation areas, the most common soils are Fluvaquents (frequently flooded, wet), Marlake loamy fine sand (frequently ponded), Massie silty clay loam (frequently ponded), Inavale fine sand (channeled, frequently flooded, wet), Sarpy-Grable variant complex (occasionally flooded), Barney loam (frequently flooded), and Barney variant fine sand (frequently flooded).

3.2. Case Study

Here we analyzed four CE sites in Nebraska as case studies. The locations of the four study sites are shown in Figure 6. The four sites were selected based on four of Nebraska’s wetland complexes. The first site was a sandhill site in northwest Nebraska (site ID: 6665260800MW5) located in Cherry County, with a total area of 0.61 km², which is the only site located in the sandhill region. The second site was a playa site consisting of three parcels (site ID: 5465261701MZG, 6665260400B9V, and 6665260100B5Y) located in Hamilton County within the Rainwater Basin, with a total area of 0.66 km². The third site was a riverine site (site ID: 6665260100B6R) located within Merrick County along the Platte River, with a total area of 3.80 km². The Rainwater Basin is a typical area for the playa and riverine wetland complexes. The last site was a saline site consisting of three parcels (site ID: 6665260200B6Q, 6665260300B74, and 6665261000Y8H) located in Lancaster County in the Salt Creek watershed, with a total of 1.44 km². The Salt Creek watershed is also a typical area for Nebraska’s Eastern saline wetland complex. Comparisons of the inundation frequency and land-cover rate change among different years are presented in Figure 7, Figure 8, Figure 9 and Figure 10.

The sandhill site is located in northwest Nebraska within the sandhill land type. Those sandhill wetlands are formed in depressions in the sandhills areas where the groundwater intercepts the land’s surface. Figure 7 illustrates the inundation frequency and water cover rate changes over different years. The highest inundation frequency was 96%, and 5.25% of the total area of this site presented an inundation frequency greater than 90% while 74.10% of this site was never inundated during the study period. The maximum water cover rate was 18.12% in June 2020, while the minimum water cover rate was 0.05% in June 2018. The total area of this sandhill wetland site was 0.61 km². Figure 7 shows that this site had a very high inundation frequency in the central wetland area. This site also presented surface water in all four years.

The playa site located in the basin is illustrated in Figure 8. Playa wetlands are wind-formed wetlands with nearly circular depressions in semiarid areas, and they are distributed across three-quarters of Nebraska. Data were not available for May for all four years because of cloud cover. The highest inundation frequency was 48.00%, and it accounted for approximately 302 m² of the area. An inundation frequency of 15.85% was observed for 20 to 50% of the total playa area, while inundation was not observed in the four years for 23.03% of the playa site. The highest water cover (61.95%) occurred in June 2019, while the lowest water cover (0%) was observed over six months during the past four years. The mean water cover rate was 9.50% according to the data for 25 months. Figure 8 shows that this site was highly inundated during the wet season. The results for this site showed that substantial differences occurred over the years. The highest mean surface water cover rate was 21.91% in 2019, which was much higher than that of the other three years, while the lowest was 1.23% in 2018.

The riverine site was also located in the Rainwater Basin, the location, and inundation frequency are shown in Figure 9. Data were not observed in May for all four years because of the climatic conditions and cloud coverage. The highest inundation frequency (100%) covered approximately 0.64% of the land at this site, while no inundation was observed over 85.69% of the site during the four years. The mean water cover rate was 2.35% according to the data for 25 months. The highest water cover rate was recorded in July 2020 (12.54%), while the lowest water cover rate was 0.80% in April 2020. The riverine site is located near Aurora, Nebraska. Changes in water coverage were very smooth at this site. Riverine wetlands are closely associated with floodplains. The inundation frequency map shows that water ponds occur year-round. However, surface water only appeared in a very limited area at this site. The mean surface water cover rate was very stable for the four years. The monthly surface water cover rate was mostly between 1 and 3%, while the yearly surface water cover rate showed little difference from the highest of 3.03% (2020) to the lowest of 1.68% (2019).

The inundation frequency of the saline site located in northern Lancaster County is illustrated in Figure 10. Saline wetlands are a unique natural resource distributed worldwide. The study site was located within the eastern Nebraska saline wetland area in the Salt Creek watershed. The highest inundation frequency at this site was 76.00%, which covered an area of 226 m², while inundation did not occur in the four years over 92.73% of the site. The highest water cover rate occurred in August 2021 (3.69%), while the lowest of 0% occurred in four months during the four years (May 2018, August 2019, September 2019, and April 2020). The mean water cover rate at the saline site was 0.84%, and the inundation frequency map shows that surface water was limited to a small area at this site. The highest surface water cover rate (1.60%) occurred in 2021.

4. Discussion

This study proposes a methodology that uses Sentinel-2 imagery, GEE, and machine learning models to detect the surface water inundation status at 662 CE sites in Nebraska. This study explored an efficient and low-cost approach for long-term monitoring of large-scale land areas using high-resolution satellite imagery and showed that the approach is reliable and can be used for long-term and continuous studies. Three major lessons were drawn regarding the use of machine learning and GEE for CE site assessments, monitoring, and wetland conservation research.

First, this study provides a long-term and cost-effective wetland mapping tool that can be used in hydrological performance monitoring for wetland conservation programs [49]. The findings by Pekel et al. [15] and Donchyts et al. [16] are based on the Landsat satellite dataset, with up to a 30 m spatial resolution, which is not high enough for CE monitoring. Sentinel-2 offers imagery with a spatial resolution of up to 10 m, which can significantly improve the monitoring range, especially for small sites that cannot be covered by the Landsat satellite. The mapping approach used in this study could be a valuable supplement to conservation monitoring and assessment programs. This study’s classification results support previous studies on model selection on LULC classification, especially on wetlands [27,35,38,39,40]. Particularly, this technology can help identify and prioritize wetland CEs that may need additional restoration.

Second, the findings of this study indicate that the inundation conditions of wetland CE sites vary and they are subject to a series of combined factors. The affecting factors for hydrological performance include topographical factors (e.g., upland or lower lands, slopes), human activities, especially agricultural activities (e.g., irrigation and agricultural pits), conservation practices (e.g., pumping and sediment removal), and watershed context (e.g., soil types, vegetation cover, drainage pattern). As a large number of CE sites in this study contain wetlands, the hydrological characteristics of these sites become much more critical. Changes in hydrological characteristics significantly affect the water depth and hydrophyte community and may cause changes in the edges of wetlands [50]. The inundation maps developed from this study provide solid evidence to understand the surface water changes of the wetland-related CEs. We also compared the surface inundation status with local climate data, and the results show that the surface water cover rate of case study sites was not significantly related to temperature and precipitation.

Third, Sentinel-2 data limitations should be recognized, and associated improvements should be made in future studies. The limitations of this approach can be summarized in terms of climate and resolution. Climate conditions have a significant impact on satellite data accessibility. This study focused on 662 CE sites from 2018 to 2021 and divided all 662 CE sites into five zones to minimize the effect of climate. However, cloud-free images still could not be acquired for each site each month. The worst conditions were observed in May, which is typically the spring rainy season in Nebraska. This study was initially designed to obtain weekly satellite imagery for a much more accurate analysis of all CE sites across the state. For example, riverine and playa sites lack data from May every year. Other sites had missing data from May for at least one year. Considering the rainy season in Nebraska, missing data in May, maybe an unavoidable problem. This limitation can be resolved by long-term data collection or manually collecting data from piloted aerial or field surveys. Sentinel-1 SAR is another option for land observation with cloud cover in our future study [34,51]. The newly published Dynamic World dataset is a powerful tool for handling global LULC studies, which can partially resolve this limitation [52]. The second limitation is spatial resolution. For example, imagery with a resolution of up to 10 m is insufficient for some small streams with widths smaller than 10 m. The 10 m resolution may also cause a false negative in a small inundation area with algae cover, and it is not sufficiently high for vegetation and soil classification. The spatial resolution was 10 m for most of the Sentinel-2 bands. Although this spatial resolution is quite high compared with that of the Landsat series, it is not sufficient for vegetation classification. This study can be improved by continuing data collection and using more high spatial resolution data collection methods, such as drone-based imagery.

Policy Recommendations for CEs

The findings of this study show that adaptive management is needed for CE sites. The results showed that water boundaries and inundation conditions are continually changing. Owing to the high temporal resolution of Sentinel-2, the inundation conditions of all CE sites across the state can be calculated and mapped more frequently. As mentioned above, many wetlands in remote areas are not easy to access when conducting traditional systematic research. Because these wetlands may be subject to degradation, the proposed method can be used to rapidly determine the impact of nearby agricultural properties or road construction that can significantly influence a wetland’s hydrological conditions. The lands with a high level of inundation frequency can be considered for future conservation programs. Based on the findings of this study, the following recommendations are provided for CE management.

The first recommendation is to perform hydrological restoration at the watershed scale for wetland-related CEs. Full hydrological restoration or partial hydrological recovery is a key step for protecting wetlands at the watershed scale. This study found a high inundation percentage among wetland sites, showing regular or partially ponded water over the past four years. The findings also showed that limited inundation areas were observed at the CE sites (35.36% of the total area with inundation vs. 64.64% of the total area without inundation during the past four years). Thus, the ponding size, frequency, and duration can be improved through hydrological restoration at the watershed scale. Tang et al. [53] recommended using restoration treatments to increase the function of wetlands. Sediment removal, drain closure, irrigation reuse pit closure, and many other treatments can also be applied to CE sites to maintain and restore wetland functions across the state. These treatments can promote sustainable conservation at the CE sites studied in this research.

The second recommendation is to keep the protection for the uplands to reduce soil erosion at the watershed level. Many of the CEs contain upland, non-wetland soils. Uplands are an important part of downstream wetland protection. Protecting uplands associated with wetlands will help better manage the erosion, sedimentation, and nutrient loading in the associated wetland. Uplands are an important part of wetland CE protection. Wetlands in Nebraska suffer from degradation due to sedimentation generated by cropping practices [53]. The current upland conservation among the CEs is a critical component to be continued to serve the needs for soil erosion reduction.

5. Conclusions

This study proposes a long-term monitoring method for inundation conditions at 662 CE sites across the entire state of Nebraska. The results showed that GEE is a powerful and reliable tool for monitoring surface water changes in large-scale areas. This study explored the possibility of using GEE to monitor the long-term hydrology and inundation conditions of CEs, especially conservation wetlands. The advantage of using satellite-based data, including Sentinel-2 data, to monitor CEs includes (but is not limited to) high temporal and spectral resolutions. GEE makes it possible to handle data with both high temporal and spectral resolutions on a large scale.

Long-term monitoring of the surface water of CE sites in Nebraska can provide land information for the further detection of associated ecosystem services. The findings of this study provide the contemporary status of CEs, especially conservation wetlands, and such data can help identify potential opportunities for future conservation practices. This study is valuable as a long-term monitoring tool for 30-year or permanent easements. The findings of this study will also contribute to CE management for wetland restoration and upland protection at the watershed level. Moreover, the hydrology function of downstream wetlands and the soil erosion reduction of uplands are the two essential management aspects that future conservation programs should focus on.

Author Contributions

Conceptualization, L.Z. and Z.T.; methodology, L.Z.; software, L.Z.; validation, L.Z. and Q.H.; formal analysis, L.Z.; investigation, L.Z.; resources, L.Z.; data curation, L.Z.; writing—original draft preparation, L.Z.; writing—review and editing, Z.T. and Q.H.; visualization, L.Z.; supervision, Z.T.; project administration, Z.T.; funding acquisition, Z.T. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

A GEE example for water classification on conservation lands in Nebraska in 2020/3 is available at https://code.earthengine.google.com/df95fccd00fd4d43f64f5fa77fd1268f?noload=true (accessed on 26 July 2022). The data used to support the findings of this study are available from the corresponding author upon request.

Acknowledgments

The research is also supported by the Robert B. Daugherty Water for Food Global Institute and the University of Nebraska Collaboration Initiative Grant. The contents do not necessarily reflect the views and policies of the funding agencies, and the mention of trade names or commercial products does not constitute endorsement or recommendation for use. The research team sincerely appreciated the great support from Shawn McVey (USDA-NRCS), Ted LaGrange and Randy Stutheit (Nebraska Game and Parks Commission), Tom Malmstrom (City of Lincoln), Dan Schulz, Andy Bishop (Rainwater Basin Joint Venture), Jeff Drahota (U.S. Fish & Wildlife Service), and Dick Ehrman (Lower Platte South Natural Resource District).

Conflicts of Interest

The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of the data; in the writing of the manuscript, or in the decision to publish the results.

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Figures and Tables

Figure 1. Location map of CE sites in Nebraska.

Figure 2. Research workflow to calculate inundation areas in each CE site.

Figure 3. The mean water cover rate by area for all CEs during 2018–2021.

Figure 4. Surface water cover rate for each site in the four-year study period.

Figure 5. Surface water cover rate by area for each category of CE sites.

Figure 6. Location of the four case study sites. The sandhill site (red box) is located in Cherry County in northern Nebraska; the playa site (yellow box) is located in Hamilton County in the Rainwater Basin; the riverine site (blue box) is located in Merrick County in the Rainwater Basin; the saline site (pink box) is located in Lancaster County in eastern Nebraska.

Figure 7. Inundation frequency and surface water cover rate in the sandhill site.

Figure 8. Inundation frequency and surface water cover rate in the playa site.

Figure 9. Inundation frequency and surface water cover rate in the riverine site.

Figure 10. Inundation frequency and surface water cover rate in the saline site.

Table 1

The geospatial data sources from the Sentinel-2 satellites.

Data Sources	Bands	Resolution	Date
Sentinel-2	B2, B3, B4, B8	10 m	March 2018–November 2018March 2019–November 2019March 2020–November 2020March 2021–November 2021
Sentinel-2	B5, B6, B7, B8, B8a, B11, B12	20 m	March 2018–November 2018March 2019–November 2019March 2020–November 2020March 2021–November 2021

Table 2

The percentage of inundated areas in each category of CE sites during 2018–2021.

	Site Count	2018	2019	2020	2021	Category Mean
ACEP-WRE	22	2.88%	5.36%	3.68%	4.56%	4.12%
EWPP	11	19.75%	44.60%	8.33%	2.21%	18.72%
GRP	13	0.91%	0.60%	1.00%	1.35%	0.96%
WRP	600	6.99%	12.99%	6.44%	5.87%	8.07%
Other	16	1.81%	4.76%	1.89%	2.97%	2.86%
Year Mean		6.82%	12.82%	6.17%	5.61%	7.85%

References

1. Eichenwald, A.J.; Evans, M.J.; Malcom, J.W. US Imperiled Species Are Most Vulnerable to Habitat Loss on Private Lands. Front. Ecol. Environ.; 2020; 18, pp. 439-446. [DOI: https://dx.doi.org/10.1002/fee.2177]

2. Hilty, J.; Merenlender, A.M. Studying Biodiversity on Private Lands. Conserv. Biol.; 2003; 17, pp. 132-137. [DOI: https://dx.doi.org/10.1046/j.1523-1739.2003.01361.x]

3. Kareiva, P.; Bailey, M.; Brown, D.; Dinkins, B.; Sauls, L.; Todia, G. Documenting the Conservation Value of Easements. Conserv. Sci. Pract.; 2021; 3, e451. [DOI: https://dx.doi.org/10.1111/csp2.451]

4. Bastian, C.T.; Keske, C.M.H.; McLeod, D.M.; Hoag, D.L. Landowner and Land Trust Agent Preferences for Conservation Easements: Implications for Sustainable Land Uses and Landscapes. Landsc. Urban Plan.; 2017; 157, pp. 1-13. [DOI: https://dx.doi.org/10.1016/j.landurbplan.2016.05.030]

5. Keske, C.M.H.; Arnold, P.; Cross, J.E.; Bastian, C.T. Does Conservation Ethic Include Intergenerational Bequest? A Random Utility Model Analysis of Conservation Easements and Agricultural Landowners. Rural. Sociol.; 2021; 86, pp. 703-727. [DOI: https://dx.doi.org/10.1111/ruso.12370]

6. Thompson, A.W.; Wadleigh, R. Factors Motivating Forest Conservation Easement Adoption in Wisconsin’s Northwoods. Soc. Nat. Resour.; 2022; 35, pp. 129-148. [DOI: https://dx.doi.org/10.1080/08941920.2021.2005854]

7. Kemink, K.M.; Adams, V.M.; Pressey, R.L.; Walker, J.A. A Synthesis of Knowledge about Motives for Participation in Perpetual Conservation Easements. Conserv. Sci. Pract.; 2021; 3, e323. [DOI: https://dx.doi.org/10.1111/csp2.323]

8. Lewis, K.E.; Rota, C.T.; Anderson, J.T. A Comparison of Wetland Characteristics between Agricultural Conservation Easement Program and Public Lands Wetlands in West Virginia, USA. Ecol. Evol.; 2020; 10, pp. 3017-3031. [DOI: https://dx.doi.org/10.1002/ece3.6118]

9. Farmer, J.R.; Meretsky, V.; Knapp, D.; Chancellor, C.; Fischer, B.C. Why Agree to a Conservation Easement? Understanding the Decision of Conservation Easement Granting. Landsc. Urban Plan.; 2015; 138, pp. 11-19. [DOI: https://dx.doi.org/10.1016/j.landurbplan.2015.01.005]

10. Lewis, K.E.; Rota, C.T.; Lituma, C.M.; Anderson, J.T. Influence of the Agricultural Conservation Easement Program Wetland Practices on Winter Occupancy of Passerellidae Sparrows and Avian Species Richness. PLoS ONE; 2019; 14, e0210878. [DOI: https://dx.doi.org/10.1371/journal.pone.0210878]

11. Braza, M. Effectiveness of conservation easements in agricultural regions. Conserv. Biol.; 2017; 31, pp. 848-859. [DOI: https://dx.doi.org/10.1111/cobi.12909]

12. Zhang, W.; Mei, B.; Izlar, R.L. Impact of Forest-Related Conservation Easements on Contiguous and Surrounding Property Values. For. Policy Econ.; 2018; 93, pp. 30-35. [DOI: https://dx.doi.org/10.1016/j.forpol.2018.05.008]

13. Gallant, A.L. The Challenges of Remote Monitoring of Wetlands. Remote Sens.; 2015; 7, pp. 10938-10950. [DOI: https://dx.doi.org/10.3390/rs70810938]

14. Yamazaki, D.; Trigg, M.A. The Dynamics of Earth’s Surface Water. Nature; 2016; 540, pp. 348-349. [DOI: https://dx.doi.org/10.1038/nature21100]

15. Pekel, J.-F.; Cottam, A.; Gorelick, N.; Belward, A.S. High-Resolution Mapping of Global Surface Water and Its Long-Term Changes. Nature; 2016; 540, pp. 418-422. [DOI: https://dx.doi.org/10.1038/nature20584]

16. Donchyts, G.; Baart, F.; Winsemius, H.; Gorelick, N.; Kwadijk, J.; van de Giesen, N. Earth’s Surface Water Change over the Past 30 Years. Nat. Clim. Chang.; 2016; 6, pp. 810-813. [DOI: https://dx.doi.org/10.1038/nclimate3111]

17. Keshtkar, H.; Voigt, W.; Alizadeh, E. Land-Cover Classification and Analysis of Change Using Machine-Learning Classifiers and Multi-Temporal Remote Sensing Imagery. Arab. J. Geosci.; 2017; 10, 154. [DOI: https://dx.doi.org/10.1007/s12517-017-2899-y]

18. Mahdavi, S.; Salehi, B.; Granger, J.; Amani, M.; Brisco, B.; Huang, W. Remote Sensing for Wetland Classification: A Comprehensive Review. GISci. Remote Sens.; 2018; 55, pp. 623-658. [DOI: https://dx.doi.org/10.1080/15481603.2017.1419602]

19. Rodriguez-Galiano, V.F.; Ghimire, B.; Rogan, J.; Chica-Olmo, M.; Rigol-Sanchez, J.P. An Assessment of the Effectiveness of a Random Forest Classifier for Land-Cover Classification. ISPRS J. Photogramm. Remote Sens.; 2012; 67, pp. 93-104. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2011.11.002]

20. Ludwig, C.; Walli, A.; Schleicher, C.; Weichselbaum, J.; Riffler, M. A Highly Automated Algorithm for Wetland Detection Using Multi-Temporal Optical Satellite Data. Remote Sens. Environ.; 2019; 224, pp. 333-351. [DOI: https://dx.doi.org/10.1016/j.rse.2019.01.017]

21. Ozesmi, S.L.; Bauer, M.E. Satellite Remote Sensing of Wetlands. Wetl. Ecol. Manag.; 2002; 10, pp. 381-402. [DOI: https://dx.doi.org/10.1023/A:1020908432489]

22. Pena-Regueiro, J.; Sebastiá-Frasquet, M.-T.; Estornell, J.; Aguilar-Maldonado, J.A. Sentinel-2 Application to the Surface Characterization of Small Water Bodies in Wetlands. Water; 2020; 12, 1487. [DOI: https://dx.doi.org/10.3390/w12051487]

23. Bwangoy, J.-R.B.; Hansen, M.C.; Roy, D.P.; Grandi, G.D.; Justice, C.O. Wetland Mapping in the Congo Basin Using Optical and Radar Remotely Sensed Data and Derived Topographical Indices. Remote Sens. Environ.; 2010; 114, pp. 73-86. [DOI: https://dx.doi.org/10.1016/j.rse.2009.08.004]

24. Davranche, A.; Poulin, B.; Lefebvre, G. Mapping Flooding Regimes in Camargue Wetlands Using Seasonal Multispectral Data. Remote Sens. Environ.; 2013; 138, pp. 165-171. [DOI: https://dx.doi.org/10.1016/j.rse.2013.07.015]

25. Islam, M.A.; Thenkabail, P.S.; Kulawardhana, R.W.; Alankara, R.; Gunasinghe, S.; Edussriya, C.; Gunawardana, A. Semi-automated Methods for Mapping Wetlands Using Landsat ETM+ and SRTM Data. Int. J. Remote Sens.; 2008; 29, pp. 7077-7106. [DOI: https://dx.doi.org/10.1080/01431160802235878]

26. Kulawardhana, R.W.; Thenkabail, P.S.; Vithanage, J.; Biradar, C.; Islam, M.A.; Gunasinghe, S.; Alankara, R. Evaluation of the Wetland Mapping Methods Using Landsat ETM+ and SRTM Data. J. Spat. Hydrol.; 2007; 7, pp. 62-96.

27. Ma, L.; Fu, T.; Blaschke, T.; Li, M.; Tiede, D.; Zhou, Z.; Ma, X.; Chen, D. Evaluation of Feature Selection Methods for Object-Based Land Cover Mapping of Unmanned Aerial Vehicle Imagery Using Random Forest and Support Vector Machine Classifiers. ISPRS Int. J. Geo Inf.; 2017; 6, 51. [DOI: https://dx.doi.org/10.3390/ijgi6020051]

28. Tang, Z.; Li, Y.; Gu, Y.; Jiang, W.; Xue, Y.; Hu, Q.; LaGrange, T.; Bishop, A.; Drahota, J.; Li, R. Assessing Nebraska Playa Wetland Inundation Status during 1985–2015 Using Landsat Data and Google Earth Engine. Environ. Monit. Assess.; 2016; 188, 654. [DOI: https://dx.doi.org/10.1007/s10661-016-5664-x]

29. Abdel-Hamid, A.; Dubovyk, O.; Abou El-Magd, I.; Menz, G. Mapping Mangroves Extents on the Red Sea Coastline in Egypt Using Polarimetric SAR and High Resolution Optical Remote Sensing Data. Sustainability; 2018; 10, 646. [DOI: https://dx.doi.org/10.3390/su10030646]

30. Franklin, S.E.; Skeries, E.M.; Stefanuk, M.A.; Ahmed, O.S. Wetland Classification Using Radarsat-2 SAR Quad-Polarization and Landsat-8 OLI Spectral Response Data: A Case Study in the Hudson Bay Lowlands Ecoregion. Int. J. Remote Sens.; 2018; 39, pp. 1615-1627. [DOI: https://dx.doi.org/10.1080/01431161.2017.1410295]

31. Maxwell, A.E.; Warner, T.A.; Fang, F. Implementation of Machine-Learning Classification in Remote Sensing: An Applied Review. Int. J. Remote Sens.; 2018; 39, pp. 2784-2817. [DOI: https://dx.doi.org/10.1080/01431161.2018.1433343]

32. Kordelas, G.A.; Manakos, I.; Aragonés, D.; Díaz-Delgado, R.; Bustamante, J. Fast and Automatic Data-Driven Thresholding for Inundation Mapping with Sentinel-2 Data. Remote Sens.; 2018; 10, 910. [DOI: https://dx.doi.org/10.3390/rs10060910]

33. Lefebvre, G.; Davranche, A.; Willm, L.; Campagna, J.; Redmond, L.; Merle, C.; Guelmami, A.; Poulin, B. Introducing WIW for Detecting the Presence of Water in Wetlands with Landsat and Sentinel Satellites. Remote Sens.; 2019; 11, 2210. [DOI: https://dx.doi.org/10.3390/rs11192210]

34. Huang, W.; DeVries, B.; Huang, C.; Lang, M.W.; Jones, J.W.; Creed, I.F.; Carroll, M.L. Automated Extraction of Surface Water Extent from Sentinel-1 Data. Remote Sens.; 2018; 10, 797. [DOI: https://dx.doi.org/10.3390/rs10050797]

35. Shao, Y.; Lunetta, R.S. Comparison of Support Vector Machine, Neural Network, and CART Algorithms for the Land-Cover Classification Using Limited Training Data Points. ISPRS J. Photogramm. Remote Sens.; 2012; 70, pp. 78-87. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2012.04.001]

36. Whyte, A.; Ferentinos, K.P.; Petropoulos, G.P. A New Synergistic Approach for Monitoring Wetlands Using Sentinels-1 and 2 Data with Object-Based Machine Learning Algorithms. Environ. Model. Softw.; 2018; 104, pp. 40-54. [DOI: https://dx.doi.org/10.1016/j.envsoft.2018.01.023]

37. Xu, X.; Li, W.; Ran, Q.; Du, Q.; Gao, L.; Zhang, B. Multisource Remote Sensing Data Classification Based on Convolutional Neural Network. IEEE Trans. Geosci. Remote Sens.; 2018; 56, pp. 937-949. [DOI: https://dx.doi.org/10.1109/TGRS.2017.2756851]

38. Thanh Noi, P.; Kappas, M. Comparison of Random Forest, k-Nearest Neighbor, and Support Vector Machine Classifiers for Land Cover Classification Using Sentinel-2 Imagery. Sensors; 2018; 18, 18. [DOI: https://dx.doi.org/10.3390/s18010018]

39. Sheykhmousa, M.; Mahdianpari, M.; Ghanbari, H.; Mohammadimanesh, F.; Ghamisi, P.; Homayouni, S. Support Vector Machine Versus Random Forest for Remote Sensing Image Classification: A Meta-Analysis and Systematic Review. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2020; 13, pp. 6308-6325. [DOI: https://dx.doi.org/10.1109/JSTARS.2020.3026724]

40. Zhang, L.; Hu, Q.; Tang, Z. Assessing the Contemporary Status of Nebraska’s Eastern Saline Wetlands by Using a Machine Learning Algorithm on the Google Earth Engine Cloud Computing Platform. Environ. Monit. Assess.; 2022; 194, 193. [DOI: https://dx.doi.org/10.1007/s10661-022-09850-8]

41. Dronova, I. Object-Based Image Analysis in Wetland Research: A Review. Remote Sens.; 2015; 7, pp. 6380-6413. [DOI: https://dx.doi.org/10.3390/rs70506380]

42. Jia, K.; Jiang, W.; Li, J.; Tang, Z. Spectral Matching Based on Discrete Particle Swarm Optimization: A New Method for Terrestrial Water Body Extraction Using Multi-Temporal Landsat 8 Images. Remote Sens. Environ.; 2018; 209, pp. 1-18. [DOI: https://dx.doi.org/10.1016/j.rse.2018.02.012]

43. Mahdianpari, M.; Brisco, B.; Granger, J.E.; Mohammadimanesh, F.; Salehi, B.; Banks, S.; Homayouni, S.; Bourgeau-Chavez, L.; Weng, Q. The Second Generation Canadian Wetland Inventory Map at 10 Meters Resolution Using Google Earth Engine. Can. J. Remote Sens.; 2020; 46, pp. 360-375. [DOI: https://dx.doi.org/10.1080/07038992.2020.1802584]

44. Tamiminia, H.; Salehi, B.; Mahdianpari, M.; Quackenbush, L.; Adeli, S.; Brisco, B. Google Earth Engine for Geo-Big Data Applications: A Meta-Analysis and Systematic Review. ISPRS J. Photogramm. Remote Sens.; 2020; 164, pp. 152-170. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2020.04.001]

45. Farda, N.M. Multi-Temporal Land Use Mapping of Coastal Wetlands Area Using Machine Learning in Google Earth Engine. IOP Conf. Ser. Earth Environ. Sci.; 2017; 98, 012042. [DOI: https://dx.doi.org/10.1088/1755-1315/98/1/012042]

46. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-Scale Geospatial Analysis for Everyone. Remote Sens. Environ.; 2017; 202, pp. 18-27. [DOI: https://dx.doi.org/10.1016/j.rse.2017.06.031]

47. LaGrange, T. Wetland Program Plan for Nebraska; Nebraska Game and Parks Commission: Lincoln, NE, USA, 2010.

48. Gersib, R.A. Nebraska Wetlands Priority Plan; Nebraska Game and Parks Commission: Lincoln, NE, USA, 1991.

49. Lang, M.W.; McCarty, G.W. Lidar Intensity for Improved Detection of Inundation below the Forest Canopy. Wetlands; 2009; 29, pp. 1166-1178. [DOI: https://dx.doi.org/10.1672/08-197.1]

50. Tang, Z.; Li, R.; Li, X.; Jiang, W.; Hirsh, A. Capturing LiDAR-Derived Hydrologic Spatial Parameters to Evaluate Playa Wetlands. JAWRA J. Am. Water Resour. Assoc.; 2014; 50, pp. 234-245. [DOI: https://dx.doi.org/10.1111/jawr.12125]

51. Mullissa, A.; Vollrath, A.; Odongo-Braun, C.; Slagter, B.; Balling, J.; Gou, Y.; Gorelick, N.; Reiche, J. Sentinel-1 SAR Backscatter Analysis Ready Data Preparation in Google Earth Engine. Remote Sens.; 2021; 13, 1954. [DOI: https://dx.doi.org/10.3390/rs13101954]

52. Brown, C.F.; Brumby, S.P.; Guzder-Williams, B.; Birch, T.; Hyde, S.B.; Mazzariello, J.; Czerwinski, W.; Pasquarella, V.J.; Haertel, R.; Ilyushchenko, S. et al. Dynamic World, Near Real-Time Global 10 m Land Use Land Cover Mapping. Sci. Data; 2022; 9, 251. [DOI: https://dx.doi.org/10.1038/s41597-022-01307-4]

53. Tang, Z.; Drahota, J.; Hu, Q.; Jiang, W. Examining Playa Wetland Contemporary Conditions in the Rainwater Basin, Nebraska. Wetlands; 2018; 38, pp. 25-36. [DOI: https://dx.doi.org/10.1007/s13157-017-0953-x]