Time-series cloud noise mapping and reduction algorithm for improved vegetation and drought monitoring

Figures & data

Figure 1. Overall methodology of the study. For full colour versions of the figures in this paper, please see the online version.

Table 1. Attribute class rating in MODIS 500-m state flag (16 bit).

*Bit no.	Band no.	*Parameter name	*Bit combination.	Cloud rating code	*Attribute class
0–1	Band 1	Cloud state	00	0	Clear
			11	1	Not set, assumed clear
			10	2	Mixed
			01	3	Cloudy
2	Band 2	Cloud shadow	0	0	No
			1	2	Yes
8–9	Band 3	Cirrus detected	00	0	None
			01	1	Small
			10	2	Average
			11	3	High
10	Band 4	Internal cloud algorithm flag	0	0	No cloud
			1	2	Cloud

*Source: Modis user guide (Vermote, Kotchenova, and Ray 2015)

Figure 2.  Graphical representation of time-series NDVI and CPI values extracted from MODIS 500 m data of 2014: (a) time-series NDVI, (b) time-series CPI, and (c) relationship between time-series NDVI and CPI values.

Table 2. Six cloud noise detection conditions used in the study.

Threshold based condition	Value of TCF	Actual form of masking condition
TBC 1	-
TBC 2	1
TBC 3	0.5
TBC 4	0
TBC 5	−0.5
TBC 6	−1

Figure 3.  Simulated annual NDVI profile having different percentages of random gaps: (a) 10%, (b) 20%, (c) 30%, (d) 40%, and (e) 50%

Figure 4.  Spatial distribution of cloud-related parameters over Sri Lanka (25th composite of h25v08 and h26v08 MODIS tiles).

Figure 5. Graphical example of different threshold level over (a) natural vegetation and (b) agriculture.

Figure 6. Raw NDVI and smoothed NDVI profile (based on different conditions) for (a) tropical dry evergreen forest, (b) tropical moist semi-evergreen forest, (c) tropical moist deciduous forest, (d) tropical dry deciduous forest, (e) rain-fed crop (single growing cycle), (f) intensive irrigated crop (double growing cycle), and (g) intensive irrigated crop (triple growing cycle).

Table 3. Reconstruction model error statistics (RMSE, MAE, MBE, and MAPE) for five gap levels (10%, 20%, 30%, 40%, and 50%).

	Descriptive statistics	% Data gap
		10%	20%	30%	40%	50%
RMSE	Min	0.0023	0.0017	0.0053	0.0053	0.0103
	Max	0.0229	0.0281	0.0273	0.0351	0.0423
	Mean	0.0215	0.0219	0.0227	0.0234	0.0252
	SD	0.0026	0.0029	0.0024	0.0028	0.0037
MAE	Min	0.0001	0.0009	0.0011	0.0016	0.0012
	Max	0.0044	0.0106	0.0098	0.0158	0.0199
	Mean	0.0014	0.0025	0.0035	0.0045	0.0064
	SD	0.0009	0.0013	0.0015	0.0022	0.0031
MBE	Min	−0.0022	−0.0039	−0.0041	−0.0043	−0.0068
	Max	0.0027	0.0069	0.0054	0.0104	0.0149
	Mean	0.0000	−0.0002	0.0001	−0.0001	0.0000
	SD	0.0008	0.0015	0.0019	0.0025	0.0035
MAPE	Min	0.0086	0.1100	0.1459	0.2118	0.1604
	Max	0.6408	1.5738	1.4422	2.3581	3.0005
	Mean	0.1983	0.3539	0.4963	0.6249	0.8878
	SD	0.1294	0.1954	0.2239	0.3297	0.4720

Figure 7. Mean bias error (MBE) distribution and error statistics from simulated annual profiles with (a) 10% gap, (b) 20% gap, (c) 30% gap, (d) 40% gap, and (e) 50% gap.

Figure 8. Spatial distribution of MODIS FCC, raw NDVI and corrected NDVI (based on TBC 3) of different composites of different months: (a) 2nd (2nd week of January), (b) 6th (2nd week of February), (c) 10th (2nd week of March), (d) 14th (2nd week of April), (e) 17th (2nd week of May), (f) 21st (2nd week of June), (g) 25th (2nd week of July), (g) 29th (2nd week of August), (i) 33rd (2nd week of September), (j) 37th (2nd week of October), (k) 40th (2nd week of November), and (l) 44th composite (2nd week of December).

Table 4. Sampling based performance evaluation of different TBCs results in estimating number of growth cycles in natural vegetation.

Result based on different conditions	Heterogeneous sampling			Homogeneous sampling
	Total number of samples	Single growing cycle		Total number of samples	Single growing cycle
		Quantitative estimation	% correctly estimated pixel		Quantitative estimation	% correctly estimated pixel
TBC 1	440	405	93.75	183	180	98.36
TBC 2		408	94.44		176	96.17
TBC 3		412	95.37		176	96.17
TBC 4		386	89.35		172	93.99
TBC 5		349	80.79		160	87.43
TBC 6		318	73.61		139	75.96
Without CPI		228	52.78		107	58.47

Figure 9. The spatial distribution of extracted growing cycle in forested region based on smoothed results from (a) TBC 1, (b) TBC 2, (c) TBC 3, (d) TBC 4, (e) TBC 5, (f) TBC 6, and (g) without CPI.

Figure 10. The spatial distribution of number of growing cycle in agricultural region based on smoothed results from (a) TBC 1, (b) TBC 2, (c) TBC 3, (d) TBC 4, (e) TBC 5, (f) TBC 6, and (g) without CPI.

Table 5. Performance evaluation of different TBCs result based on number of growth cycles in heterogeneous agriculture samples.

Result based on different conditions	Single growing cycle			Double growing cycle			Triple growing cycle			Overall accuracy
	Number of sample pixel		% of correctly estimated pixels	Number of sample pixel		% of correctly estimated pixels	Number of sample pixel		% of correctly estimated pixel	Number of sample pixel		% of correctly estimated pixels
	Observed	Estimated		Observed	Estimated		Observed	Estimated		Observed	Estimated
TBC 1	83	76	91.57	102	58	56.86	15	7	46.67	200	141	70.5
TBC 2		73	87.95		72	70.59		8	53.33		153	76.5
TBC 3		77	92.77		64	62.75		7	46.67		148	74
TBC 4		73	87.95		75	73.53		9	60.00		157	78.5
TBC 5		68	81.93		81	79.41		9	60.00		158	79
TBC 6		71	85.54		89	87.25		15	100.00		175	87.5
Without CPI		34	40.96		76	74.51		15	100.00		125	62.5

Table 6. Performance evaluation of different TBCs result based on number of growth cycles in homogeneous agriculture samples.

Result based on different conditions	Single growing cycle			Double growing cycle			Triple growing cycle			Overall accuracy
	Number of sample pixel		% of correctly estimated pixels	Number of sample pixel		% of correctly estimated pixel	Number of sample pixel		% of correctly estimated pixels	Number of sample pixel		% of correctly estimated pixels
	Observed	Estimated		Observed	Estimated		Observed	Estimated		Observed	Estimated
TBC 1	76	66	86.84	103	46	44.66	7	0	0.00	186	112	60.22
TBC 2		64	84.21		64	62.14		1	14.29		129	69.35
TBC 3		70	92.11		57	55.34		2	28.57		129	69.35
TBC 4		70	92.11		63	61.17		2	28.57		135	72.58
TBC 5		66	86.84		73	70.87		4	57.14		143	76.88
TBC 6		65	85.53		79	76.70		7	100.00		151	81.18
Without CPI		34	44.74		54	52.43		6	85.71		94	50.54

Figure 11. Spatial distribution of cloud corrected VCI and non-cloud corrected VCI over Sri Lanka for (a) 2nd week of July, (b) 3rd week of July, (c) 4th week of July, (d) 1st week of August, (e) 2nd week of August, and (f) 3rd week of August in 2014.

Figure 12. Class-wise % of agriculture area extracted using cloud-corrected VCI and non-cloud-corrected VCI over Sri Lanka for six continuous weeks from 2nd week of July to 3rd week of August, 2014.

Vermote, E. F., S. Y. Kotchenova, and J. P. Ray. 2015. “MODIS Surface Reflectance User’s Guide.” MODIS Land Surface Reflectance Science Computing Facility, Version 1.4: 1-35.

 Google Scholar