Optimal season for discriminating C3 and C4 grass functional types using multi-date Sentinel 2 data

Figures & data

Figure 1. The location of the study area: (a) in KwaZulu-Natal Province, South Africa and (b) its overview from Sentinel 2 natural band composition (bands 4, 3 and 2), the spatial distribution of the sampled points and the location of a weather station, from which some of the data used were recorded.

Figure 2. Flowchart showing the methodological procedure followed in the study. The highlighted boxes indicated the stages at which statistical tests were performed.

Table 1. Sentinel 2A image data characterization.

Band name (−)	Band	Central λ (nm)	GSD (m)
Blue	2	490	10
Green	3	560
Red	4	665
NIR	8	842
RE	5	705	20
RE	6	740
RE	7	783
RE	8a	865
SWIR	11	1610
SWIR	12	2190
Coastal aerosol	1	443	60
Water vapour	9	945
Cirrus	10	1380

Table 2. Sentinel 2A image acquisition date (Julian day format) in 2016 and their characteristics.

Season	Acquisition date	Sun zenith angle (°)	Sun azimuth angle (°)
Summer	38	41.57	44.02
	64	46.94	36.32
	307	24.59	60.95
	337	22.41	77.97
Winter	147	55.99	28.94
	177	58.34	29.29
	197	47.16	37.1
	237	36.49	43.65

Table 3. Test of significant difference results in spectral response between the two species over time.

Season	Julian day	B	G	R	RE	RE	RE	NIR	RE	SWIR 1	SWIR 2
Summer	038	0.97	0.40	0.16	0.31	0.01	0.01	0.05	0.01	0.29	0.01
	064	0.79	0.32	0.55	0.04	0.01	0.01	0.02	0.09	0.11	0.01
	307	0.34	0.01	0.01	0.01	0.01	0.01	0.01	0.11	0.44	0.01
	337	0.04	0.21	0.04	0.01	0.01	0.48	0.01	0.14	0.04	0.08
Winter	147	0.06	0.39	0.18	0.02	0.02	0.01	0.74	0.23	0.01	0.01
	177	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01
	197	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.03
	237	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01

p values in bold and italics indicate no significant differences (α = 0.05) in spectral response between the two species using that particular band.

Figure 3. Species spectral response curves for summer (a-d) and winter e-h. Red boxes show the bands with separable spectral response.

Figure 4. Derived: (a) overall accuracies, (b) classification errors and (c) species areal coverage in relation to monthly rainfall and temperature.

Figure 5. Spectral bands frequency in discriminating C3 and C4 over time.

Figure 6. The spatial distribution of C3 and C4 grass species across the study area using images acquired at different seasonal periods in 2016.