Advertisement

Remote Sensing in Earth Systems Sciences

, Volume 2, Issue 4, pp 173–182 | Cite as

Mapping Urbanization Trends in a Forested Landscape Using Google Earth Engine

  • Hamdi A. ZurqaniEmail author
  • Christopher J. Post
  • Elena A. Mikhailova
  • Jeffrey S. Allen
Original Paper
  • 500 Downloads

Abstract

Land cover change is one of the most important issues facing the landscape of the Southeastern United States. Land use change impacts the natural environment, and it is critical to understand the location and rate of change in forests near rapidly urbanizing areas. The objectives of this study are to determine the classes and the distribution of land cover using classification of high-resolution satellite imagery for the upstate region of South Carolina and identify urbanization trends (conversion of forested areas to residential or industrial developments). Rapid urbanization has occurred in the Southeastern U.S. because of economic development and population growth. The challenge is to develop methodologies to quickly identify how the landscape is being altered as forested areas are developed. Remote sensing techniques using newly available high-resolution imagery have great potential for providing up-to-date spatial information about the land cover change. In this study, a framework has been developed to regularly monitor land cover change using a new geospatial technology platform: Google Earth Engine (GEE). The overall accuracy assessments of the 3 years were 91.21% (2013), 90.46% (2015), and 91.01% (2017), respectively. Based on the classification results, urbanization activities have resulted in a gradual change of land cover classes. The predominant land cover alteration at each time interval was changes from forested areas to the new development lands, bare lands, and non-forested areas.

Keywords

Ecosystem High-resolution imagery Land cover Unsupervised classification Urban 

1 Introduction

Land use change detection is essential for better understanding the interactions and relationships between human activities and the environment over time. Land use change is a common and accelerating process due to rapid urbanization, forest conversion, and agricultural expansion [1, 2]. Land use change is the most remarkable indicator of human footprint, and it is considered the most important driver of loss of biodiversity and land degradation [3]. The effect of the change in land use varies by region and geographical location. At a smaller spatial scale, it was observed that conversion of land from forested and agricultural uses to urban and suburban uses can degrade aquatic ecosystems in small streams in the upstate region of South Carolina, with the impacts being particularly destructive during the actual land conversion process [4, 5, 6].

Forest disturbance is a key process in the development of forest ecosystems, and advances in remote sensing and geospatial technology have greatly increased the ability to visualize and record this disturbance [6, 7, 8, 9]. Monitoring forest disturbance over time is crucial to better understand the history of land use. Cohen et al. [8] studied forest disturbance from 1985 to 2012 across the USA and reported that harvest areas were more evident in the more heavily forested regions where the national rates of disturbance ranged between 1.5 and 4.5% of forest area per year. Huang et al. [9] were able to detect most stand clearing disturbance events, including harvest, fire, and urban development in the Western U.S. using an automated algorithm called the vegetation change tracker (VCT) that uses Landsat time series data. Zurqani et al. [6] studied the loss and gain of land cover in the Savannah River basin, Southeastern U.S., using a historical Landsat imagery, and they revealed that the major land use change observed in the area was the deforestation and reforestation of forest areas during the entire study period.

South Carolina has limited open space protections and was ranked third among 50 states in related regulations, according to a Sierra Club report [10]. Campbell et al. [4] pointed out that developed land in the upstate South Carolina grew from 222,745 acres to 576,336 acres between the years 1990 and 2000, and they anticipated these areas to grow to 1,523,667 acres by the year 2030 based on a GIS-based logistic regression model.

The use of remote sensing techniques for land use analysis in the urban areas has been examined in many studies [7, 11, 12, 13]. These studies indicate that using remote sensing data such as Landsat, SPOT, and National Agriculture Imagery Program (NAIP) imagery can help characterize the change in urbanization on a large scale. To improve urban forest planning and management at the fine or neighborhood scale, high spatial resolution imagery is appropriate to identify individual tree locations [14, 15]. Zhang [15] pointed out that the use of the high spatial resolution data for small regions can produce accurate land cover maps.

Traditional techniques of mapping and monitoring land cover/use change such as deforestation, urban growth, agricultural expansion, and wetland loss require downloading remote sensing imagery and having adequate CPU processing power to classify imagery. This is often expensive in terms of both financial cost and time, especially when large areas and/or long-time periods are analyzed. On December 2, 2010, Google launched a new technology named Google Earth Engine (GEE) (U.S. Geological Survey 2010) [16]. This new geospatial analysis platform made much of the freely available satellite imagery available online so that researchers can analyze changes to the Earth’s surface in near real time [6, 17, 18]. Google Earth Engine allows users to run algorithms on a large archive of georeferenced images and other data within Google’s infrastructure. This study utilizes this new geospatial technology and the NAIP high-resolution imagery that are freely available in GEE to investigate the land use change within the urbanizing areas of Greenville County in upstate South Carolina, USA.

The objectives of the study were to: (1) determine the classes and the distribution of land cover using classification of high-resolution satellite imagery for the upstate region of South Carolina, (2) identify where forested areas have been converted to residential or industrial developments, and (3) discuss the potential impacts of land use change in this region.

2 Materials and Methods

2.1 Study Area

The study area (Fig. 1) is approximately 2114.47 km2. The region is characterized by a humid subtropical climate with a mean annual temperature of 15.7 °C and a mean annual precipitation of 1267 mm [19]. High rainfall amounts may contribute to a greater risk of soil erosion by water [20]. Topography ranges from gently rolling hills and pasture land to mountainous terrain in the northern portions of the county [20]. Soils in the region are typically Ultisols (containing high levels of silicate clay such as kaolinite), Entisols, and Inceptisols [21].
Fig. 1

Location of the study area: Greenville County, SC

2.2 Data Processing

The change detection technique for identifying a change in land cover requires image preprocessing and normalization, a reference dataset, and land cover classification. The land use change detection technique was applied and evaluated by developing code in the GEE platform using the unsupervised clustering algorithm which was used to classify high-resolution NAIP aerial imagery individually for each study year. The NAIP imagery was also used to generate the reference dataset for validating the classifications, which is possible because of its high spatial resolution. Finally, the type of land use change detection was determined using a post-classification approach by overlying the land cover classification maps, and the change detection was described and identified using NDVI Change Ratio to Previous Year (RPNDVI) approach. The general procedures are summarized in the flowchart illustrated in (Fig. 3).

2.3 Image Preprocessing

The approach is based on the unsupervised classification k-mean algorithm [22], using the high-resolution NAIP imagery. Unsupervised classification is a method often used for the quantitative analysis of remote sensing images. Google Earth Engine offers reduced time in performing analysis by utilizing Google’s computing infrastructure. Preprocessed NAIP imagery, available through GEE, was used to assess land use/land cover change across the study area (Table 1). All the data processing was conducted using the cloud-computing technology in the GEE platform (https://earthengine.google.org/). The NAIP imagery scenes were used in this study for the years 2013, 2015, and 2017, respectively (Table 1; Fig. 3). The plot of the reflectance values against the chosen NAIP imagery bands and reflectance values against wavelengths of the land cover types at known points in our study area (Fig. 2) shows a clear spectral separability of the land cover types. The near-infrared band has the highest spectral separability to distinguish among the different land cover types. The red band provides a useful feature for delineating forests and non-forests based on chlorophyll reflectance. Other indices derived from spectral band combinations were also used to distinguish features that are more representative of vegetation greenness, such as the Normalized Difference Vegetation Index (NDVI) [6, 23], the Enhanced Vegetation Index (EVI) [7, 24], the Green Ratio Vegetation Index (GRVI) [25], and the Modified Soil-Adjusted Vegetation Index (MSAVI) [7, 26]. The Normalized Difference Water Index (NDWI) [27] was used to better distinguish water areas [6]. All these indices were performed for each image and stacked for later classification (Fig. 3). These indices are expressed in the following equations (Eqs. (1), (2), (3), (4), and (5)):
Table 1

Data sources and description

Data layer

Source

Spatial resolution

Date

Aerial imagery (NAIP)

Google Earth Engine (GEE)

1 m

2013, 2015, and 2017

Study area boundary

Google Earth Engine (GEE)

n/a

2017

Fig. 2

The spectral values of each land cover type in the study area. a Spectral band reflectance (R, red; G, green; B, blue; and NIR, near infrared) and b spectral indices (NDVI, Normalized Difference Vegetation Index; EVI, Enhanced Vegetation Index; GRVI, Green–Red Vegetation Index; MSAVI, Modified Soil-Adjusted Vegetation Index; and NDWI, Normalized Difference Water Index

Fig. 3

Flowchart of data processing and classification

$$ NDVI=\frac{\left(\mathrm{NIR}-\mathrm{Red}\right)}{\left(\mathrm{NIR}+\mathrm{Red}\right)} $$
(1)
$$ EVI=G\frac{\left(\mathrm{NIR}-R\right)}{\left(\mathrm{NIR}+C1\times \mathrm{Red}-C2\times \mathrm{Blue}+L\right)}\times 100 $$
(2)
where red, green, blue, and NIR are the NAIP imagery bands. The coefficients adopted in the MODIS-EVI algorithm are L = 1, C1 = 6, C2 = 7.5, and G (gain factor) = 2.5.
$$ GRVI=\mathrm{NIR}/\mathrm{Green}\times 100 $$
(3)
$$ MSAVI=\frac{2\times \mathrm{NIR}+1-\kern0.5em \sqrt{{\left(2\times \mathrm{NIR}+1\right)}^2-8\times \left(\mathrm{NIR}-\mathrm{Red}\right)}}{2}\times 100 $$
(4)
$$ NDWI=\frac{\left(\mathrm{Green}-\mathrm{NIR}\right)}{\left(\mathrm{Green}+\mathrm{NIR}\right)} $$
(5)

2.4 Image Classification

A wide variety of classification algorithms have been used to map the changes in land cover/land use from remotely sensed data [3, 28]. The unsupervised classification has been shown effective in terms of classification accuracy and stronger noise resistance and is less time-consuming than other techniques [29, 30, 31]. For example, Sader et al. [29] obtained an overall accuracy of 80% for forest and wetland classifications in Maine using Landsat TM and GIS rule–based model with supervised classifications. An 8% improvement was obtained when the statistical clustering functions of unsupervised classifications were applied. Unsupervised classification is a classification of groups (groupings of pixels with common characteristics), and no extensive prior knowledge of the region is required. In this study, an unsupervised classification procedure was applied by employing a k-mean clustering algorithm in Google Earth Engine (GEE) platform (Appendix A in the Electronic Supplementary Material (ESM)). In this approach, pixel reflectance values are separated into clusters which represent the most general natural spectral groupings. The unsupervised classification technique is sensitive to the number of clusters, and therefore, the optimum number of clusters was determined to be 100 clusters after experimenting and visually evaluating a range of from 10 to 1000 clusters. The number of clusters for each of the land cover classes was selected based on visual assessment. Six land cover classes were identified across the study area using the National Land Cover Database (NLCD) maps and description [32], and the high-resolution imagery National Agriculture Imagery Program (NAIP) (1-m resolution) as a reference [6]. The identified classes were: (1) open water with rivers, lakes, and standing water bodies; (2) low–medium-intensity urban with paved roads, concrete, and warehouses; (3) high-intensity urban with densely populated areas; (4) barren land; (5) forest which includes the deciduous and evergreen forests, and woody wetlands; (6) non-forest which includes the shrub/scrub, grassland/herbaceous, pasture/hay, and cultivated crops.

2.5 Accuracy Assessments

Many factors affect the accuracy of image classification. Accuracy assessments are useful and effective techniques to determine how well the classification process was accomplished [29, 33, 34]. The accuracy assessment process allows the analyst to compare certain pixel values in a raster layer to the reference pixels for which the class is known. The produced land cover classification maps were validated using high-resolution imagery (NAIP) (1-m resolution). A total of about 600 reference points were delineated using a random stratified sample design for the years 2013, 2015, and 2017, respectively, with no less than 50 reference points per land cover category. Following the previous studies [6, 29, 33, 34], a confusion matrix of land cover maps was calculated to evaluate the accuracy of the results using producer’s accuracy, user’s accuracy, the overall accuracy, kappa statistics [35], and F1 score which shows how good the classifier is in the context of both producer’s and user’s by weighting the average of producer’s and user’s [6, 36] as shown in Eq. (6):

$$ \mathrm{F}1\ \mathrm{score}=\frac{2}{\frac{1}{\mathrm{producer}'\mathrm{s}} + \frac{1}{\mathrm{user}'\mathrm{s}\ }}\kern0.5em =2.\frac{\mathrm{user}'\mathrm{s}\times \mathrm{producer}'\mathrm{s}\ }{\mathrm{user}'\mathrm{s}+\mathrm{producer}'\mathrm{s}\ } $$
(6)

2.6 Land Use Change Detection

Change detection analysis is the process that designates differences between images of the same scene at different times. The post-classification comparison approach was used by using the land cover classified images to calculate the area of each land cover class and observe the changes that are taking place at that time [37]. This analysis identifies the various changes that occur in land cover for each class. To detect one type of land cover change such as forest change, a single change index with a fixed threshold is typically sufficient [38]. The change detection to identify the land use change in the area was identified by image difference technique. The multi-date differencing NDVI bands were used to determine the change in each pixel by using NDVI Change Ratio to Previous Year (RPNDVI) [39] as shown in the following equation (Eq. (7)):
$$ RPNDVI=\frac{NDVI_i\left(x,y\right)-{NDVI}_{i-1}\left(x,y\right)}{NDVI_{i-1}\left(x,y\right)}\times 100\% $$
(7)
where NDVIi (x, y) is the NDVI value to the current year at location (x, y) and NDVIi − 1 (x, y) is the NDVI value to the previous year.

The change detection approach was applied with a threshold value of negative (1.1) to the RPNDVI layer to detect the change between the years 2013 to 2015 and 2015 to 2017, and over the whole study period between the years 2013 to 2017. The noise of NDVI change ratio was filtered by grouping the pixels based on size and eliminating the small patches. This approach provides information that illustrates “from–to” change that has occurred which can be easily calculated and mapped [6]. The pixels with change in this approach will be flagged as “probable change.”

3 Results and Discussion

3.1 Classes and the Distribution of Land Cover Using Classification of High-Resolution Satellite Imagery for the Upstate Region of South Carolina

The land cover classification maps were produced using unsupervised classification for the years 2013, 2015, and 2017 in a total of six land cover categories were identified and classified in the study area: high-intensity urban, low–medium-intensity urban, forested areas, non-forested areas, bare soil, and water as shown in Fig. 4. The individual land cover class areas for the 3 years are summarized in Table 3. The overall accuracies were 91.21% (2013), 90.46% (2015), and 91.01% (2017), respectively (Table 2). Based on the classification results, urbanization activities have resulted in a gradual spread of land cover classes. Therefore, the information on the maps point to a repeated dispersion of the loss and gain in land use classes within the Greenville County (Fig. 5). The majority of the land use change in the study area was a result of urbanization and increase of deforestation.
Fig. 4

Distribution of land cover for the 3 years: a 2013, b 2015, and c 2017

Table 2

The percentage of producer and user accuracy, F1 score, overall accuracy, and kappa statistic for land cover classification

Land cover classes

2013

2015

2017

User’s accuracy

Producer’s accuracy

F1 score

User’s accuracy

Producer’s accuracy

F1 score

User’s accuracy

Producer’s accuracy

F1 score

Forested (FOR)

96.93

94.47

93.59

95.12

82.97

88.63

98.52

89.59

93.84

Non-forested

86.92

98.26

92.24

73.21

97.61

83.67

74.62

98.00

84.73

High-intensity urban

89.47

81.01

85.00

93.22

91.45

92.33

92.37

91.16

91.76

Low–medium-intensity urban

91.30

80.76

85.71

89.85

74.69

81.57

91.71

73.09

81.36

Bare land

88.88

61.53

72.72

78.57

64.70

70.96

75.61

72.09

73.80

Water

94.11

100.00

96.96

95.34

99.65

97.45

96.91

99.73

98.30

Overall accuracy

91.21

  

90.46

  

91.01

  

Kappa coefficient

87.38

  

87.44

  

88.70

  
Fig. 5

An association plot of the land cover–type loss and gain in each time interval (FOR, forested; NFOR, non-forested; HIU, high-intensity urban; LMIU, low–medium-intensity urban; BL, bare land; W, water)

Over the whole study period, the urban areas experienced a remarkable change where in the first year (2013), the high-intensity urban and low–medium-intensity urban were 68.10 km2 (3.22%) and 76.53 km2 (3.62%), respectively. In the second year (2015), these areas increased to 77.77 km2 (3.68%) and 92.11 km2 (4.36%), respectively. In 2017, these areas showed another rise to 89.72 km2 (4.24%) and 110.85 km2 (5.24%), respectively. The urban areas and infrastructure expansion were continuously increasing over time. The barren land area was 13.12 km2 (0.62% of the study area) in 2013. The barren land category area experienced a substantial increase to 25.73 km2 (1.22%) in 2015, and this area decreased to 14.33 km2 (0.68%) in the last year of the study period in (2017, Table 3).
Table 3

Distribution of land cover in 2013, 2015, and 2017

Land cover classes

Areas

2013

2015

2017

km2

(%)

km2

(%)

km2

(%)

Forested

1402.52

(66.33)

1354.31

(64.05)

1326.93

(62.75)

Non-forested

532.85

(25.20)

543.76

(25.71)

553.19

(26.16)

High-intensity urban

68.10

(3.22)

77.77

(3.68)

89.72

(4.24)

Low–medium-intensity urban

76.53

(3.62)

92.11

(4.36)

110.85

(5.24)

Bare land

13.12

(0.62)

25.73

(1.22)

14.33

(0.68)

Water

21.35

(1.01)

20.79

(0.98)

19.45

(0.92)

Total

2114.47

(100.00)

2114.47

(100.00)

2114.47

(100.00)

Forested areas experienced the highest decline while the non-forested areas were increasing (Table 3). In the first study year (2013), the forested areas were 1402.52 km2 (66.33%), while the non-forested areas were 532.85 km2 (25.20%); in the second year (2015), forested areas declined to 1354.31 km2 (64.05%), while the non-forested areas increased to 543.76 km2 (25.71%). In the last year of the study period (2017), the forested areas decreased to 1326.93 km2 (62.75%), while the non-forested increased to 553.19 km2 (26.16%), respectively. Water bodies showed little change over the study period where it was 21.35 km2 or 1.01% in the first year (2013) and 20.79 km2 or 0.98% in the second year (2015). However, the water areas declined slightly to 19.45 km2 (0.92% of the study area) over the whole study period (2017).

3.2 Land Use Change Detection Using Post-Classification Comparison Approach

For the whole study period, the predominant land cover alteration at each time interval was generally caused by changes from forested areas to the new development lands, bare lands, and non-forested areas. The association plot in Fig. 5 indicates that loss of the forested areas is greater than would be expected especially during the whole study period especially during the relatively short time interval between 2013 and 2017. While the non-forested areas showed little overall gain and loss during the whole study period, the urban areas show gains over the whole study period which was clearly evident between 2013 and 2017. Bare lands show greater than would be expected gains in the first and the second intervals (2013–2015 and 2015–2017), while these areas show an unexpected loss during the whole study period between 2013 and 2017. The water bodies showed little loss and gain in all time intervals.

3.3 Land Use Change Detection Using NDVI Change Ratio

NDVI change detection is a powerful alternative to more traditional approaches used in land cover classification [6, 12, 38, 40]. In this study, NDVI Change Ratio to Previous Year (RPNDVI) change detection approach helped clarify and provide a better understanding of the urbanization activities in the forested areas. However, the impact of the tonal variations in NAIP images affects the result of the NDVI change ratio approach. This approach worked better to detect the overall change (2013–2017), while showing more noisy results in the first interval (2013–2015) and the second interval (2015–2017) (Fig. 6).
Fig. 6

The hotspot area of the land use change from 2013 to 2015 and from 2015 to 2017 and the overall change from 2013 to 2017. a First interval (2013–2015). b Second interval (2015–2017). c Overall change (2013–2017)

The results indicated that the main land cover change during the study period occurred in forested areas. The change detection results indicate that the land use change areas in the first interval between 2013 and 2015 were about 124.2 km2 (5.87%). In the second interval between 2015 and 2017, the land use change areas were about 82.33 km2 (3.89%). Over the whole study period between 2013 and 2017, the land use change areas were approximately 216.51 km2 (10.23%) (Table 4; Fig. 6).
Table 4

Land use change areas from 2013 to 2015, 2015 to 2017, and the overall change from 2013 to 2017

Time interval

Land use change

Coverage area

km2 (%)

2013–2015

Land use change

124.20 (5.87)

2015–2017

Land use change

82.33 (3.89)

2013–2017

Land use change

216.51 (10.23)

3.4 Potential Impacts of Land Use Change

Understanding the land use change and its potential impact is key to the future success of increasing environmental awareness and monitoring [6]. Producing accurate maps of land cover over time can provide important information for a better understanding of urban ecosystems and help improve environmental quality and human health in urban areas. Unlike traditional GIS software, utilizing Google Earth Engine makes it much easier to integrate remote sensing data as well as the image normalization process to improve the classification accuracy [12]. In this study, the NAIP imagery and its normalization results have a significant contribution to decreasing the confusion between diverse types of land cover (Figs. 2 and 4).

New residential development areas in the Southeastern United States peaked in early 2006, but since then declined to leave many residential developments stalled in various stages of construction [41]. Greenville County is the most populous county in South Carolina, and it continues growing up at an average rate of 2.1% per year (https://www.greenvillecounty.org/). Comparison of the land cover classification results over the entire study period indicates a clear increase in the urban areas coupled with a decrease in the forested area (Table 3). Changes in the forested area were illustrated as where the deforestation and the urbanization activities occur using the PNDVI-revealed evidence supporting the case of this study. The accelerated development of urban areas has replaced natural vegetation with impervious surfaces such as buildings, paved roads, and parking lots [6]. The impervious surfaces are associated with various anthropogenic and economic activities generating pollutants, such as the discharge of residential and industrial sewage [6]. In addition, a higher percentage of impervious surfaces prevent rainfall infiltration into the soil, leading to contaminating the nearby streams by the soluble and particulate forms of pollutants through surface runoff [6]. These paved areas affect the temperature of stormwater runoff and increase the environmental hazard in the area especially on the flooding season [42]. Forested areas typically experience little to no management and largely need to have constant attention over time. In future urban development, urban planners should be aware of the adverse impacts of impervious surfaces.

4 Conclusions

The accessibility of historical remotely sensed data as well as the new geospatial technology of Google Earth Engine represents a significant improvement for monitoring and evaluating land use change over time. This study successfully developed a regional scale analysis using high-resolution imagery, determine the classes and the distribution of land cover in Greenville, SC, and identified the spatial and the temporal changes of the land cover types that occurred as a consequence of land use change over time. Multiple layers were used including the original four bands RGB and NIR, NDVI, EVI, GRVI, MSAVI, and NDWI which provided reliable results in classifying six general land cover types.

Classifying high spatial resolution aerial imagery using remote sensing analysis is an efficient way to get accurate land cover maps of individual cities. This study successfully produced a general land cover classification maps for the city of Greenville, SC, for the years 2013, 2015, and 2017. The overall accuracy of the 3 years exceeded 90% for the six classes of high-intensity urban, low–medium-intensity urban, forested areas, non-forested areas, bare soil, and water. The potential impact of the land use change took place near the center of the Greenville County, and most of the direct impact on the study area appeared to be clearing of the trees and the establishment of new development areas.

This methodology is available at no cost to the users, and the advantage of this proposed approach can be useful for describing detailed land cover types within the locations of different areas of the city. This type of information facilitates the research and management of the urban ecosystem to improve environmental quality and human health in the area. The limitation of this approach is that the NAIP imagery is not yearly available at some locations. In future research, continuous monitoring of land use change is needed to better understand its impact in the region and which returns more effective management strategies.

Notes

Funding Information

Financial support for this project was provided by Libyan Government (Ministry of Higher Education and Scientific Research) on behalf of the University of Tripoli and Clemson University.

Supplementary material

41976_2019_20_MOESM1_ESM.docx (39 kb)
ESM 1 (DOCX 39 kb)

References

  1. 1.
    Drummond A, Loveland R (2010) Land-use pressure and a transition to forest-cover loss in the eastern United States. BioScience 60:286–298CrossRefGoogle Scholar
  2. 2.
    Agaton M, Setiawan Y, Effendi H (2016) Land use/land cover change detection in an urban watershed: a case study of upper Citarum Watershed, West Java Province, Indonesia. Procedia Environ Sci 33:654–660CrossRefGoogle Scholar
  3. 3.
    Butt A, Shabbir R, Ahmad S, Aziz N (2015) Land use change mapping and analysis using remote sensing and GIS: a case study of Simly watershed, Islamabad, Pakistan. Egypt J Remote Sens Space Sci 18:251–259Google Scholar
  4. 4.
    Campbell E, Allen J, Lu S (2007) Modeling growth and predicting future developed land in the upstate of South Carolina. Report submitted to the Saluda-Reedy Watershed Consortium. Strom Thurmond Institute, Clemson University, Clemson, South CarolinaGoogle Scholar
  5. 5.
    Sciera K, Smink J, Morse J, Post C, Pike J, English W, Karanfil T, Hayes J, Schlautman M, Klaine S (2008) Impacts of land disturbance on aquatic ecosystem health: quantifying the cascade of events. Integr Environ Asses 4(4):431–442CrossRefGoogle Scholar
  6. 6.
    Zurqani H, Post C, Mikhailova E, Schlautman A, Sharp J (2018) Geospatial analysis of land use change in the Savannah River Basin using Google Earth Engine. Int J Appl Earth Obs Geoinf 69:175–185CrossRefGoogle Scholar
  7. 7.
    Zurqani HA, Post, CJ, Mikhailova EA, Cope MP, Allen JS, Lytle B (2019) Evaluating the integrity of forested riparian buffers over a large area using LiDAR data and Google Earth Engine; Graduate Research and Discovery Symposium (GRADS): Clemson, SC, USA. 225Google Scholar
  8. 8.
    Cohen B, Yang Z, Stehman V, Schroeder A, Bell M, Masek G, Huang C, Meigs W (2016) Forest disturbance across the conterminous United States from 1985 to 2012: the emerging dominance of forest decline. For Ecol Manag 360:242–252CrossRefGoogle Scholar
  9. 9.
    Huang C, Goward N, Masek G, Thomas N, Zhu Z, Vogelmann E (2010) An automated approach for reconstructing recent forest disturbance history using dense Landsat time series stacks. Remote Sens Environ 114(1):183–198CrossRefGoogle Scholar
  10. 10.
    The Sierra Club (1999). http://vault.sierraclub.org/sprawl/report99/openspace.asp. Accessed 20 Nov 2018
  11. 11.
    Davies G, Barbosa O, Fuller RA, Tratalos J, Burke N, Lewis D, Warren PH, Gaston KJ (2008) City-wide relationships between green spaces, urban land use and topography. Urban Ecosyst 11(3):269CrossRefGoogle Scholar
  12. 12.
    Huang H, Chen Y, Clinton N, Wang J, Wang X, Liu C, Gong P, Yang J, Bai Y, Zheng Y, Zhu Z (2017) Mapping major land cover dynamics in Beijing using all Landsat images in Google Earth Engine. Remote Sens Environ 202:166–176CrossRefGoogle Scholar
  13. 13.
    Romero H, Ihl M, Rivera A, Zalazar P, Azocar P (1999) Rapid urban growth, land-use changes and air pollution in Santiago, Chile. Atmos Environ 33(24–25):4039–4047CrossRefGoogle Scholar
  14. 14.
    Myeong S, Nowak J, Hopkins F, Brock H (2001) Urban cover mapping using digital, high-spatial resolution aerial imagery. Urban Ecosyst 5(4):243–256CrossRefGoogle Scholar
  15. 15.
    Zhang Y (2001) Texture-integrated classification of urban treed areas in high-resolution color-infrared imagery. Photogramm Eng Rem Sci 67(12):1359–1366Google Scholar
  16. 16.
    U.S. Geological Survey, 2010. Thousands of Landsat scenes in Google’s Earth Engine. URL https://landsat.usgs.gov/google-earth-engine (Accessed 04.16.2018)
  17. 17.
    Zurqani, HA, Post CJ, Mikhailova EA, Ozalas, K, Allen JS, (2019) Geospatial analysis of flooding from Hurricane Florence in the Coastal South Carolina using Google Earth Engine; Graduate Research and Discovery Symposium (GRADS): Clemson, SC, USA, 2019. 230Google Scholar
  18. 18.
    Housman I, Tanpipat V, Biswas T, Clark A, Stephen P, Maus P, Megown K (2015) Monitoring forest change in Southeast Asia: case studies for USAID lowering emissions in Asia’s forests. RSAC-10108-RPT1. U.S. Department of Agriculture, Forest Service, Remote Sensing Applications Center, Salt Lake City, UT, 16Google Scholar
  19. 19.
    SC-DNR. (2005) South Carolina State Climatology Office. http://dnr.sc.gov/climate/sco/Publications/temp_study/temp_precip_main.php (Accessed 06.23.2018)
  20. 20.
    Werts J, Mikhailova E, Post C, Sharp J (2013) Sediment pollution assessment of abandoned residential developments using remote sensing and GIS. Pedosphere 23(1):39–47CrossRefGoogle Scholar
  21. 21.
    Soil Survey Staff (2009) Soil Survey Geographic (SSURGO) database for Greenville County, SC. Natural Resources Conservation Service, United States Department of Agriculture. https://sdmdataaccess.nrcs.usda.gov/ (Accessed 06.20.2018)
  22. 22.
    Arthur D, and Vassilvitskii S (2007) k-means++: the advantages of careful seeding, In: Proc. of the 18th annual ACM-SIAM symposium on discrete algorithms. 1027–1035Google Scholar
  23. 23.
    Tucker CJ (1979) Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens Environ 8(2):127–150CrossRefGoogle Scholar
  24. 24.
    Huete A, Didan K, Miura T, Rodriguez P, Gao X, Ferreira G (2002) Overview of the radiometric and biophysical performance of the MODIS vegetation indices. ISPRS J Photogramm Remote Sens 83(1–2):195–213Google Scholar
  25. 25.
    Li-Hong U, Wei-Xing A, Lin-Zhang A (2007) Predicting grain yield and protein content in winter wheat at different N supply levels using canopy reflectance spectra. Pedosphere 17(5):646–653CrossRefGoogle Scholar
  26. 26.
    Qi J, Chehbouni A, Huete R, Kerr H, Sorooshian S (1994) A modified soil adjusted vegetation index. Remote Sens Environ 48(2):119–126CrossRefGoogle Scholar
  27. 27.
    Gao C (1996) NDWI—a normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens Environ 58(3):257–266CrossRefGoogle Scholar
  28. 28.
    Lam NS (2008) Methodologies for mapping land cover/land use and its change. In: Liang S (ed) Advances in land remote sensing. Springer, Netherlands, pp 341–367CrossRefGoogle Scholar
  29. 29.
    Sader SA, Ahl D, Liou WS (1995) Accuracy of Landsat TM and GIS rule-based methods for forest wetland classification in Maine. Remote Sens Environ 53:133–144CrossRefGoogle Scholar
  30. 30.
    Duda T, Canty M (2002) Unsupervised classification of satellite imagery: choosing a good algorithm. Int J Remote Sens 23(11):2193–2212CrossRefGoogle Scholar
  31. 31.
    Grinias I, Panagiotakis C, Tziritas G (2016) MRF-based segmentation and unsupervised classification for building and road detection in peri-urban areas of high-resolution satellite images. ISPRS J Photogramm Remote Sens 122:145–166CrossRefGoogle Scholar
  32. 32.
    Wickham JD, Stehman SV, Gass L, Dewitz J, Fry JA, Wade TG (2013) Accuracy assessment of NLCD 2006 land cover and impervious surface. Remote Sens Environ 130:294–304CrossRefGoogle Scholar
  33. 33.
    Congalton G (1991) A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens Environ 37(1):35–46CrossRefGoogle Scholar
  34. 34.
    Tsutsumida N, Comber A (2015) Measures of spatio-temporal accuracy for time series land cover data. Int J Appl Earth Obs Geoinf 41:46–55CrossRefGoogle Scholar
  35. 35.
    Congalton G, Green K (2009) Assessing the accuracy of remotely sensed data: principles and practices, 2nd edn. CRC Press, Boca RatonGoogle Scholar
  36. 36.
    Sasaki Y (2007) The truth of the F-measure. Teach Tutor Mat 1:5Google Scholar
  37. 37.
    Yuan F, Sawaya E, Loeffelholz C, Bauer E (2005) Land cover classification and change analysis of the Twin Cities (Minnesota) Metropolitan Area by multitemporal Landsat remote sensing. Remote Sens Environ 98(2–3):317–328CrossRefGoogle Scholar
  38. 38.
    Zhu Z, Woodcock E, Olofsson P (2012) Continuous monitoring of forest disturbance using all available Landsat imagery. Remote Sens Environ 122:75–91CrossRefGoogle Scholar
  39. 39.
    Yang Z, Di L, Yu G, Chen Z (2011) Vegetation condition indices for crop vegetation condition monitoring. In Geoscience and Remote Sensing Symposium (IGARSS), 2011 IEEE International 3534–3537Google Scholar
  40. 40.
    Johansen K, Phinn S, Taylor M (2015) Mapping woody vegetation clearing in Queensland, Australia from Landsat imagery using the Google Earth Engine. Remote Sens Appl: Soc Environ 1:36–49Google Scholar
  41. 41.
    Post C, Ritter B, Akturk E, Breedlove A, Buchanan R, Che C, Fravel J, Hammett L, Kirby T, Mikhailova E, Qiao X (2015) Analysis of factors contributing to abandoned residential developments using remote sensing and geographic information systems (GIS). Urban Ecosyst 18(3):701–713CrossRefGoogle Scholar
  42. 42.
    Perry T, Nawaz R (2008) An investigation into the extent and impacts of hard surfacing of domestic gardens in an area of Leeds, United Kingdom. Landsc Urban Plan 86(1):1–13CrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  1. 1.Department of Forestry and Environmental ConservationClemson UniversityClemsonUSA
  2. 2.Department of Soil and Water SciencesUniversity of TripoliTripoliLibya
  3. 3.South Carolina Water Resources CenterClemson UniversityPendletonUSA

Personalised recommendations