Assessment of values and trends in coarse spatial resolution NDVI datasets in Southeast Asia landscapes

Figures & data

Table 1. NDVI applications used in this study with spatial and temporal resolution and period of data availability.

Data	Source	Spatial resolution	Temporal resolution	Data archive
GIMMS3G ver. 0	ECOCAST NASA	8 km	Bi-monthly	Aug 1981–Dec 2013
STAR AVHRR	NESDIS	4 km	Weekly	Aug 1981–Present*
STAR VIIRS	NESDIS	4 km	Weekly	Nov 2012–Present*
SPOT VGT	VGT VITO	1 km	10 days	Jan 1999–May 2014
METOP	VGT VITO	1 km	10 days	Apr 2007–Present*
MERIS	VGT VITO	1 km	10 days	May 2002–March 2012
TERRA (MOD13A3) ver. 6	LPDAAC	1 km	Monthly	Feb 2000–Present*
AQUA (MYD13A3) ver. 6	LPDAAC	1 km	Monthly	July 2002–Present*
Proba – V	VGT VITO	1 km	10 days	Nov 2013–Present*

*Data as current as February, 2017

Figure 1. Study area includes all of Indonesia and parts of other South East Asian countries.

Table 2. Type 3 classification scheme of MOD12Q1 and the associated notation for each classes in this study.

Category	Notation	Category	Notation
Grasses/Cereal crops	Grass/Cer	Deciduous Broadleaf Forests	Deciduous BF
Shrubs	Shrubs	Evergreen Needleleaf Forests	Evergreen NF
Broadleaf Crops	Broadleaf Cr	Deciduous Needleleaf Forests	Deciduous NF
Savanna	Savanna	Non-Vegetated	Non-Veg
Evergreen Broadleaf Forests	Evergreen BF	Urban	Urban

Table 3. Trend map categories derived using the Mann–Kendall test.

Trend’s type	Slope (S)	P-value	Trend map categories
Significant negative	S < 0	p ≤ 0.05	SN
Non-significant negative		0.05 < p ≤ 0.1	NS
No trend with negative tendency		p > 0.1	NT
No trend with positive tendency	S > 0	p > 0.1
Non-significant positive		0.05 < p ≤ 0.1	NS
Significant positive		p ≤ 0.05	SP

Table 4. Matrix to calculate the percentage of similarity.

	Categories	Trend Map A
		SN	NS	NT	SP
Trend map	SN	A	B	C	D
	NS	E	F	G	H
	NT	I	J	K	L
	SP	M	N	O	P

Figure 2. Distribution of average NDVI values from time series Landsat 5 TM and 8 OLI used for validating NDVI from coarse resolution data.

Figure 3. Scatter plot of monthly average and maximum NDVI values at Pasoh Rainforest, Malaysia.

Table 5. NDVI data used in the regression analysis.

Data	Time range	Original data observation	Processing	Number of data
Landsat 5 TM	April 2007–March 2012	16 days	1.) Averaging the pixel values of more than 3 Landsat imageries with continuous time acquisition within the time range and 2.) resampling to 8-km (2007–2013) and 4-km (2014–2015)	255 scenes averaged into 46 imageries
Landsat 8 OLI	January 2014–December 2015			472 scenes averaged into 73 imageries
Flux Tower data	October 2008–2015	30-minutes data during the non-rainy day	Averaging the 30 min data to represent monthly data	1.) 40 months (2008–2012), and 2.) 24 months (2014–2015)

Table 6. Correlation coefficients for pairwise comparison of different NDVI datasets at spatial resolutions of 1, 4 and 8 km during April 2007 to March 2012 (60 months) and January 2014 to December 2015 (24 months).

Pixel Size	Period	April 2007–Dec 2012							Jan 2014–Dec 2015
	Data	MA	MRS	MTP	MT	SV			Data	MA	MTP	PV	MTP
1 km	MA	100							MA	100
	MRS	36.64	100						MTP	35.19	100
	MTP	36.05	36.01	100					PV	36.05	37.98	100
	MT	39.96	36.68	37.44	100				MT	35.81	34.94	35.92	100
	SV	36.21	34.71	40.94	38.52	100
	Data	MA	MRS	MTP	MT	SV	AN		Data	AN	VN	MA	MTP	PV	MT
4 km	MA	100							AN	100
	MRS	37.77	100						VN	38.23	100
	MTP	36.62	35.61	100					MA	34.84	36.13	100
	MT	42.88	35.62	39.25	100				MTP	35.52	35.74	35.22	100
	SV	37.49	36.42	44.12	42.25	100			PV	35.99	37.31	37.42	40.25	100
	AN	37.04	34.76	36.05	36.26	36.17	100		MT	33.37	34.75	38.04	35.37	38.01	100
	Data	GM	MA	MRS	MTP	AN	MT	SV	Legend
8 km	GM	100							GM : GIMMS3G			AN : AVHRR NESDIS
	MA	37.45	100						PV : PROBA V			VN : VIIRS NESDIS
	MRS	35.12	38.05	100					MRS : MERIS			MT : MODIS TERRA
	MTP	34.23	36.06	35.30	100				MTP : METOP			MA : MODIS AQUA
	AN	35.31	37.84	35.12	35.90	100			SV : SPOT VGT
	MT	36.10	42.99	38.71	39.59	37.08	100			Data with relatively low percentage
	SV	33.13	36.56	36.30	46.44	36.89	42.30	100

Figure 4. Average correlation of coefficient map of NDVI products estimated from the pairwise comparison between existing NDVI products.

Figure 5. Boxplot showing the statistical properties of correlation coefficient for different land cover types at different spatial resolutions of 1, 4 and 8 km.

Table 7. Percentage of spatial similarities in trend maps in vegetation areas estimated from different NDVI time series applications at spatial resolution of 1, 4 and 8 km.

	Period	April 2007–Dec 2012							Jan 2014–Dec 2015
Pixel Size	Data	MA	MRS	MTP	MT	SV			Data	MA	MTP	PV	MTP
1 km	MA	100							MA	100
	MRS	36.64	100						MTP	35.19	100
	MTP	36.05	36.01	100					PV	36.05	37.98	100
	MT	39.96	36.68	37.44	100				MT	35.81	34.94	35.92	100
	SV	36.21	34.71	40.94	38.52	100
	Data	MA	MRS	MTP	MT	SV	AN		Data	AN	VN	MA	MTP	PV	MT
4 km	MA	100							AN	100
	MRS	37.77	100						VN	38.23	100
	MTP	36.62	35.61	100					MA	34.84	36.13	100
	MT	42.88	35.62	39.25	100				MTP	35.52	35.74	35.22	100
	SV	37.49	36.42	44.12	42.25	100			PV	35.99	37.31	37.42	40.25	100
	AN	37.04	34.76	36.05	36.26	36.17	100		MT	33.37	34.75	38.04	35.37	38.01	100
	Data	GM	MA	MRS	MTP	AN	MT	SV	Legend
8 km	GM	100							GM : GIMMS3G			AN : AVHRR NESDIS
	MA	37.45	100						PV : PROBA V			VN : VIIRS NESDIS
	MRS	35.12	38.05	100					MRS : MERIS			MT : MODIS TERRA
	MTP	34.23	36.06	35.30	100				MTP : METOP			MA : MODIS AQUA
	AN	35.31	37.84	35.12	35.90	100			SV : SPOT VGT
	MT	36.10	42.99	38.71	39.59	37.08	100			Data with generally smaller percentages
	SV	33.13	36.56	36.30	46.44	36.89	42.30	100

Figure 6. Greening and browning trends over 60 months mapped using different NDVI time series applications at different spatial resolutions.

Table 8. Matrix of average distribution of spatial similarities in vegetation areas from trend maps derived using different NDVI applications.

	Categories	SP	SN	NS	NT
1 k	SP	6.21 ± 3.6	2.65 ± 1.3	1.25 ± 0.4	5.31 ± 1.8
	SN	8.2 ± 3.1	12.46 ± 2.5	3.95 ± 0.3	19.86 ± 0.8
	NS	1.76 ± 0.6	2.42 ± 0.7	0.79 ± 0.1	3.76 ± 0.4
	NT	4.72 ± 1.9	9.96 ± 2.1	2.76 ± 0.5	13.94 ± 1.6
4 km	SP	6.88 ± 4	3.06 ± 1.9	1.46 ± 0.6	6.04 ± 2.7
	SN	7.9 ± 3.1	12.58 ± 2.5	3.96 ± 0.3	19.75 ± 1
	NS	1.78 ± 0.6	2.51 ± 0.7	0.81 ± 0.1	3.78 ± 0.4
	NT	3.98 ± 1.8	9.79 ± 2.5	2.6 ± 0.5	13.11 ± 2
8 km	SP	8.89 ± 4.1	2.19 ± 1.5	1.43 ± 0.6	5.53 ± 2.5
	SN	10.28 ± 3.5	11.04 ± 2.7	3.89 ± 0.3	19.85 ± 1.5
	NS	2.18 ± 0.7	1.92 ± 0.5	0.72 ± 0	3.51 ± 0.3
	NT	5.09 ± 2.3	8 ± 1.7	2.4 ± 0.4	13.09 ± 1.7
		: percentage of same trends at both trend maps
		: percentage of contrasting significant trends

Table 9. Similarity percentages of breakpoints detected from coarse resolution NDVI products during a) 2007–2012, and b) 2014–2015.

a) April 2007–Dec 2012								b) Jan 2014–Dec 2015
Data	GM	MA	MRS	MTP	AN	MT	SV	Data	MA	MTP	PV	MT	VN	AN
GM	100							MA	100
MA	29.58	100						MTP	57.50	100
MRS	29.63	32.35	100					PV	56.62	57.98	100
MTP	29.30	32.87	33.74	100				MT	55.13	55.21	55.87	100
AN	29.12	34.14	32.19	32.29	100			VN	52.54	54.12	53.65	52.20	100
MT	29.72	37.97	34.86	35.25	34.22	100		AN	51.32	51.70	52.12	51.10	58.42	100
SV	29.30	35.74	35.63	38.66	35.66	40.53	100		Data with relatively smaller percentages

Figure 7. NDVI values for the vegetation cover taken from a.) SPOT-VGT (during 2007 – 2012) and b.) MODIS Terra (during 2014 – 2015) displayed a strong fit with NDVI from Landsat 5 TM and Landsat 8 OLI.

Table 10. Regression results from NDVI coarse resolution products versus Landsat systems.

a) Regression with Landsat 5 TM during 2007–2012 in 8-km pixel size
Category	GIMMS			Modis Terra			Modis Aqua			SPOT VGT			METOP			MERIS			NESDIS
	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n
All pixels	0.12	0.35	13,055	0.1	0.54	13,055	0.1	0.52	13,055	0.09	0.64	13,055	0.1	0.53	13,055	0.11	0.45	13,055	0.12	0.31	5710
Veg. only pixels	0.09	0.32	11,747	0.07	0.58	11,745	0.07	0.53	11,746	0.06	0.68	11,746	0.08	0.51	11,746	0.09	0.34	11,746	0.12	0.28	5181
CRO	0.07	0.05	215	0.05	0.61	214	0.05	0.55	216	0.05	0.59	216	0.05	0.52	216	0.06	0.43	216	0.05	0.46	48
SAV	0.12	0.31	1642	0.1	0.54	1638	0.11	0.48	1638	0.07	0.76	1640	0.1	0.53	1639	0.11	0.4	1640	0.1	0.6	1430
EBF	0.06	0.09	9693	0.05	0.25	9696	0.06	0.18	9695	0.05	0.34	9692	0.05	0.25	9693	0.06	0.11	9692	0.06	0.01	3644
b) regression with Landsat 8 OLI during 2014–2015 in 4-km pixel size
	Modis Terra			Modis Aqua			METOP			Proba-V			VIIRS			NESDIS
Category	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n	RMSE	r2	n
All pixels	0.12	0.59	56,566	0.13	0.55	56,566	0.13	0.54	56,551	0.11	0.64	59,127	0.15	0.19	51,330	0.13	0.1	45,305
Veg. only pixels	0.07	0.67	46,999	0.09	0.54	47,012	0.09	0.49	47,010	0.07	0.67	49,582	0.11	0.2	44,409	0.12	0.06	43,047
CRO	0.06	0.71	1271	0.07	0.65	1272	0.07	0.62	1276	0.07	0.67	1271	0.09	0.44	1259	0.09	0.33	1245
SAV	0.07	0.62	10,859	0.07	0.59	10,865	0.09	0.45	10,859	0.07	0.6	10,877	0.09	0.46	10,740	0.09	0.35	10,520
EBF	0.06	0.26	32,830	0.07	0.11	32,841	0.06	0.2	32,836	0.06	0.31	35,400	0.07	0.01	30,382	0.08	0	29,267

Table 11. Linear regression and corresponding R-squared and RMSE derived from comparison between NDVI values from satellite data and flux tower datasets.

Data (n)	Regression (average)	R^2	RMSE	Regression (max)	R^2	RMSE
GIMMS3G Ver.0 (40)	Y = 0.81–0.03x	0.02	0.06	Y = 0.83–0.02x	0.01	0.06
AVHRR NESDIS (40)	Y = 0.8–0.05x	0.03	0.41	Y = 0.81 + 0.01x	0	0.44
AVHRR VIIRS (24)	Y = 0.69 + 0.13x	0.18	0.16	Y = 0.76 + 0.08x	0.08	0.2
SPOT VGT (40)	Y = 0.85–0.08x	0.03	0.06	Y = 0.9–0.1x	0.08	0.04
METOP (40)	Y = 0.75 + 0.04x	0.02	0.07	Y = 0.81 + 0.01x	0	0.05
MERIS (40)	Y = 0.76 + 0.02x	0.01	0.01	Y = 0.82–0.01x	0	0.01
MODIS TERRA (40)	Y = 0.84–0.07x	0.03	0.07	Y = 0.87–0.06x	0.04	0.04
MODIS AQUA (40)	Y = 0.79–0.01x	0	0.09	Y = 0.9–0.1x	0.06	0.05
PROBA V (24)	Y = 0.8–0.04x	0.05	0.11	Y = 0.84–0.04x	0.04	0.1