1
00:00:00,000 --> 00:01:11,033


2
00:01:11,033 --> 00:01:15,099
Today we will have a Webinar that presents 4 topics.

3
00:01:15,100 --> 00:01:20,400
As a fifth topic, your Questions as typed in the Q&A-box, will be addressed.

4
00:01:20,400 --> 00:01:25,100
First, the basics of mapping Agro-EcoSystems will be presented, 

5
00:01:25,100 --> 00:01:33,266
after which I will show the strengths of using the dimension ‘time’ for mapping and monitoring Agro-EcoSystems.

6
00:01:33,266 --> 00:01:38,532
Then, a series of maps derived from NDVI data-cubes will be presented,

7
00:01:38,533 --> 00:01:46,199
followed by more details on data sources, pre-processing, visualization, and interpretation. 

8
00:01:46,200 --> 00:01:49,466


9
00:01:49,466 --> 00:01:55,366
Section-1. Hyper-Temporal RS gives you a rich data-cube that can be interpreted

10
00:01:55,366 --> 00:01:58,866
to provide information on what is growing where, 

11
00:01:58,866 --> 00:02:05,399
when it is growing, how it is performing, and when/how it is changing.

12
00:02:05,400 --> 00:02:09,000
This information is essential: ...

13
00:02:09,000 --> 00:02:14,666


14
00:02:14,666 --> 00:02:21,366


15
00:02:21,366 --> 00:02:28,266


16
00:02:28,266 --> 00:02:35,432


17
00:02:35,433 --> 00:02:40,099


18
00:02:40,100 --> 00:02:45,466


19
00:02:45,466 --> 00:02:51,299


20
00:02:51,300 --> 00:02:53,033
The Index

21
00:02:53,033 --> 00:03:02,499
NDVI is the most widely used and possibly most successful ‘spectral difference’ index.

22
00:03:02,500 --> 00:03:10,400
Which is the ratio of: the difference of the near infrared and red reflection, over their sum.

23
00:03:10,400 --> 00:03:16,200
We see in the left figure that the green line, which represents vegetation 

24
00:03:16,200 --> 00:03:20,066
[factually, it represents the amount of active chlorophyl present], 

25
00:03:20,066 --> 00:03:28,732
absorbs most light in the visual red band and reflects most light in the 
Near Infra Red [NIR] band.

26
00:03:28,733 --> 00:03:34,799
By comparing both bands, we can thus measure the amount of active chlorophyl.

27
00:03:34,800 --> 00:03:42,966


28
00:03:42,966 --> 00:03:47,032
An NDVI data-cube

29
00:03:47,033 --> 00:03:50,566
A data-cube can be presented as a movie.

30
00:03:50,566 --> 00:03:59,832
The long-term median statistics (or: climatology) of actual (cleaned) NDVI-data are presented in the left figure. 

31
00:03:59,833 --> 00:04:05,199
The movie consists of 36 sequential images, 

32
00:04:05,200 --> 00:04:11,000
each representing a specific dekad, which is a 10-day period.

33
00:04:11,000 --> 00:04:16,633
The figure displays a high and dense amount of vegetation in orange/red, 

34
00:04:16,633 --> 00:04:20,199
and bare-soil (no vegetation) as dark-blue/black. 

35
00:04:20,200 --> 00:04:24,600
In between light to dark-green colours are used. 

36
00:04:24,600 --> 00:04:31,166
The movie clearly shows spatial-temporal changes in vegetation phenology. 

37
00:04:31,166 --> 00:04:37,766
These are caused by climate differences, soil/terrain/landform differences, 

38
00:04:37,766 --> 00:04:41,099
land use differences, etc.

39
00:04:41,100 --> 00:04:47,900
All differences are reflected through land cover (read: NDVI) differences.

40
00:04:47,900 --> 00:04:53,066
Because of these differences, a specific land use, for instance: 

41
00:04:53,066 --> 00:04:56,199
cropping of common wheat, can be mapped.  

42
00:04:56,200 --> 00:05:02,000
In the right figure, an example of such a map is presented. 

43
00:05:02,000 --> 00:05:06,833
The “common-wheat crop-intensity map” is produced on the basis of this movie, 

44
00:05:06,833 --> 00:05:11,133
on available agricultural statistics (by administrative areas), 

45
00:05:11,133 --> 00:05:14,966
on field-observations (the Copernicus-Lucas data), 

46
00:05:14,966 --> 00:05:21,566
and the EU-Corine map (based on a 30m TM-imagery interpretation). 

47
00:05:21,566 --> 00:05:29,799
Thus, satellite imagery and tabular data can be fully integrated and transformed into maps 

48
00:05:29,800 --> 00:05:33,366
containing information of importance.

49
00:05:33,366 --> 00:05:42,999


50
00:05:43,000 --> 00:05:45,466
Let us recapture:

51
00:05:45,466 --> 00:05:49,866
Land Cover (NDVI) relates, amongst others, to:

52
00:05:49,866 --> 00:06:26,599


53
00:06:26,600 --> 00:06:31,100
Since many of the above have a strong temporal behaviour, 

54
00:06:31,100 --> 00:06:34,500
also the NDVI varies over time. 

55
00:06:34,500 --> 00:06:40,600
Often, clear crop growing periods (seasonality) can be recognized 

56
00:06:40,600 --> 00:06:45,966
through plotting NDVI over time; see the top-right figure, 

57
00:06:45,966 --> 00:06:50,532
which displays a specific NDVI-Profile.

58
00:06:50,533 --> 00:06:57,633
Such NDVI-profiles may differ over years because crop-performance will differ 

59
00:06:57,633 --> 00:07:03,266
due to the severity of adverse growing conditions (=perils).

60
00:07:03,266 --> 00:07:08,999
The bottom-right figure displays such annual (seasonal) differences. 

61
00:07:09,000 --> 00:07:16,600
Extreme heat, combined with drought, caused severe crop-failure during the 2010 season; 

62
00:07:16,600 --> 00:07:20,900
2003 however delivered a bumper crop. 

63
00:07:20,900 --> 00:07:30,700
If perils hit a relatively large agricultural area, it often leads to severe price hikes followed by unrest.

64
00:07:30,700 --> 00:07:39,533


65
00:07:39,533 --> 00:07:45,133
Section-2 will zoom-in on differences in behaviour of individual pixels, 

66
00:07:45,133 --> 00:07:49,433
but also of specific areas, over time.

67
00:07:49,433 --> 00:07:55,299
Comparing NDVI-profiles between pixels, between areas, and between seasons (years) 

68
00:07:55,300 --> 00:07:59,733
provides useful information on the actual land cover present, 

69
00:07:59,733 --> 00:08:03,333
of the variability of the vegetation performance, 

70
00:08:03,333 --> 00:08:07,699
and of gradual to abrupt changes in land cover.

71
00:08:07,700 --> 00:08:14,433


72
00:08:14,433 --> 00:08:19,266
Visual inspection of NDVI time-series by pixel

73
00:08:19,266 --> 00:08:27,632
The shown visual inspection of 3 individual pixels clearly shows that land cover of the 3 pixels truly differs.

74
00:08:27,633 --> 00:08:33,933
The shown NDVI-Profiles (DN-values) cover the period 2000-2013

75
00:08:33,933 --> 00:08:37,799
(13 years of MODIS-Terra imagery).

76
00:08:37,800 --> 00:08:46,933
- The red-line displays a continuously high NDVI-amount (DN-values) that indicates an evergreen situation. 

77
00:08:46,933 --> 00:08:52,033
Factually, the pixel is located in the Mau Forest National reserve of Kenya, 

78
00:08:52,033 --> 00:08:58,799
which consists predominantly of a dense forest complex which includes dense bamboo patches.

79
00:08:58,800 --> 00:09:08,966
- The dark-green line displays clear seasonality with 2 peaks annually indicating 2 crop-growing seasons. 

80
00:09:08,966 --> 00:09:15,732
The respective landscape is dominated by arable fields with dispersed some trees

81
00:09:15,733 --> 00:09:20,766
- The light-green line displays a 1-season per year NDVI-profile, 

82
00:09:20,766 --> 00:09:25,466
with relatively low NDVI-values indicating sparse vegetation. 

83
00:09:25,466 --> 00:09:30,166
On the respective Google-Earth image, a bare-landscape is shown; 

84
00:09:30,166 --> 00:09:36,932
likely, seasonal vegetation (short grasses) dominate the specific area.

85
00:09:36,933 --> 00:09:41,799


86
00:09:41,800 --> 00:09:51,566
Each pixel will have its own ‘history’

87
00:09:51,566 --> 00:10:18,899


88
00:10:18,900 --> 00:10:26,200
…and, each pixel will have its own ‘variability’

89
00:10:26,200 --> 00:11:01,266


90
00:11:01,266 --> 00:11:08,632
Each area will have its own ‘variability’

91
00:11:08,633 --> 00:11:14,233
Plotting on top of each-other all 13 annual NDVI-Profiles (DN-values), 

92
00:11:14,233 --> 00:11:19,199
representing whole areas in and around the Masai Mara Park in Kenya, 

93
00:11:19,200 --> 00:11:26,066
displays how reliable the seasonality of the respective areas are.  

94
00:11:26,066 --> 00:11:34,266
Inspect: 
- If there are shifts in timing of seasons causing early or late season starts and/or season ends, and

95
00:11:34,266 --> 00:11:38,966
Inspect:
- If complete absence of seasons do occur.

96
00:11:38,966 --> 00:11:46,066
Where area 48, in general, is dryer and with less vegetation than area 35, 

97
00:11:46,066 --> 00:11:52,899
both areas only confirm that from April to June a clear ‘wet’ season occurs. 

98
00:11:52,900 --> 00:12:03,600
The remainder of the year can remain completely dry or can experience another early or late wet season (with every option in-between).

99
00:12:03,600 --> 00:12:11,966
The shown unreliability in seasonality makes for this area arable agriculture rather difficult; 

100
00:12:11,966 --> 00:12:16,999
good years will be fully mixed with years experiencing severe crop failures. 

101
00:12:17,000 --> 00:12:25,700
Semi-extensive animal husbandry with local trans-migration options is thus the best option to build ones livelihood. 

102
00:12:25,700 --> 00:12:28,800
The Masai thus got it right !

103
00:12:28,800 --> 00:12:34,033


104
00:12:34,033 --> 00:12:38,133
Monitoring an area through its NDVI data

105
00:12:38,133 --> 00:13:17,333


106
00:13:17,333 --> 00:13:25,699
… The drop in cultivated area was caused by a major overhaul of the irrigation infrastructures upstream. 

107
00:13:25,700 --> 00:13:30,400
Thus, the shown ‘drop’ was only temporary.

108
00:13:30,400 --> 00:13:36,366


109
00:13:36,366 --> 00:13:41,099
Monitoring an area through its NDVI data

110
00:13:41,100 --> 00:13:43,200
Another example.

111
00:13:43,200 --> 00:14:10,100


112
00:14:10,100 --> 00:14:23,466
Section-3. The third section shows several mapping-examples, and reviews a series of aspects that relate to creating area-strata from NDVI data-cubes.

113
00:14:23,466 --> 00:14:26,666
The mapping examples cover: ...

114
00:14:26,666 --> 00:14:52,299


115
00:14:52,300 --> 00:14:56,466
Pixels are Mixels

116
00:14:56,466 --> 00:15:03,266
This example relates to wheat & barley systems in Andalusia, Spain.

117
00:15:03,266 --> 00:15:07,666
When an NDVI data-cube is converted to a strata map, 

118
00:15:07,666 --> 00:15:13,399
and if only the key-strata with wheat and barley areas are extracted, 

119
00:15:13,400 --> 00:15:17,633
then we get the shown 7 NDVI-profiles. 

120
00:15:17,633 --> 00:15:24,266
It shows profiles of 5 wheat strata and of 2 barley strata.  

121
00:15:24,266 --> 00:15:31,899
Where barley profiles mainly differ due to the cropping intensity (22 versus 13% in average), 

122
00:15:31,900 --> 00:15:39,666
the wheat strata differ both in wheat cropping intensities (29 to 47%), 

123
00:15:39,666 --> 00:15:45,666
and in the presence of sunflower cropping (22 to 40%).

124
00:15:45,666 --> 00:15:53,599
As an inset, a map of Andalusia, showing the overall wheat cropping intensity (area-fractions) is presented; 

125
00:15:53,600 --> 00:15:59,333
it varies, if present, from 2 to 47%.

126
00:15:59,333 --> 00:16:08,233
A key message of this slide is that preparing a ‘library’ of NDVI-profiles makes no sense. 

127
00:16:08,233 --> 00:16:13,266
Crop mixes vary, responses over time vary, etc. etc. 

128
00:16:13,266 --> 00:16:20,799
A profile will be very-very area-and-season specific and very specific in what mixes it represents.  

129
00:16:20,800 --> 00:16:28,566
We need thus skills to stratify a space-time cube and to relate each created strata with secondary data 

130
00:16:28,566 --> 00:16:31,399
(or as a base to carry out fieldwork) 

131
00:16:31,400 --> 00:16:35,500
in order to make a legend as required.

132
00:16:35,500 --> 00:16:40,133


133
00:16:40,133 --> 00:16:45,033
Mapping at which resolution ?

134
00:16:45,033 --> 00:16:51,966
A general misconception is that use of higher spatial resolution data, produces better mapping opportunities. 

135
00:16:51,966 --> 00:16:59,266
However, please note that the sought-after strata must differentiate land cover & land use patterns 

136
00:16:59,266 --> 00:17:07,932
in the space-time domain that relates best to differentiating different climate, soil, terrain, and landforms.

137
00:17:07,933 --> 00:17:17,533
Differentiating dominant mixtures of Agro-EcoSystems happens to relate better to landscapes when using coarse resolution pixels.

138
00:17:17,533 --> 00:17:28,333
‘Mixels’ at, for instance 100m resolution, will be less ‘generalised’ and will produce an increasing mix of land cover type combinations.

139
00:17:28,333 --> 00:17:36,166
Note also that we mostly analyse NDVI-based data-cubes at country and regional scales. 

140
00:17:36,166 --> 00:17:46,666
Using a too fine spatial resolution of such huge areas will also defeat the effort to create generalizations at the required scale.

141
00:17:46,666 --> 00:17:52,432
As example, for the Mekong area of Vietnam, two stratified maps are presented, 

142
00:17:52,433 --> 00:17:56,799
the left one at 1km resolution (using Spot-VGT) 

143
00:17:56,800 --> 00:18:02,500
and the right one at 250m resolution (using MODIS-Terra.

144
00:18:02,500 --> 00:18:11,433
The 1km map does not suffer from various inclusions (specific mixels) within mapped landscapes 

145
00:18:11,433 --> 00:18:15,499
and provides a better generalization (as required).  

146
00:18:15,500 --> 00:18:19,533
The right map shows a serious increase in specific mixels, 

147
00:18:19,533 --> 00:18:23,833
each with a different unique combination of land cover types. 

148
00:18:23,833 --> 00:18:29,233
As such, linear features like canals and roads, create many of these mixels. 

149
00:18:29,233 --> 00:18:34,533
They clearly interfere with the generalization effort envisaged.

150
00:18:34,533 --> 00:18:39,866


151
00:18:39,866 --> 00:18:46,766
Mapping the gradient between a desert and continental humid areas

152
00:18:46,766 --> 00:18:55,532
Stratifying, for instance an NDVI data-cube of Mongolia, creates a series of strata that follow a beautiful gradient: 

153
00:18:55,533 --> 00:19:02,233
from desert and steppe to continental humid (severe cold winters and warm/hot summers).

154
00:19:02,233 --> 00:19:12,066
Capturing such gradients is simply impossible without the use of an NDVI-based spatial-temporal data-cube.

155
00:19:12,066 --> 00:19:17,132


156
00:19:17,133 --> 00:19:20,699
A down-hill gradient

157
00:19:20,700 --> 00:19:25,533
Another nice example to map a gradient occurs in Java, Indonesia. 

158
00:19:25,533 --> 00:19:29,633
The white arrow points to an area East of Jakarta. 

159
00:19:29,633 --> 00:19:34,399
In the whole area cropping occurs during two seasons annually. 

160
00:19:34,400 --> 00:19:43,933
The timing of each season gets delayed when one moves to the North, or better: when one moves downhill towards the Java Sea. 

161
00:19:43,933 --> 00:19:50,599
This specific gradient is divided into 4 strata, but it could have been divided into more strata, 

162
00:19:50,600 --> 00:19:57,000
each with a ‘narrower’ temporal window regarding the start and end dates of each season.

163
00:19:57,000 --> 00:20:02,500
Exactly, because of the existence of many criss-crossing gradients all over the globe, 

164
00:20:02,500 --> 00:20:06,966
it becomes impossible to specify “how many strata to differentiate”. 

165
00:20:06,966 --> 00:20:14,599
It is as difficult as to state that there just x-shades of grey (between white and black).  

166
00:20:14,600 --> 00:20:22,266
Just consider that you aim to create a map (for monitoring) which must contain a certain level of ‘generalization’. 

167
00:20:22,266 --> 00:20:29,132
Only the expert (you) can decide on the required level (pending on your objectives).

168
00:20:29,133 --> 00:20:35,033


169
00:20:35,033 --> 00:20:40,366
Base-Line Mapping

170
00:20:40,366 --> 00:20:45,999
Base-line mapping, through an NDVI-based space-time data-cube, 

171
00:20:46,000 --> 00:20:48,433
at country or regional level,

172
00:20:48,433 --> 00:20:55,366
is meant to differentiate areas that are clearly different regarding land use and land cover.

173
00:20:55,366 --> 00:21:04,732
The created strata or map-units must be relatively homogeneous within regarding the land cover mixes present, 

174
00:21:04,733 --> 00:21:07,933
and rather heterogeneous between the strata. 

175
00:21:07,933 --> 00:21:12,499
A base-line map can be used, amongst others, for:

176
00:21:12,500 --> 00:21:18,666
- To adopt the strata as a first-level survey sample scheme (i.e. stratified sampling), 

177
00:21:18,666 --> 00:21:30,432
e.g. for Area Frame Sampling (AFS) to collect data that relate to land cover (biodiversity) and/or to land use (agricultural statistics),

178
00:21:30,433 --> 00:21:37,466
- For developing a subsequent monitoring scheme to detect changes in land cover and land use,

179
00:21:37,466 --> 00:21:45,932
- For subsequent mapping individual strata at higher spatial scales through use of high resolution earth observation imagery,

180
00:21:45,933 --> 00:21:59,066
- To use the created NDVI-profiles by strata for subsequent monitoring the performance of those strata as required e.g. for early warning or for insurance schemes

181
00:21:59,066 --> 00:22:04,466


182
00:22:04,466 --> 00:22:11,632
Note that mapped ”forests” of Korea (top-right figure with the green lines) 

183
00:22:11,633 --> 00:22:21,233
differentiates a clear gradient capturing differences in forest-densities (as average for a 1km pixel).

184
00:22:21,233 --> 00:22:30,666
Note also that the ‘dent’ in the green NDVI-profiles reflects the fact that during the ‘long’ growing season of trees,

185
00:22:30,666 --> 00:22:34,966
haze and clouds influenced (dented) the NDVI-readings. 

186
00:22:34,966 --> 00:22:40,999
The ‘dent’ is thus an artifact specific for the forests and the prevailing weather conditions.

187
00:22:41,000 --> 00:22:48,833


188
00:22:48,833 --> 00:22:54,866
A second example of a base-line map concerns Pakistan.

189
00:22:54,866 --> 00:23:00,399
Using 11 years of MODIS NDVI-imagery at 250m resolution,

190
00:23:00,400 --> 00:23:04,566
80 different clusters were created.

191
00:23:04,566 --> 00:23:12,199
Subsequently, the NDVI-profiles of the 80 clusters were grouped into 13 major-groups. 

192
00:23:12,200 --> 00:23:23,633
The map shows that the colour scheme of the Agro-Ecosystems map of Pakistan follows the 13 colours used for plotting the NDVI-profiles (legend).

193
00:23:23,633 --> 00:23:30,699
This map clearly reflects (i) an altitude gradient (up North),

194
00:23:30,700 --> 00:23:36,166
(ii) Irrigated areas versus Barani (rainfed) areas, 

195
00:23:36,166 --> 00:23:41,699
(iii) the performance of the two cropping seasons per year.

196
00:23:41,700 --> 00:23:46,866
From ‘purple’ to ‘red’, to ‘blue’, the performance of the first season was best, 

197
00:23:46,866 --> 00:23:48,799
then both seasons performed equally, 

198
00:23:48,800 --> 00:23:54,566
and more South (downstream) the 2nd season performed best.

199
00:23:54,566 --> 00:24:00,666


200
00:24:00,666 --> 00:24:05,966
The 3rd Base-Line map concerns a specific land use: 

201
00:24:05,966 --> 00:24:09,166
rice-cropping in India and Pakistan.

202
00:24:09,166 --> 00:24:13,232
The rice-systems are differentiated based on 

203
00:24:13,233 --> 00:24:16,266
(i) the water management system, and 

204
00:24:16,266 --> 00:24:20,366
(ii) differences in crop calendars practiced.

205
00:24:20,366 --> 00:24:27,432
The latter differences were nicely captured through the respective area-specific NDVI-profiles.

206
00:24:27,433 --> 00:24:31,466
A small selection of these profiles is shown; 

207
00:24:31,466 --> 00:24:36,832
note specifically the ‘sudden dip’ in NDVI at the start of a growing season 

208
00:24:36,833 --> 00:24:41,566
as caused by flooding of the concerned paddies.

209
00:24:41,566 --> 00:24:47,132


210
00:24:47,133 --> 00:24:54,299
The last base-line mapping example concerns an effort by IWMI-staff, Colombo, 

211
00:24:54,300 --> 00:24:58,366
to map the global cropping areas of five major crops. 

212
00:24:58,366 --> 00:25:04,432
They are: wheat, rice, maize, barley and soybeans. 

213
00:25:04,433 --> 00:25:13,133
Also in this case, the authors made ample use of differentiating NDVI-profiles.

214
00:25:13,133 --> 00:25:22,199
Using the link provided at the bottom of the map will play a video by ESA titled “Changing Lands”. 

215
00:25:22,200 --> 00:25:31,200
This video further illustrates the strength to map actual land cover and to monitor actual land cover

216
00:25:31,200 --> 00:25:36,000
through the use of remotely sensed imagery ‘over time’.

217
00:25:36,000 --> 00:30:22,500


218
00:30:22,500 --> 00:30:29,566
Topic-4. Our 4th topic will present a series of examples to illustrate: 

219
00:30:29,566 --> 00:30:35,799
acquisition, pre-processing, visualization, analysis and interpretation, 

220
00:30:35,800 --> 00:30:41,333
of Hyper-Temporal NDVI-based data-cubes.

221
00:30:41,333 --> 00:30:47,933
But first, let us list some typical characteristics of Hyper-Temporal (HT) NDVI-catalogues.  

222
00:30:47,933 --> 00:30:49,399
They are: ...

223
00:30:49,400 --> 00:31:42,166


224
00:31:42,166 --> 00:31:50,332
Maximum Value Composites (=MVCs)

225
00:31:50,333 --> 00:31:58,999
MVC means: The ‘best’ pixel-specific, daily, NDVI-value of a given period is used 

226
00:31:59,000 --> 00:32:02,433
to represent that specific period.  

227
00:32:02,433 --> 00:32:10,033
This follows the logic that haze and clouds ‘allways’ reduce the actual NDVI-value 

228
00:32:10,033 --> 00:32:13,699
which leads to a lower ‘recorded’ NDVI-value.  

229
00:32:13,700 --> 00:32:18,966
As it happens, the highest recorded NDVI-value in a given period 

230
00:32:18,966 --> 00:32:26,232
thus suffered least from haze and cloud contamination (of course assuming that during the given period,

231
00:32:26,233 --> 00:32:29,499
the NDVI remained relatively static).

232
00:32:29,500 --> 00:32:37,266
Since during a 10-day period, the ‘greenness’ of vegetation is not supposed to fluctuate heavily, 

233
00:32:37,266 --> 00:32:41,766
use of a 10-day MVC-period is very common.  

234
00:32:41,766 --> 00:32:53,199
16-day and monthly composites already (partially) defeat the purpose of creating a Hyper-Temporal NDVI-based data-cube.

235
00:32:53,200 --> 00:33:00,433


236
00:33:00,433 --> 00:33:11,699
The very best 10-daily MVC NDVI-series, as made available by the Copernicus Global Land Service (CGLS), are:

237
00:33:11,700 --> 00:33:23,866
For the period 1 Jan.’1999 – 30 June 2020:   The 1km SPOT-VGT & Proba-V catalogue (version 3; final version!), and

238
00:33:23,866 --> 00:33:32,699
For the period 1 July 2020 – present:   The 300m Sentinel-3 OLCI-based catalogue (version 2).

239
00:33:32,700 --> 00:33:46,700
Both series are fully cross-calibrated (BRDF-adjusted), and carry a guarantee that they will be continued till at least 2040.

240
00:33:46,700 --> 00:33:52,433


241
00:33:52,433 --> 00:33:59,633
Why a BRDF-adjusted NDVI ?

242
00:33:59,633 --> 00:34:06,166
The overpass times of the various platforms used, differed and varied.

243
00:34:06,166 --> 00:34:14,999
See for instance the ‘drifting’ of Proba-V in the figure on: “The evolution of local overpass time at nadir”.

244
00:34:15,000 --> 00:34:24,466
In addition, the viewing angle of the overpassing satellite to the specific ‘pixel’ varies between image rows & columns, 

245
00:34:24,466 --> 00:34:31,299
and pending calendar dates, to make it worse, also the sun follows an annual North-South cycle.

246
00:34:31,300 --> 00:34:39,066
This all impacts on absorption-reflection aspects of VIS and NIR light. 

247
00:34:39,066 --> 00:34:44,799
It also impacts on the shade pattern thrown by objects ‘catching’ the sunlight. 

248
00:34:44,800 --> 00:34:50,966
To make it worse, also the slope and slope-position of the surveyed ‘pixel’ matters.

249
00:34:50,966 --> 00:35:02,199
All the above is adjusted by the data-provider using a Bidirectional_Reflectance_Distribution_Function (BRDF). 

250
00:35:02,200 --> 00:35:16,500
The bottom-right figure shows the results of the BRDF exercise, while the top-right figure shows the considerable mismatch between catalogues when a BRDF-adjustments is omitted.

251
00:35:16,500 --> 00:35:23,533


252
00:35:23,533 --> 00:35:29,199
… more pre-processing by ‘you'

253
00:35:29,200 --> 00:35:37,666
First note that in the CGLS-imagery, all DN-values above 250  are Quality Flags, 

254
00:35:37,666 --> 00:35:44,199
and when used for a pixel, that pixel has thus a ‘missing’ NDVI-value.

255
00:35:44,200 --> 00:35:50,200
Then after the required BRDF-adjustment, and after creating the MVC, 

256
00:35:50,200 --> 00:35:59,000
the resulting NDVI-data can still be influenced by ‘rather persistent’ haze and clouds.  

257
00:35:59,000 --> 00:36:04,966
The brown-line in the shown figure, displays for a specific pixel, 

258
00:36:04,966 --> 00:36:09,432
the NDVI data-series covering 2006. 

259
00:36:09,433 --> 00:36:15,133
Many negative ‘outlyers’ and downward anomalies do occur.

260
00:36:15,133 --> 00:36:25,566
Theoretically, the NDVI pattern, covering the specific pixel and year, would follow the dotted green line. 

261
00:36:25,566 --> 00:36:33,232
That line can partially be estimated through a function called ‘Timesat’ (by Ecklund et.al). 

262
00:36:33,233 --> 00:36:41,766
Timesat is iterative and uses the Savitsky-Golay smoothing algorithm to interpolate between NDVI-readings. 

263
00:36:41,766 --> 00:36:50,099
If the estimated ‘smoothing’ value is higher than the actual value at time-x, then Timesat uses the higher value. 

264
00:36:50,100 --> 00:36:54,800
This process is repeated thrice with a growing temporal window. 

265
00:36:54,800 --> 00:37:00,633
It results in an estimation of the “Upper-Envelope”.  

266
00:37:00,633 --> 00:37:07,866
At the peak of the growing season, the red-line did not fully manage to approach the dotted green line; 

267
00:37:07,866 --> 00:37:16,499
persistent haze and clouds made that impossible, and the Upper-Envelope is thus never completely perfect.

268
00:37:16,500 --> 00:37:22,066


269
00:37:22,066 --> 00:37:29,232
The TIMESAT filter will deliver a ‘clean’ data-cube

270
00:37:29,233 --> 00:37:36,766
Besides filling values for missing records and adjusting values for negative outliers and anomalies, 

271
00:37:36,766 --> 00:37:42,299
Timesat will also remove smaller ‘peaks’; see the figure. 

272
00:37:42,300 --> 00:37:46,533
This removal is achieved through a final iteration, 

273
00:37:46,533 --> 00:37:54,533
in which the actual estimate of the last Savitsky-Golay smoothing is used and not the possible ‘higher’ actual.

274
00:37:54,533 --> 00:38:00,066
Real “Upward” NDVI-anomalies occur only seldomly. 

275
00:38:00,066 --> 00:38:02,132
We call it ‘Glint’.

276
00:38:02,133 --> 00:38:08,266
It happens when the distance between the pixel and the sensor is very high and has a very low angle

277
00:38:08,266 --> 00:38:12,232
(the ‘signal’ has to pass lots of atmosphere); 

278
00:38:12,233 --> 00:38:16,966
in addition, it only occurs when the pixel consists of snow or ice. 

279
00:38:16,966 --> 00:38:24,399
This type of anomaly thus occurs rather rarely, and should not concern us the least.

280
00:38:24,400 --> 00:38:38,166
Once a BRDF-adjusted data-cube of MVC-values has passed the Timesat filter, we can start using that data-cube for our ‘science’.

281
00:38:38,166 --> 00:38:44,866


282
00:38:44,866 --> 00:38:52,666
Once cleaned, do visualize the NDVI-imagery as a movie

283
00:38:52,666 --> 00:39:01,732
Viewing an original cleaned data-cube is of importance, because ‘data-implosion’ through clustering or classification is not yet done. 

284
00:39:01,733 --> 00:39:09,799
Viewing the data-cube as a movie will display and show lots of specifics in which one might be interested.

285
00:39:09,800 --> 00:39:17,166
Do take time to detect and understand features that later must also show in the analysis results.

286
00:39:17,166 --> 00:39:28,199
The movie can be ‘sharpened’ by overlaying a semi-transparent hill-shade image (prepared from a 30m resolution SRTM).

287
00:39:28,200 --> 00:39:36,100
Also add roads, rivers, etc. as required to identify and recognize features of the movie.

288
00:39:36,100 --> 00:39:48,200
Shown is a ‘movie’ of Mozambique prepared through 36 images representing 36 sequential dekads (these are 10-day periods).

289
00:39:48,200 --> 00:40:01,300
The NDVI-value of each pixel of a specific dekad-image represents the median value of all annual repeats of the cleaned NDVI-values for that dekad and pixel.

290
00:40:01,300 --> 00:40:06,066
The shown ‘movie’ thus represents the ‘climatology’ of Mozambique: 

291
00:40:06,066 --> 00:40:14,499
it is a cloudless representative of the average green vegetation present over time covering the past 20 years (2000-2019). 

292
00:40:14,500 --> 00:40:24,466
It is thus the ‘best’ base-line country-level map to study present differences and monitor future changes in land cover and land use.

293
00:40:24,466 --> 00:40:30,999


294
00:40:31,000 --> 00:40:37,466
After one has studied the movie, it is time for a computer-guided ‘data-implosion’. 

295
00:40:37,466 --> 00:40:47,332
The ‘movie’ must become a 2D-map containing map-units, and its legend must become a collection of NDVI-profiles. 

296
00:40:47,333 --> 00:40:53,766
Each NDVI-profile represents one specific map-unit of the prepared map. 

297
00:40:53,766 --> 00:41:01,699
Note that, through the clustering or classification process of a time-series of data by pixel, 

298
00:41:01,700 --> 00:41:08,333
the pixel has received a map-unit code and the pixel-data (its time-series of NDVI-values) 

299
00:41:08,333 --> 00:41:11,066
has shifted to the legend.

300
00:41:11,066 --> 00:41:22,699
Actually: each NDVI-profile in the legend represents the averaged values of all pixels classified to that specific cluster or class.

301
00:41:22,700 --> 00:41:33,200
The produced map is an intermediate-map, ready for use during fieldwork or for integration with secondary data. 

302
00:41:33,200 --> 00:41:39,366
Note that this map can, for 100%, be reproduced at will. 

303
00:41:39,366 --> 00:41:46,699
The only influence one had is by defining the number of clusters that the 2D map must contain. 

304
00:41:46,700 --> 00:41:52,500
That represents the only ‘unknown’ when clustering (classifying) the movie. 

305
00:41:52,500 --> 00:41:58,766
Supervised classification, based a list of a-priori specified land-cover types, 

306
00:41:58,766 --> 00:42:06,832
is rather impossible to carry out, because the ‘coarse’ pixels will often contain complexes of land-cover types, 

307
00:42:06,833 --> 00:42:18,033
and one may easily overlook unique land-cover types that an unsupervised clustering algorithm like ISODATA will properly identify.

308
00:42:18,033 --> 00:42:22,966
Note in the legend the “gradient” of NDVI-profiles. 

309
00:42:22,966 --> 00:42:29,332
Their mutual distances indicates that preparing 50 clusters ‘did the job’.

310
00:42:29,333 --> 00:42:34,899


311
00:42:34,900 --> 00:42:38,466
Another ‘movie'

312
00:42:38,466 --> 00:42:43,166
Shown now is the climatology of Luzon, The Philippines, 

313
00:42:43,166 --> 00:42:47,666
representing the period 1998-2011.

314
00:42:47,666 --> 00:42:53,966
Using the Iterative Self-Organizing Data Analysis Technique (ISODATA), 

315
00:42:53,966 --> 00:42:59,999
an unsupervised clustering of the Hyper-Temporal data-cube is performed.

316
00:43:00,000 --> 00:43:09,566
Using patterns identification and classification, the ‘movie’ is ‘imploded’ to a 2D-map with 79 map-units; 

317
00:43:09,566 --> 00:43:14,566
each map-unit is described through an NDVI-profile.

318
00:43:14,566 --> 00:43:19,766


319
00:43:19,766 --> 00:43:33,332
Shown now are some NDVI-profiles that clearly differ and that can each be labelled using some expert knowledge and field-experience in Luzon.

320
00:43:33,333 --> 00:43:39,866
The red line, for instance, shows 2 distinct growing periods within one year; 

321
00:43:39,866 --> 00:43:42,899
they suggests arable-cropping. 

322
00:43:42,900 --> 00:43:50,366
Indeed, in Luzon, the ‘red’ areas happen to be cropped to irrigated rice twice annually.

323
00:43:50,366 --> 00:43:56,699
The light- and darker-green  profiles show a distinct “dry-period” 

324
00:43:56,700 --> 00:44:03,600
and a rather long (extended) growing season, during which NDVI values are rather high. 

325
00:44:03,600 --> 00:44:11,100
Such findings relate likely to evergreen cover-types with a high density of trees and bushes.

326
00:44:11,100 --> 00:44:16,900
Likely, the orange-line represents something in-between the red and green lines.

327
00:44:16,900 --> 00:44:25,433
It depicts just one growing season. Likely it represents a mixel of rainfed rice and trees and/or bushes.

328
00:44:25,433 --> 00:44:31,366
Verification through high resolution imagery and fieldwork is clearly required.

329
00:44:31,366 --> 00:44:37,199
Gradually, through use of field data, expert knowledge, and secondary data,

330
00:44:37,200 --> 00:44:44,866
the legend of the intermediate map will become a legend that contains factual land cover and land use terminology.

331
00:44:44,866 --> 00:44:49,999
The intermediate NDVI-profiles will thus disappear.

332
00:44:50,000 --> 00:44:55,166


333
00:44:55,166 --> 00:45:01,166
Detecting past changes

334
00:45:01,166 --> 00:45:10,199
In case, annual repeats are NOT generalized using their median values, and thus the movie does not represent ‘climatology’, 

335
00:45:10,200 --> 00:45:19,533
ISODATA clustering will produce a legend with NDVI-profiles covering the full time-period of the original data-cube.

336
00:45:19,533 --> 00:45:25,466
Now, the variability between years is thus NOT removed from the dataset

337
00:45:25,466 --> 00:45:29,566
- This also represents a choice made by the researcher

338
00:45:29,566 --> 00:45:38,599
- Detecting ‘past’ changes or defining ‘past extremes’ will now be possible through studying the produced legend.

339
00:45:38,600 --> 00:45:54,300
As example, shown are NDVI-profiles covering a 10-year period (1998-2008) and representing 3 specific irrigated rice areas in the Mekong area of Vietnam.

340
00:45:54,300 --> 00:46:02,366
- In the blue area the extent of annual flooding in the area increased between 1998 and 2000.

341
00:46:02,366 --> 00:46:10,866
Afterwards, a very steady sequence ensued of flooding and cropping twice a year.

342
00:46:10,866 --> 00:46:18,066
- In the green area, the sequence of cropping and flooding was rather haphazard till early 2002.

343
00:46:18,066 --> 00:46:28,599
Afterwards, annually, 3 crops were cultivated, with no time remaining for any serious flooding.  Till 2007!

344
00:46:28,600 --> 00:46:37,133
Then, famers claimed to have opened the dike that was completed in 2002 and that surrounds their villages,

345
00:46:37,133 --> 00:46:41,599
to ‘harvest’ silt and thus to fertilize their paddies in a very natural way.

346
00:46:41,600 --> 00:46:50,066
In was by intent and not an unforeseen disaster that created a loss of the specific crop.

347
00:46:50,066 --> 00:46:57,799
- The brown line displays a very gradual increase in performance of the rice crop until 2003.

348
00:46:57,800 --> 00:47:03,133
Otherwise, no shocks or abrupt changes did occur.

349
00:47:03,133 --> 00:47:09,066


350
00:47:09,066 --> 00:47:16,299
The last example will be presented through a sequence of slides. 

351
00:47:16,300 --> 00:47:22,633
It concerns mapping rice systems and their cropping calendars, in Odisha, India.

352
00:47:22,633 --> 00:47:32,666
Step-1 shows the climatology movie, which was converted through ISODATA into a map with 100 clusters.

353
00:47:32,666 --> 00:47:45,799
Step-2. Knowing that 100 clusters is rather ‘many’, the 100 NDVI-profiles were grouped into 21 distinct types of profiles.  

354
00:47:45,800 --> 00:47:50,933
This was partly achieved through k-means grouping using the profile-data, 

355
00:47:50,933 --> 00:48:00,966
and partly through ‘expert-interpretation’ considering both the ‘relevant’ parts of the NDVI-Profiles and the locations of related ‘profiles’, 

356
00:48:00,966 --> 00:48:05,966
that is: ‘where’ does a profile occur (and its extent).

357
00:48:05,966 --> 00:48:11,432


358
00:48:11,433 --> 00:48:18,766
This in-between slide shows that ‘expert-guided’ step by showing the spatial relation 

359
00:48:18,766 --> 00:48:24,832
of 4 created NDVI-groups, each having 2 distinct growing seasons:

360
00:48:24,833 --> 00:48:34,666
- Groups [E-E1-I] occur closely together and they form together a gradient (see the profiles and the map)

361
00:48:34,666 --> 00:48:42,632
- Group J (the red area) stands spatially separate and seems unrelated to the others.

362
00:48:42,633 --> 00:48:47,333


363
00:48:47,333 --> 00:48:55,766
Step-3 shows the NDVI-profiles of the 21 groups.

364
00:48:55,766 --> 00:49:06,932
These profiles are used to identify the critical periods within one year that ISODATA used to differentiate the clusters (groups).

365
00:49:06,933 --> 00:49:11,866
Inspection identified 4 distinct periods.

366
00:49:11,866 --> 00:49:17,066
Knowing the 4 periods, a 2nd phase of ‘mapping’ started.

367
00:49:17,066 --> 00:49:23,299


368
00:49:23,300 --> 00:49:28,233
Steps 4 to 6

369
00:49:28,233 --> 00:49:38,033
Step 4.  For the identified 4-periods, NDVI-averages from all available cloud-free medium-resolution imagery, 

370
00:49:38,033 --> 00:49:41,133
covering recent years, were extracted. 

371
00:49:41,133 --> 00:49:46,533
This delivered 4 images that contained mean NDVI-values. 

372
00:49:46,533 --> 00:49:54,199
These 4 images were then used by ISODATA to once more generate 21 classes, 

373
00:49:54,200 --> 00:49:58,200
but now at a 30m spatial resolution.

374
00:49:58,200 --> 00:50:04,200
The super-imposed figure illustrates some results of Step-4.

375
00:50:04,200 --> 00:50:12,166
The 10 profiles shown all have just 4 datapoints each, matching the 4 earlier identified periods.

376
00:50:12,166 --> 00:50:20,266
- Classes 11 & 15 relate to cropping during 2 seasons, where 11 occurs in the lowlands and 15 in the uplands.

377
00:50:20,266 --> 00:50:26,299
- Classes 4, 6, 9 and 10 relate to cropping in the upland areas.

378
00:50:26,300 --> 00:50:32,266
- Classes 5, 7, 8, and 12 relate to cropping  in the lowlands.

379
00:50:32,266 --> 00:50:37,099


380
00:50:37,100 --> 00:50:55,500
Step 5:   The extent of each of the 21 classes was inspected and compared with published  seasonal area statistics on rice by administrative area (Ref. EARAS).

381
00:50:55,500 --> 00:51:02,100
This created an improved legend containing area fractions by NDVI-class.

382
00:51:02,100 --> 00:51:16,066
The shown table contains these area fractions as area-percentages, and does so by season, by water management system (irrigated versus rainfed), and by terrain unit (uplands versus lowlands).

383
00:51:16,066 --> 00:51:25,066
Besides these fractions, also the fractions of the total net area sown annually to any crop is reported.

384
00:51:25,066 --> 00:51:34,599
Please do note that in Odisha, used season-identifiers relate to the harvesting periods of the respective crops.

385
00:51:34,600 --> 00:51:39,366


386
00:51:39,366 --> 00:51:52,832
Step 6: The legend was further improved by matching created legend-items with published crop calendar information.

387
00:51:52,833 --> 00:52:01,466
Note once more that the season-identifiers relate to the harvesting periods of the respective crops:

388
00:52:01,466 --> 00:52:08,899
- Summer rice is harvested in June and has a Crop Growing Period Length of 120 to 135 days,

389
00:52:08,900 --> 00:52:18,833
- Autumn rice is harvested in September, with a relatively short growing period of 80 to 100 days, and

390
00:52:18,833 --> 00:52:27,466
- Winter rice is harvested from October to December, with a growing period that can extend from 100 to 150 days.

391
00:52:27,466 --> 00:52:34,566
Finally do note, that I as author of this map, never set foot in Odisha. 

392
00:52:34,566 --> 00:52:45,566
Agronomic expert knowledge, great secondary data sources, and smart combination of NDVI-data in the spatial and temporal domains, did the job !

393
00:52:45,566 --> 00:52:52,599


394
00:52:52,600 --> 00:52:57,000
Section-5. This brings us to the end of the presentation. 

395
00:52:57,000 --> 00:53:02,900
Now it is time to address Questions posed by you in the Q&A chat-box.

396
00:53:02,900 --> 00:53:07,666
As a repeat, the topics covered are repeated on this slide.

397
00:53:07,666 --> 00:53:53,399