Extracting water-related features using reflectance data and principal component analysis of Landsat images

Figures & data

Figure 1. Location of the study area.

Table 1. Derived spectral indices (B2: green band; B3: red band; B4: near-infrared band; B5: shortwave infrared band of the Landsat 7 image).

Index	Landsat band	Authors
AVI: Ashburn vegetation index	2B4 − B3	Ashburn (1979)
NDVI: normalized difference vegetation index	B4 − B3/B4 + B3	Rouse et al. (1974)
RVI: ratio vegetation index	B3/B4	Richardson and Wiegand (1977)
NRVI: normalized ratio vegetation index	RVI − 1/RVI + 1	Baret and Guyot (1991)
TVI: transformed vegetation index		Deering et al. (1975)
CTVI: corrected transformed vegetation index		Perry and Lautenschlager (1984)
TTVI: Thiam’s transformed vegetation index		Thiam (1997)
NDWI: normalized difference water index	B4 − B2/B4 + B2	McFeeters (1996)
NDWI: normalized difference water index	B5 − B4/B5 + B4	Gao (1996)
MNDWI: modified normalized difference water index	B5 − B2/B5 + B2	Xu (2006)

Figure 2. Flowchart of the classification process.

Figure 3. RF classification results using (a) 4–5–7 TM bands; (b) PCA, all TM bands; (c) all WIs; and (d) MNDWI.

Table 2. Producer’s accuracy of water features and overall accuracy (OA) % of the classifications performed on original Landsat 7 bands using RF and SVM classifiers.

Class	Original bands		4–5–7 bands
	RF	SVM	RF	SVM
Non-water	98	100	100	100
Saturated soil	74	76	78	81
Water	100	97	100	100
OA	91	91	93	94

Table 3. Producer’s accuracy of water features and overall accuracy (OA) % of the classifications performed on WIs derived from Landsat 7 bands using RF and SVM classifiers.

Class	MNDWI		AAVI		NDVI		NDWI Gao		NDWI McFeeters
	RF	SVM	RF	SVM	RF	SVM	RF	SVM	RF	SVM
Non-water	100	100	87	86	99	98	91	92	93	99
Saturated soil	93	91	3	4	9	6	14	5	7	0
Water	100	100	39	56	93	86	83	85	99	99
OA	98	97	43	49	64	63	63	61	67	65
Class	NRVI		RVI		TVI		CTVI		TTVI
	RF	SVM	RF	SVM	RF	SVM	RF	SVM	RF	SVM
Non-water	92	97	98	97	92	97	92	97	92	97
Saturated soil	8	4	9	7	9	7	9	7	9	7
Water	94	86	93	85	94	85	94	85	94	86
OA	65	62	64	63	65	63	65	63	65	63

Table 4. Producer’s accuracy of water features and overall accuracy (OA) % of the classifications with all WIs and with a set of WIs (McF: McFeeters).

Class	All WI		Gao + NDWI(McF) + MNDWI		NDWI(McF) + MNDWI
	RF	SVM	RF	SVM	RF	SVM
Non-water	100	100	100	100	100	100
Saturated soil	94	59	94	72	90	75
Water	100	93	100	100	100	100
OA	98	84	98	91	97	92

Table 5. Producer’s accuracy of water features and overall accuracy (OA) %of the classifications performed on PCs derived from original Landsat 7 bands (TM bands) and WIs, using RF and SVM classifiers.

Class	PCA (WIs)		PCA (TM bands)
	RF	SVM	RF	SVM
Non-water	100	100	100	100
Saturated soil	99	99	100	100
Water	98	100	99	100
OA	99	100	100	100

Table 6. Distribution of water-related categories by land-cover classes of CLC 2000 (minimum and maximum values).

CLC code	CLC type	Proportion of CLC type (%)	Non-water (%)	Saturated soil (%)	Water (%)	NDVI mean ± SD
112	Discontinuous urban fabric	4.30	3.83–4.20	0.09–0.45	0.01–0.02	–0.08 ± 0.10
211	Non-irrigated arable land	68.66	57.67–59.66	8.33–10.17	0.68–0.88	–0.05 ± 0.18
222	Fruit trees and berry plantations	1.46	1.26–1.43	0.03–0.2	0	0.11 ± 0.10
231	Pastures (grassland)	15.38	11.50–12.02	3.12–3.66	0.22–0.25	0.13 ± 0.15
242	Complex cultivation	2.76	2.41–2.48	0.25–0.32	0.02–0.03	–0.01 ± 0.13
243	Agriculture with natural vegetation	3.12	2.20–2.22	0.78–0.80	0.11–0.12	0.02 ± 0.15
311	Broad-leaved forest	1.62	1.03–1.20	0.39–0.56	0.03–0.04	0.14 ± 0.15
324	Transitional woodland shrub	0.50	0.34–0.35	0.12–0.16	0.01	0.16 ± 0.13
0.13	Inland marshes	1.75	0.97–1.03	0.67–0.73	0.06–0.07	0.01 ± 0.13
512	Water bodies	0.44	0.02–0.03	0.07–0.08	0.33–0.34	–0.26 ± 0.14

Figure 4. Bi-plot diagram of the PCs derived from TM bands (categories are grouped by colours: (a) non-water; (b) saturated soil; (c) water; 95% ellipses).

Figure 5. Bi-plot diagram of the PCs derived from WIs (categories are grouped by colours: (a) non-water; (b) saturated soil; (c) water; 95% ellipses).

Figure 6. Map of standard deviation (SD) calculated from the six most accurate classifications including the cumulative histogram of the SD values.

Table 7. Classification uncertainty of the best six classified maps (PCs derived from TM bands with RF; TM bands with RF; 4–5–7 TM bands with SVM; PCs derived from WIs with SVM; WIs with RF; MNDWI with RF) considering the standard deviation (SD) of the categories by CLC class.

CLC code	CLC class	Average SD by CLC class
512	Water body	0.252
324	Transitional woodland shrub	0.210
222	Fruit trees and berry plantations	0.145
311	Broad-leaved forest	0.138
112	Discontinuous urban fabric	0.133
231	Pastures (grassland)	0.126
243	Land principally occupied by agriculture, with significant areas of natural vegetation	0.120
211	Non-irrigated arable land	0.117
242	Complex cultivation	0.116
411	Inland marsh	0.079

Table 8. Classification accuracies of the Landsat 5 image of 19 June 2006 using different input data (WIs: water indices; TM bands; and PCs of PCA solutions) and classification algorithms. RF: random forest; SVM: support vector machine.

Category	MNDWI (RF)	All WI (RF)	Gao + NDWI + MNDWI (RF)	PCA WIs (RF)	PCA WIs (SVM)	PCA TM bands (RF)	PCA TM bands (SVM)
Dry area	97	99	100	99	100	100	100
Saturated soil	96	96	96	93	91	96	95
Water body	98	100	99	99	100	99	100
OA	97	98	98	97	97	98	98

Table 9. Accuracy of image analysis techniques by the data types involved in previous studies.

Author	Study area	Image	Accuracy	Note
Original bands
Keuchel et al. (2003)	Tenerife, Spain	Landsat	90%	Land-cover detection
Cingolani et al. (2004)	Argentina	Landsat	78%	Vegetation mapping
Atkinson (2004)	–	Simulated images	89–93%
Fichera et al. (2012)	Avellino, Italy	Landsat	86–95%	Land-cover change
Churches et al. (2014)	Haiti	Landsat	78%	Forest detection
Szabó et al. (2014)	Debrecen, Hungary	Airborne hyperspectral data	~85%	Asbestos roof identification
Spectral indices
de Alwis et al. (2007)	Town Brook, USA	Landsat	78–79%	Land-cover detection
Xu (2006)	Xiamen City; Luoyuan Bay; Min River, China	Landsat	99.85%	Water detection
Fisher and Danaher (2013)	New South Wales, Australia	SPOT	98%	LDAWI for water detection
PCA-based
Byrne et al. (1980)	Batemans Bay, New South Wales, Australia	Landsat		Land-cover change detection
Richards (1984)	New South Wales, Australia	Landsat		Change detection (bush fire)
Hirosawa et al. (1996)	Arizona, USA	AVHRR		Vegetation mapping
Munyati (2004)	Kaufe Flats, Zambia	Landsat
Khan et al. (2005)	Indus Basin, Pakistan	IRS LISS-II		Salt-affected soils detection
Koutsias et al. (2009)	Island of Samos, Greece	Landsat	Separability value was used: 1.89	Burnt land mapping, change detection
Forzieri et al. (2011)	Avisio River, Italy	SPOT-5	69.29%, Kappa 0.78	Land-cover extraction
Kit et al. (2012)	Hyderabad, India	QuickBird	Identification probability: 83.33%	Slum detection
Mills et al. (2013)	Southern Rockies–Colorado Plateau, USA	MODIS		Change detection
Dronova et al. (2015)	Poyang Lake, China	Beijing-1; ASAR	94.7% OA, Kappa 0.938	Land-cover mapping

Ashburn, P., 1979. The vegetative index number and crop identification. In: Proceedings of the technical session, Houston, TX, Vols. 1–2. NASA. Johnson Space Center Proc. of Tech., 843–855.

 Google Scholar

Rouse, Jr., J.W., et al., 1974. Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation. In: NASA/GSFC type III final report, Greenbelt, MD, 371. Texas A&M University, Remote Sensing Center.

 Google Scholar

Richardson, A.J. and Wiegand, C.L., 1977. Distinguishing vegetation from soil background information. Photogramnetric Engineering and Remote Sensing, 43, 1541–1552.

 Web of Science ®Google Scholar

Baret, F. and Guyot, G., 1991. Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sensing of Environment, 35, 161–173. doi:10.1016/0034-4257(91)90009-U

 Web of Science ®Google Scholar

Deering, D.W., et al., 1975. Measuring “forage production” of grazing units from Landsat MSS data. In: Proceedings of the 10th International Symposium on Remote Sensing of Environment, Vol. 2, Ann Arbor: ERIM, 1169–1178.

 Google Scholar

Perry Jr., C. and Lautenschlager, L.F., 1984. Functional equivalence of spectral vegetation indices. Remote Sensing of Environment, 14, 169–182. doi:10.1016/0034-4257(84)90013-0

 Web of Science ®Google Scholar

Thiam, A.K., 1997. Geographic information systems and remote sensing methods for assessing and monitoring land degradation in the Sahel: the case of Southern Mauritania. Dissertation. Clark University, Worcester Massachusetts.

 Google Scholar

McFeeters, S.K., 1996. The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing, 17, 1425–1432. doi:10.1080/01431169608948714

 Web of Science ®Google Scholar

Gao, B.-C., 1996. NDWI—a normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58, 257–266. doi:10.1016/S0034-4257(96)00067-3

 Web of Science ®Google Scholar

Xu, H., 2006. Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing, 27, 3025–3033. doi:10.1080/01431160600589179

 Web of Science ®Google Scholar

Keuchel, J., et al., 2003. Automatic land cover analysis for Tenerife by supervised classification using remotely sensed data. Remote Sensing of Environment, 86, 530–541. doi:10.1016/S0034-4257(03)00130-5

 Web of Science ®Google Scholar

Cingolani, A.M., et al., 2004. Mapping vegetation in a heterogeneous mountain rangeland using Landsat data: an alternative method to define and classify land-cover units. Remote Sensing of Environment, 92, 84–97. doi:10.1016/j.rse.2004.05.008

 Web of Science ®Google Scholar

Atkinson, P.M., 2004. Spatially weighted supervised classification for remote sensing. International Journal of Applied Earth Observation and Geoinformation, 5, 277–291. doi:10.1016/j.jag.2004.07.006

 Google Scholar

Fichera, C.R., Modica, G., and Pollino, M., 2012. Land cover classification and change-detection analysis using multi-temporal remote sensed imagery and landscape metrics. European Journal of Remote Sensing, 45, 1–18. doi:10.5721/EuJRS20124501

 Web of Science ®Google Scholar

Churches, C.E., et al., 2014. Evaluation of forest cover estimates for Haiti using supervised classification of Landsat data. International Journal of Applied Earth Observation and Geoinformation, 30, 203–216. doi:10.1016/j.jag.2014.01.020

 Web of Science ®Google Scholar

Szabó, S., et al., 2014. Testing algorithms for the identification of asbestos roofing based on hyperspectral data. Environmental Engineering and Management Journal, 143, 2875–2880.

 Web of Science ®Google Scholar

de Alwis, D.A., et al., 2007. Unsupervised classification of saturated areas using a time series of remotely sensed images. Hydrology and Earth System Sciences, 11, 1609–1620. doi:10.5194/hess-11-1609-2007

 Web of Science ®Google Scholar

Fisher, A. and Danaher, T., 2013. A water index for SPOT5 HRG satellite imagery, New South Wales, Australia, determined by linear discriminant analysis. Remote Sensing, 5 (11), 5907–5925. doi:10.3390/rs5115907

 Web of Science ®Google Scholar

Byrne, G.F., Crapper, P.F., and Mayo, K.K., 1980. Monitoring land-cover change by principal component analysis of multitemporal Landsat data. Remote Sensing of Environment, 10, 175–184. doi:10.1016/0034-4257(80)90021-8

 Web of Science ®Google Scholar

Richards, J.A., 1984. Thematic mapping from multitemporal image data using the principal components transformation. Remote Sensing of Environment, 16, 35–46. doi:10.1016/0034-4257(84)90025-7

 Web of Science ®Google Scholar

Hirosawa, Y., Stuart, E.M., and Kliman, D.H., 1996. Application of standardized principal component analysis to land-cover characterization using multitemporal AVHRR data. Remote Sensing of Environment, 58, 267–281. doi:10.1016/S0034-4257(96)00068-5

 Web of Science ®Google Scholar

Munyati, C., 2004. Use of principal component analysis (PCA) of remote sensing images in wetland change detection on the Kafue Flats, Zambia. Geocarto International, 19, 11–22. doi:10.1080/10106040408542313

 Google Scholar

Khan, N.M., et al., 2005. Assessment of hydrosaline land degradation by using a simple approach of remote sensing indicators. Agricultural Water Management, 77, 96–109. doi:10.1016/j.agwat.2004.09.038

 Web of Science ®Google Scholar

Koutsias, N., Mallinis, G., and Karteris, M., 2009. A forward/backward principal component analysis of Landsat-7 ETM+ data to enhance the spectral signal of burnt surfaces. ISPRS Journal of Photogrammetry and Remote Sensing, 64, 37–46. doi:10.1016/j.isprsjprs.2008.06.004

 Web of Science ®Google Scholar

Forzieri, G., et al., 2011. Satellite multispectral data for improved floodplain roughness modelling. Journal of Hydrology, 407, 41–57. doi:10.1016/j.jhydrol.2011.07.009

 Web of Science ®Google Scholar

Kit, O., Lüdeke, M., and Reckien, D., 2012. Texture-based identification of urban slums in Hyderabad, India using remote sensing data. Applied Geography, 32, 660–667. doi:10.1016/j.apgeog.2011.07.016

 Web of Science ®Google Scholar

Mills, R.T., et al., 2013. Identification and visualization of dominant patterns and anomalies in remotely sensed vegetation phenology using a parallel tool for principal components analysis. Procedia Computer Science, 18, 2396–2405. doi:10.1016/j.procs.2013.05.411

 Google Scholar

Dronova, I., et al., 2015. Mapping dynamic cover types in a large seasonally flooded wetland using extended principal component analysis and object-based classification. Remote Sensing of Environment, 158, 193–206. doi:10.1016/j.rse.2014.10.027

 Web of Science ®Google Scholar