ChapterPDF Available

Global Cropland Area Database (GCAD) derived from Remote Sensing in Support of Food Security in the Twenty-first Century: Current Achievements and Future Possibilities

Authors:

Abstract and Figures

Abstract The precise estimation of the global agricultural cropland-extents, areas, geographic locations, crop types, cropping intensities, and their watering methods (irrigated or rainfed; type of irrigation) provides a critical scientific basis for the development of water and food security policies (Thenkabail et al., 2012, 2011, 2010). By year 2100, the global human population is expected to grow to 10.4 billion under median fertility variants or higher under constant or higher fertility variants (Table 1) with over three quarters living in ...
Content may be subject to copyright.
1
Remote Sensing Handbook:
1
Land Resources: Monitoring, Modelling, and Mapping
2
Volume II, Chapter 6
3
Global Food Security Support Analysis Data (GFSAD) at Nominal
4
1-km (GCAD) derived from Remote Sensing in Support of Food
5
Security in the Twenty-first Century: Current Achievements and
6
Future Possibilities
7
8
Pardhasaradhi Teluguntla
1,2
, Prasad S. Thenkabail
1
, Jun Xiong
1,3
,
9
Murali Krishna Gumma
4
, Chandra Giri
5
, Cristina Milesi
6
, Mutlu Ozdogan
7
,
10
Russell G. Congalton
8
, James Tilton
9
, Temuulen Tsagaan Sankey
3
, Richard Massey
3
,
11
Aparna Phalke
7
, and Kamini Yadav
8
12
1 = U. S. Geological Survey (USGS), 2255, N. Gemini Drive, Flagstaff, AZ 86001, USA
13
2 = Bay Area Environmental Research Institute (BAERI), 596 1st St West Sonoma, CA 95476, USA
14
3 = School of Earth Sciences and Environmental Sustainability (SESES), Northern Arizona University, Flagstaff,
15
AZ 86011,USA
16
4 = International Crops Research Institute for the Semi Arid Tropics (ICRISAT), Patancheru, Hyderabad, India
17
5 = U. S. Geological Survey (USGS), (EROS) Center, Sioux Falls, SD, USA
18
6 = NASA Ames Research Center, MS 242-4, Moffett Field, CA 94035, USA
19
7 = University Of Wisconsin, 1710 University Avenue, Madison, WI 53726, USA
20
8 = University of New Hampshire, 215 James Hall, 56 College Road, Durham, NH 03824, USA
21
9 = NASA Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA
22
23
Email: pteluguntla@usgs.gov, pthenkabail@usgs.gov, jxiong@usgs.gov, m.gumma@cgiar.org, cgiri@usgs.gov
24
cristina.milesi-1@nasa.gov, ozdogan@wisc.edu, russ.congalton@unh.edu, james.c.tilton@nasa.gov,
25
Temuulen.Sankey@nau.edu, rmassey@usgs.gov, phalke@wisc.edu, kaminiyadav.02@gmail.com
26
27
6.0 Introduction
28
6.1 Global distribution of croplands and other land use and land cover: Baseline
29
6.1.1 Existing global cropland maps: Remote sensing and non-remote sensing approaches
30
6.2 Key remote sensing derived cropland products: in support of global food security
31
6.3 Definition of cropland mapping using remote sensing
32
6.4Data: Remote sensing and other data for global cropland mapping
33
6.4.1 Primary satellite sensor data
34
6.4.2 Secondary data
35
6.4.3 Field-plot Data
36
6.4.4 Very high resolution imagery data
37
6.4.5 Data composition: Mega File Data Cube (MFDC) concept
38
6.5 Methods of cropland mapping
39
6.5.1 Cropland mapping methods using remote sensing at global, regional, and local
40
scales
41
6.5.2 Spectral Matching Techniques (SMTs) Algorithms
42
6.5.2.1 Generating Class Spectra
43
6.5.2.2 Ideal Spectra Data Bank on Irrigated Areas (ISDB IA)
44
6.6 Automated Cropland Classification Algorithm (ACCA)
45
6.7 Remote sensing based global cropland products: current state-of-art, their strengths, and
46
limitations
47
2
6.7.1 Global cropland extent at nominal 1-km resolution6.8 Change Analysis
1
6.9 Uncertainties of existing cropland products
2
6.10 Way forward
3
6.11 Conclusions
4
6.12 Acknowledgements
5
6.13 References
6
7
8
3
6.0 Introduction
The precise estimation of the global agricultural cropland- extents, areas, geographic locations,
crop types, cropping intensities, and their watering methods (irrigated or rainfed; type of
irrigation) provides a critical scientific basis for the development of water and food security
policies (Thenkabail et al., 2012, 2011, 2010). By year 2100, the global human population is
expected to grow to 10.4 billion under median fertility variants or higher under constant or
higher fertility variants (Table 6.1) with over three quarters living in developing countries and in
regions that already lack the capacity to produce enough food. With current agricultural
practices, the increased demand for food and nutrition would require about 2 billion hectares of
additional cropland, about twice the equivalent to the land area of the United States, and lead to
significant increases in greenhouse gas productions associated with agricultural practices and
activities (Tillman et al., 2011). For example, during 1960-2010, world population more than
doubled from 3 billion to 7 billion. The nutritional demand of the population also grew swiftly
during this period from an average of about 2000 calories per day per person in 1960 to nearly
3000 calories per day per person in 2010. The food demand of increased population along with
increased nutritional demand during this period was met by the green revolution which more
than tripled the food production; even though croplands decreased from about 0.43 ha per capita
to 0.26 ha per capita (FAO, 2009). The increase in food production during the green revolution
was the result of factors such as: (a) expansion of irrigated croplands, which had increased in
2000 from 130 Mha in the 1960s to between 278 Mha (Siebert et al., 2006) and 467 Mha
(Thenkabail et al., 2009a, 2009b, 2009c), with the larger estimate due to consideration of
cropping intensity; (b) increase in yield and per capita production of food (e.g., cereal production
from 280 kg/person to 380 kg/person and meat from 22 kg/person to 34 kg/person (McIntyre,
2008); (c) new cultivar types (e.g., hybrid varieties of wheat and rice, biotechnology); and (d)
modern agronomic and crop management practices (e.g., fertilizers, herbicide, pesticide
applications).
Although modern agriculture met the challenge to increase food production last century,
lessons learned from the 20
th
century “green revolution” and our current circumstances impact
the likelihood of another such revolution. The intensive use of chemicals have adversely
impacted the environment in many regions, leading to salinization and decreasing water quality
and degrading croplands. From 1960 to 2000, worldwide phosphorous use doubled from 10
million tons (MT) to 20 MT, pesticide use tripled from near zero to 3 MT, and nitrogen use as
fertilizer increased to a staggering 80 MT from just 10 MT (Foley et al., 2007; Khan and Hanjra,
2008). Diversion of croplands to bio-fuels is taking water away from food production (Bindraban
et al., 2009), even as the economic, carbon sequestration, environmental, and food security
impacts of biofuel production are proving to be a net negative (Lal and Pimentel, 2009; Gibbs et
al., 2008; Searchinger et al., 2008). Climate models predict that the hottest seasons on record will
become the norm by the end of the century in most regions of the world - a prediction that bodes
ill for feeding the world (Kumar and Singh, 2005). Increasing per capita meat consumption is
increasing agricultural demands on land and water (Vinnari and Tapio, 2009). Cropland areas are
decreasing in many parts of the World due to urbanization, industrialization, and salinization
(Khan and Hanjra, 2008). Ecological and environmental imperatives, such as biodiversity
conservation and atmospheric carbon sequestration, have put a cap on the possible expansion of
cropland areas to other lands such as forests and rangelands (Gordon et al., 2009). Crop yield
increases of the green revolution era have now stagnated (Hossain et al., 2005). Given these
4
factors and limitations, further increase in food production through increase in cropland areas
and\or increased allocations of water for croplands are widely considered unsustainable or
simply infeasible.
Clearly, our continued ability to sustain adequate global food production and achieve
future food security in the twenty-first century is challenged. So, how does the World continue to
meet its food and nutrition needs? Solutions may come from bio-technology and precision
farming. However, developments in these fields are not currently moving at rates that will ensure
global food security over the next few decades (Foley et al., 2011). Further, there is a need for
careful consideration of possible adverse effects of bio-technology. We should not be looking
back 3050 years from now with regrets, like we are looking back now at many mistakes made
during the green revolution. During the green revolution, the focus was only on getting more
yield per unit area. Little thought was given to the serious damage done to our natural
environments, water resources, and human health as a result of detrimental factors such as
uncontrolled use of herbicides, pesticides, and nutrients, drastic groundwater mining, and
salinization of fertile soils due to over-irrigation. Currently, there are discussions of a “second
green revolution” or even an “ever green revolution”, but definitions of what these terms actually
mean are still debated and are evolving (e.g., Monfreda et al., 2008). One of the biggest issues
that has not been given adequate focus is the use of large quantities of water for food production.
Indeed, an overwhelming proportion (60-90%) of all human water use in India, for example,
goes for producing their food (Falkenmark, M., & Rockström, 2006). But such intensive water
use for food production is no longer sustainable due to increasing competition for water in
alternative uses, such as urbanization, industrialization, environmental flows, bio-fuels, and
recreation. This has brought into sharp focus the need to grow more food per drop of water
leading to the need for a “blue revolution” in agriculture (Pennisi, E., 2008).
Table 6.1. World population (thousands) under all variants, 1950-2100.
Year
Medium
fertility variant
High
fertility variant
Low
fertility variant
Constant
fertility variant
1950
2,529,346
2,529,346
2,529,346
2,529,346
1955
2,763,453
2,763,453
2,763,453
2,763,453
1960
3,023,358
3,023,358
3,023,358
3,023,358
1965
3,331,670
3,331,670
3,331,670
3,331,670
1970
3,685,777
3,685,777
3,685,777
3,685,777
1975
4,061,317
4,061,317
4,061,317
4,061,317
1980
4,437,609
4,437,609
4,437,609
4,437,609
1985
4,846,247
4,846,247
4,846,247
4,846,247
1990
5,290,452
5,290,452
5,290,452
5,290,452
1995
5,713,073
5,713,073
5,713,073
5,713,073
2000
6,115,367
6,115,367
6,115,367
6,115,367
2005
6,512,276
6,512,276
6,512,276
6,512,276
5
2010
6,916,183
6,916,183
6,916,183
6,916,183
2015
7,324,782
7,392,233
7,256,925
7,353,522
2020
7,716,749
7,893,904
7,539,163
7,809,497
2025
8,083,413
8,398,226
7,768,450
8,273,410
2030
8,424,937
8,881,519
7,969,407
8,750,296
2035
8,743,447
9,359,400
8,135,087
9,255,828
2040
9,038,687
9,847,909
8,255,351
9,806,383
2045
9,308,438
10,352,435
8,323,978
10,413,537
2050
9,550,945
10,868,444
8,341,706
11,089,178
2055
9,766,475
11,388,551
8,314,597
11,852,474
2060
9,957,399
11,911,465
8,248,967
12,729,809
2065
10,127,007
12,442,757
8,149,085
13,752,494
2070
10,277,339
12,989,484
8,016,514
14,953,882
2075
10,305,146
13,101,094
7,986,122
15,218,723
2080
10,332,223
13,213,515
7,954,481
15,492,520
2085
10,358,578
13,326,745
7,921,618
15,775,624
2090
10,384,216
13,440,773
7,887,560
16,068,398
2095
10,409,149
13,555,593
7,852,342
16,371,225
2100
10,433,385
13,671,202
7,815,996
16,684,501
Source: UNDP (2012).
A significant part of the solution lies in determining how global croplands are currently used and
how they might be better managed to optimize use of resources in food production. This will
require development of an advanced Global Cropland Area Database (GCAD) with an ability to
map global croplands and their attributes routinely, rapidly, consistently, and with sufficient
accuracies. This in turn requires the creation of a framework of best practices for cropland
mapping and an advanced global geospatial information system on global croplands. Such a
system would need to be consistent across nations and regions by providing information on
issues such as the composition and location of cropping, cropping intensities (e.g. single, double
crop), rotations, crop health/vigor, and irrigation status. Opportunities to establish such a global
system can be achieved by fusing advanced remote sensing data from multiple platforms and
agencies (e.g., http://eros.usgs.gov/ceos/satellites_midres1.shtml; http://www.ceos-
cove.org/index.php) in combination with national statistics, secondary data (e.g., elevation,
slope, soils, temperature, precipitation), and the systematic collection of field level observations.
An example of such a system on a regional scale is USDA, NASS Cropland Data Layer (CDL),
which is a raster, geo-referenced, crop-specific land cover data layer with a ground resolution of
30 meters (Johnson and Mueller., 2010). The GCAD will be a major contribution to Group on
Earth Observations (GEO) Global Agricultural Monitoring Initiative (GLAM), to the
overarching vision of GEO Agriculture and Water Societal Beneficial Areas (GEO Ag. SBAs),
G20 Agriculture Ministers initiatives, and ultimately to the Global Earth Observation System of
Systems (GEOSS). These initiatives are also supported by the Committee on Earth Observing
Satellites (CEOS) Strategic Implementation Team (SIT).
6
Within the context of the above facts, the overarching goal of this chapter is to provide a
comprehensive overview of the state-of-art of global cropland mapping procedures using remote
sensing as characterized and envisioned by the “Global Food Security Support Analysis Data @
30 m (GFSAD30)” project working group team. First, the chapter will provide an overview of
existing cropland maps and their characteristics along with establishing the gaps in knowledge
related to global cropland mapping. Second, definitions of cropland mapping along with key
parameters involved in cropland mapping based on their importance in food security analysis,
and cropland naming conventions for standardized cropland mapping using remote sensing will
be presented. Third, existing methods and approaches for cropland mapping will be discussed.
This will include the type of remote sensing data used in cropland mapping and their
characteristics along with discussions on the secondary data, field-plot data, and cropland
mapping algorithms. Fourth, currently existing global cropland products derived using remote
sensing will be presented and discussed. Fifth, a synthesis of all existing products leading to a
composite global cropland extent version 1.0 (GCE V1.0) is presented and discussed. Sixth, a
way forward for advanced global cropland mapping is visualized.
6.1 Global distribution of croplands and other land use and land cover: Baseline for the
year 2000
The first comprehensive global map of croplands was created by Ramankutty et al in 1998. A
more current version for the year 2000 shows the spatial distribution of global croplands along
with other land use and land cover classes (Figure 6.1). This provides a first view of where
global croplands are concentrated and helps us focus on the appropriate geographic locations for
detailed cropland studies. Water and snow (Class 8 and 9, respectively) have zero croplands and
occupy 44% of the total terrestrial land surface. Further, forests (class 6) occupy 17% of the
terrestrial area and deserts (class 7) an additional 12%. In these two classes, <5% of the total
croplands exist. Therefore, in order to study croplands systematically and intensively, one must
prioritize mapping in the areas of classes 1 to 5 (26% of the terrestrial area) where 95% of all
global croplands exist, with the first 3 classes (class 1, 2, 3) having 75% and the next 2 20%.
In the future, it is likely some of the non-croplands may be converted to croplands or vice versa,
highlighting the need for repeated and systematic global mapping of croplands. Segmenting the
world into cropland versus non-cropland areas routinely will help us understand and study these
change dynamics better.
7
Figure 6.1. Global croplands and other land use and land cover: Baseline.
8
6.1.1 Existing global cropland maps: Remote sensing and non-remote sensing approaches
1
There are currently six major global cropland maps: (1) Thenkabail et al. (2009a,b), (2)
2
Ramankutty et al. (1998), (3) Goldewijk et al. (2011), (4) Portmann et al. (2008), (5) Pittman et
3
al. (2010), and (6) Yu et al. (2013). These studies estimated the total global cropland area to be
4
around 1.5 billion hectares for the year 2000 as a baseline. However, there are 2 significant
5
differences in these products: 1) spatial disagreement on where the actual croplands are, and 2)
6
Irrigated to rainfed cropland proportions and their precise spatial locations. Globally, cropland
7
areas have increased from around 265 Mha in year 1700 to around 1,471 Mha in year 1990,
8
whilst the area of pasture has increased approximately six fold from 524 to 3,451 Mha (Foley et
9
al., 2011). Ramankutty and Foley (1998) estimated the cropland and pasture to represent about
10
36% of the world's terrestrial surface (148,940,000 km
2
), of which, according to different
11
studies, roughly 12% is croplands and 24% pasture. Multiple studies (Goldewijk et al., 2011;
12
Portmann et al., 2008; Ramankutty et al., 2008) integrated agricultural statistics and census data
13
from the national systems with spatial mapping technologies involving geographic information
14
systems (GIS) to derive global cropland maps.
15
16
Thenkabail and others (2011, 2009a,b) produced the first remote sensing-based global irrigated
17
and rainfed cropland maps and statistics through multi-sensor remote sensing data fusion along
18
with secondary data and in-situ data. They further used 5 dominant crop types (wheat, rice, corn,
19
barley, and soybeans) using parcel-based inventory data (Monfreda et al., 2008; Portmann et al.,
20
2008; Ramankutty et al., 2008) to produce a classification of global croplands with crop
21
dominance (Thenkabail et al., 2012). The five crops account for about 60% of the total global
22
cropland areas. The precise spatial location of these crops is only an approximation due to the
23
coarse resolution (approx. 1 km
2
) and fractional representation (1 to 100% crop in a pixel) of the
24
crop data in each grid cell of all the maps from which this composite map is produced
25
(Thenkabail et al. 2012). The existing global cropland datasets also differ from each other due to
26
inherent uncertainties in establishing the precise location of croplands, the watering methods
27
(rainfed versus irrigated), cropping intensities, crop types and/or dominance, and crop
28
characteristics (e.g. crop or water productivity measures such as biomass, yield, and water use).
29
Improved knowledge of the uncertainties (Congalton and Green, 2009) in these estimates will
30
lead to a suite of highly accurate spatial data products in support of crop modeling, food security
31
analysis, and decision support.
32
33
6.2 Key remote sensing derived cropland products: global food security
34
The production of a repeatable global cropland product requires a standard set of metrics and
35
attributes that can be derived consistently across the diverse cropland regions of the World. Four
36
key cropland information systems attributes that have been identified for global food security
37
analysis and that can be readily derived from remote sensing include (Figure 6.2): (a) cropland
38
extent\areas, (b) watering methods (e.g., irrigated, supplemental irrigated, rainfed), (c) crop
39
types, and (d) cropping intensities (e.g., single crop, double crop, continuous crop). Although not
40
the focus of this chapter, many other parameters are also derived in local regions, such as: (e)
41
precise location of crops, (f) cropping calendar, (g) crop health\vigor, (h) flood and drought
42
information, (i) water use assessments, and (j) yield or productivity (expressed per unit of land
43
and\or unit of water). Remote sensing is specifically suited to derive the four key products over
44
large areas using fusion of advanced remote sensing (e.g., Landsat, Resourcesat, MODIS) in
45
combination with national statistics, ancillary data (e.g., elevation, precipitation), and field-plot
46
9
data. Such a system, at the global level, will be complex in data handling and processing and
1
requires coordination between multiple agencies leading to development of a seamless, scalable,
2
transparent, and repeatable methodology. As a result, it is important to have systematic class
3
labeling convention as illustrated in Figure 6.3. A standardized class identifying and labeling
4
process (Figure 6.3) will enable consistent and systematic labeling of classes, irrespective of
5
analysts. First, the area is separated into cropland versus non-cropland. Then, within the cropland
6
class, labeling will involve (Figure 6.3): (a) cropland extent (cropland vs. non-cropland), (b)
7
watering source (e.g., irrigated versus rainfed), (c) irrigation source (e.g., surface water, ground
8
water), (d) crop type or dominance, (e) scale (e.g., large or contiguous, small or fragmented), and
9
(f) cropping intensity (e.g., single crop, double crop). The detail at which one maps at each stage
10
and each parameter would depend on many factors such as resolution of the imagery, available
11
ground data, and expert knowledge. For example, if there is no sufficient knowledge on whether
12
the irrigation is by surface water or ground water, but it is clear that the area is irrigated; one
13
could just map it as irrigated without mapping greater details on the type of irrigation. But, for
14
every cropland class, one has the potential to map the details as shown in Figure 6.3.
15
16
10
1
Figure 6.2. Key global cropland area products that will support food security analysis in the
2
twenty-first century.
3
4
11
1
Figure 6.3. Cropland class naming convention at different levels. Level I is most detailed and level IV is least detailed.
2
12
6.3 Definition of remote sensing-based cropland mapping products
1
Key to effective mapping is a precise and clear definition of what will be mapped. It is the first
2
and primary step, with different definitions leading to different products. For example, irrigated
3
areas are defined and understood differently in different applications and contexts. One can
4
define them as areas which receive irrigation at least once during their crop growing period.
5
Alternatively, they can be defined as areas which receive irrigation to meet at least half their crop
6
water requirements during the growing season. One other definition can be that these are areas
7
that are irrigated throughout the growing season. In each of these cases, the irrigated area extent
8
mapped will vary. Similarly, croplands can be defined as all agricultural areas irrespective of
9
type of crops grown or they may be limited to food crops (and not the fodder crops or plantation
10
crops). So, it is obvious that having a clear understanding of the definitions of what we map is
11
extremely important for the integrity of the products developed. We defined cropland products as
12
follows:
13
Minimum mapping unit
14
The minimum mapping unit of a particular crop is an area of 3 by 3 (0.81 hectares)
15
Landsat pixels identified as having the same crop type.
16
Cropland extent
17
All cultivated plants harvested for food, feed, and fiber, including plantations (e.g.,
18
orchards, vineyards, coffee, tea, rubber).
19
What is a cropland pixel?
20
>50% of pixel is cropped
21
Irrigated areas: Irrigation is defined as artificial application of any amount of water to
22
overcome crop water stress. Irrigated areas are those areas which are irrigated one or
23
more times during crop growing season.
24
Rainfed areas: areas that have no irrigation whatsoever and are precipitation dependent.
25
Cropping intensity
26
Number of cropping cycles within a 12 month period.
27
Crop type
28
8 crops (Wheat, Corn, Rice, Barley, Soybeans, Pulses, Cotton, Potatoes)
29
6.4 Data: Remote sensing and other data for global cropland mapping
30
Cropland mapping using remote sensing involves multiple types of data: satellite data with a
31
consistent and useful global repeat cycle, secondary data, statistical data, and field plot data.
32
When these data are used in an integrated fashion, the output products achieve highest possible
33
accuracies (Thenkabail et al., 2009b,c).
34
35
6.4.1 Primary satellite sensor data
36
Cropland mapping will require satellite sensor data across spatial, spectral, radiometric, and
37
temporal resolutions from a wide array of satellite/sensor platforms (Table 6.2) throughout the
38
growing season. These satellites and sensors are “representative” of hyperspectral, multispectral,
39
and hyperspatial data. The data points per hectare (Table 6.2, last column) will indicate the
40
spatial detail of agricultural information gathered. In addition to satellite based sensors, it is
41
always valuable to gather ground based hand-held spectroradiometer data from hyperspectral
42
sensors and\or imaging spectroscopy from ground based, airborne, or space borne sensors for
43
validation and calibration purposes (Thenkabail et al., 2011). Much greater details of a wide
44
array of sensors available to gather data are presented in Chapter 1 and 2 of Volume 1 of Remote
45
Sensing Handbook.
46
13
Table 6.2. Characteristics of some of the key satellite sensor data currently used in cropland mapping.
Satellite sensor
Wavelength range (μm)
Spatial resolution
(m)
Spectral bands
(#)
Temporal
(days)
Radiometric
(bits)
Data points
(per hectare)
A. Hyperspectral
EO-1 Hyperion
VNIR
SWIR
0.43-0.93
0.93-2.40
30
30
196
16
16
11.1 points for 30 m pixel
(0.09 hectares per pixel)
B. Advanced multispectral
Landsat TM
Multispectral
Band 1
Band 2
Band 3
Band 4
Band 5
Band 6
Band 7
Panchromatic
0.45-0.52
0.53-0.61
0.63-0.69
0.78-0.90
1.55-1.75
10.40-12.50
2.09-2.35
0.52-0.90
30
30
30
30
30
120/60
30
15
7/8
16
8
44.4 points for 15 m pixel
11.1 points for 30 m pixel
2.77 points for 60 m pixel
0.69 points for 120 m pixel
EO-1 ALI
Multispectral
Band 1
Band 2
Band 3
Band 4
Band 5
Band 6
Band 7
Band 8
Band 9
Panchromatic
0.43-0.45
0.45-0.52
0.52-0.61
0.63-0.69
0.78-0.81
0.85-0.89
1.20-1.30
1.55-1.75
2.08-2.35
0.48-0.69
30
30
30
30
30
30
30
30
30
10
10
16
16
14
ASTER
VNIR
Band 1
Band 2
Band 3N/3B
SWIR
Band 4
Band 5
Band 6
Band 7
Band 8
Band 9
TIR
Band 10
Band 11
Band 12
Band 13
Band 14
0.52-0.60
0.63-0.69
0.76-0.86
1.600-1.700
2.145-2.185
2.185-2.225
2.235-2.285
2.295-2.365
2.360-2.430
8.125-8.475
8.475-8.825
8.925-9.275
10.25-10.95
10.95-11.65
15
30
90
14
16
8
MODIS
MOD09Q1
Band1
Band2
0.62-0.67
0.84-0.876
250
2
1
12
MOD09A1
Band1
Band2
Band3
Band4
Band5
Band6
Band7
0.62-0.67
0.84-0.876
0.459-0.479
0.545-0.565
1.23-1.25
1.63-1.65
2.11-2.16
500
7*/36
1
12
C. Hyperspatial
GeoEye-1
Multispectral
Band 1
Band 2
Band 3
Band 4
Panchromatic
0.45-0.52
0.52-0.60
0.63-0.70
0.76-0.90
0.45-0.90
1.65
0.41
5
<3
11
59,488 points for 0.41 m
26,874 points for 0.61 m
10,000 points for 1 m
3673 points for 1.65 m
1679 points for 2.44 m
625 points for 4 m
400 points for 5 m
236 points for 6.5 m
100 points for 10 m
IKONOS
Multispectral
Band 1
Band 2
Band 3
Band 4
Panchromatic
0.45-0.52
0.51-0.60
0.63-0.70
0.76-0.85
0.53-0.93
4
1
5
3
11
Quickbird
5
1-6
11
15
Multispectral
Band 1
Band 2
Band 3
Band 4
Panchromatic
0.45-0.52
0.52-0.60
0.63-0.69
0.76-0.90
0.45-0.90
2.44
0.61
44.4 points for 15 m
1.23 points for 90 m
0.69 points for 120 m
0.16 points for 250 m
0.04 points for 500 m
Rapideye
Band 1
Band 2
Band 3
Band 4
Band 5
0.44-0.51
0.52-0.59
0.63-0.68
0.69-0.73
0.76-0.85
5-6.5
5
1-6
16
* MODIS has 36 bands, but we considered only the first 7 bands (Mod09A1).
16
6.4.2. Secondary data: There is a wide array of secondary or ancillary data such as the ASTER-
1
derived digital elevation data (GDEM), long (50 to 100 year) records of precipitation and
2
temperature, digital maps of soil types, and administrative boundaries. Many secondary data are
3
known to improve crop classification accuracies (references?). The secondary data will also form
4
core data for the spatial decision support system and final visualization tool in many systems.
5
6
6.4.3. Field-plot data: Field-plot data (e.g., Figure 6.4) will be used for purposes such as: (i)
7
Class identification and labeling; (ii) Determining irrigated area fractions, and (iii) Establishing
8
accuracies, errors, and uncertainties. At each field point (e.g., Figure 6.3), data such as cropland
9
or non-cropland, watering method (irrigated or rainfed), crop type, and cropping intensities are
10
recorded along with GPS locations, digital photographs, and other information (e.g., yield, soil
11
type) as needed. Field plot data will also help in gathering an ideal spectral data bank of
12
croplands. One could use the precise locations and the crop characteristics and generate
13
coincident remote sensing data characteristics (e.g., MODIS time-series monthly NDVI).
14
15
16
Figure 6.4. Field plot data for cropland studies collected over the Globe.
17
18
17
6.4.4 Very high resolution imagery data
1
Very high resolution (sub-meter to 5 meter) imagery (VHRI; see hyperspatial data characteristics
2
in Table 6.2) are widely available these days from numerous sources. These data can be used as
3
ground samples in localized areas to classify as well as verify classification results of the coarser
4
resolution imagery. For example, in Figure 6.5, VHRI tiles identify uncertainties existing in
5
cropland classification of coarser resolution imagery. VHRI are specifically useful for
6
identifying croplands versus non-croplands (Figure 6.5). They can also be used for identifying
7
irrigation based on associated features such as canals and tanks.
8
9
6.4.5 Data composition: Mega File Data Cube (MFDC) concept
10
Data pre-processing requires that all the acquired imagery is harmonized and standardized in
11
known time intervals (e.g., monthly, bi-weekly). For this, the imagery data is either acquired or
12
converted to at-sensor reflectance (see Chander et al., 2009, Thenkabail et al., 2004) and then
13
converted to surface reflectance using Landsat Ecosystem Disturbance Adaptive Processing
14
System (LEDAPS) processing system codes for Landsat (Masek et al., 2006) or similar codes
15
for other sensors. All data are processed and mosaicked to required geographic levels (e.g.,
16
global, continental). One method to organize these disparate but co-located data sets is through
17
the use of a mega-file data cube (MFDC). Numerous secondary datasets are combined in a
18
MFDC, which is then stratified using image segmentation into distinct precipitation-elevation-
19
temperature-vegetation zones. Data within the MFDC can include ASTER-derived refined
20
digital elevation from SRTM (GDEM), monthly long-term precipitation, monthly thermal skin
21
temperature, and forest cover and density. This segmentation allows cropland mapping to be
22
focused; creating distinctive segments of MFDCs and analyzing them separately for croplands
23
will enhance accuracy. For example, the likelihood of croplands in a temperature zone of <280
24
degree Kelvin is very low. Similarly, croplands in elevation above 1500 m will be of distinctive
25
characteristics (e.g., patchy, on hilly terrain most likely plantations of coffee or tea). Every layer
26
of data is geo-linked (having precisely same projection and datum and are geo-referenced to one
27
another).
28
29
The purpose of mega-file data cube (MFDC; see Thenkabail et al., 2009b for details) is to ensure
30
numerous remote sensing and secondary data layers are all stacked one over the other to form a
31
data cube akin to hyper spectral data cube. This approach has been used by X to map croplands
32
in Y (reference). The MFDC allows us to have the entire data stack for any geographic location
33
(global to local) as a single file available for analysis. For example, one can classify 10s or 100s
34
or even 1000s of data layers (e.g., monthly MODIS NDVI time series data for a geographic area
35
for an entire decade along with secondary data of the same area) stacked together in a single file
36
and classify the image. The classes coming out of such a mega-file data cube inform us about the
37
phenology along with other characteristics of the crop.
38
39
18
Figure 6.5. Very high resolution imagery used to resolve uncertainties in cropland mapping of Australia.
19
6.5 Cropland mapping methods
1
2
6.5.1 Remote sensing-based cropland mapping methods for global, regional, and local
3
scales
4
There is growing literature on cropland mapping across resolutions for both irrigated and rainfed
5
crops (Gumma et al., 2011; Friedl et al., 2002; Hansen et al., 2002; Loveland et al., 2000;
6
Ozdogan and Woodcock, 2006; Thenkabail et al., 2009a; Thenkabail et al., 2009c; Wardlow and
7
Egbert, 2008; Wardlow et al., 2007; Wardlow et al., 2006). Based on these studies, an ensemble
8
of methods that is considered most efficient include: (a) spectral matching techniques (SMTs)
9
(Thenkabail et al., 2007a; Thenkabail et al., 2009a; Thenkabail et al., 2009c); (b) decision tree
10
algorithms (DeFries et al., 1998); (c) Tassel cap brightness-greenness-wetness (Cohen and
11
Goward, 2004; Crist and Cicone, 1984; Masek et al., 2008); (d) Space-time spiral curves and
12
Change Vector Analysis (Thenkabail et al., 2005); (e) Phenology (Loveland et al., 2000;
13
Wardlow et al., 2006); and (f) climate data fusion with MODIS time-series spectral indices
14
using decision tree algorithms and sub-pixel classification (Ozdogan and Gutman, 2008). More
15
recently, cropland mapping algorithms which analyze end member spectra have been used for
16
global mapping by Thenkabail et al., (2009a, 2011).
17
18
6.5.2 Spectral Matching Techniques (SMTs) Algorithms: SMTs (Thenkabail et al., 2007a,
19
2009a, 2011) are innovative methods of identifying and labeling classes (see illustration in
20
Figure 6.6, 6.7a). For each derived class, this method identifies its characteristics over time using
21
MODIS time-series data (e.g., Figure 6.6). NDVI time-series or other metrics (Thenkabail et al.,
22
2005, 2007a, Biggs et al., 2006, Dheeravath et al., 2010) are analogous to spectra, where time is
23
substituted for wavelength. The principle in SMT is to match the shape, or the magnitude or both
24
to an ideal or target spectrum (pure class or “end-member”). The spectra at each pixel to be
25
classified is compared to the end-member spectra and the fit is quantified using the following
26
SMTs (Thenkabail et al., 2007a): (a) Spectral Correlation Similarity (SCS)-a shape measure; (b)
27
Spectral Similarity Value (SSV)-a shape and magnitude measure; (c) Euclidian Distance
28
Similarity (EDS)-a distance measure; and (d) Modified Spectral Angle Similarity (MSAS)-a
29
hyper angle measure.
30
31
6.5.2.1 Generating Class Spectra: The MFDC (section 6.4.5) of each of segment (Figure 6.6,
32
6.7a) is processed using ISOCLASS K-means classification to produce a large number of class
33
spectra with a unsupervised classification technique that are then interpreted and labeled. In more
34
localized applications, it is common to undertake a field-plot data collection to identify and label
35
class spectra. However, at the global scale this is not possible due to the enormous resources
36
required to cover vast areas to identify and label classes. Therefore, spectral matching techniques
37
(Thenkabail et al., 2007a) to match similar classes or to match class spectra from the
38
unsupervised classification with a library of ideal or target spectra (e.g., Figure 6.6a) will be used
39
to identify and label the classes.
40
41
6.5.2.2 Ideal Spectra Data Bank (ISDB): The term “ideal or target” spectrum refers to time-
42
series spectral reflectivity or NDVI generated for classes for which we have precise location-
43
specific ground knowledge. From these locations, signatures are extracted using MFDC,
44
synthesized, and aggregated to generate a few hundred signatures that will constitute an ISDB
45
(e.g., Figure 6.6, 6.7a).
46
20
1
6.6 Automated Cropland Classification Algorithm (ACCA) (Thenkabail et al., 2012, Wu et
2
al., 2014a, Wu et al., 2014b): The first part of the ACCA method involves knowledge-capture to
3
understand and map agricultural cropland dynamics by: (a) identifying croplands versus non-
4
croplands and crop type\dominance based on spectral matching techniques, decision trees tassel
5
cap bi-spectral plots, and very high resolution imagery; (b) determining watering method (e.g.,
6
irrigated or rainfed) based on temporal characteristics (e.g., NDVI), crop water requirement
7
(water use by crops), secondary data (elevation, precipitation, temperature), and irrigation
8
structure (e.g., canals and wells); (c) establishing croplands that are large scale (i.e., contiguous)
9
versus small scale (i.e., fragmented); (d) characterizing cropping intensities (single, double,
10
triple, and continuous cropping); (e) interpreting MODIS NDVI Temporal bi-spectral Plots to
11
Identify and Label Classes; and (f) using in-situ data from very high resolution imagery, field-
12
plot data, and national statistics (see Figure 6.7b for details). The second part of the method
13
establishes accuracy of the knowledge-captured agricultural map and statistics by comparison
14
with national statistics, field-plot data, and very high resolution imagery. The third part of the
15
method makes use of the captured-knowledge to code and map cropland dynamics through an
16
automated algorithm. The fourth part of the method compares the agricultural cropland map
17
derived using an automated algorithm (classified data) with that derived based on knowledge
18
capture (reference map). The fifth part of the method applies the tested algorithm on an
19
independent data set of the same area to automatically classify and identify agricultural cropland
20
classes. The sixth part of the method assesses accuracy and validates the classes derived from
21
independent dataset using an automated algorithm.
22
23
24
21
Figure 6.6. Spectral matching technique (SMT). In SMTs, the class temporal profile (NDVI curves) are matched with the ideal
temporal profile (quantitatively based on temporal profile similarity values) in order to group and identify classes as illustrated for a
rice class in this figure. a) Ideal temporal profile illustrated for “irrigated- surface-water-rice-double crop”; b) some of the class
temporal profile signatures that are similar, c) ideal temporal profile signature (Fig. 6.6a) matched with class temporal profiles (Fig.
6.6b), and d) the ideal temporal profile (Fig. 6.6a, in deep green) matches with class temporal profiles of classes 17 and 33 perfectly.
Then one can label classes 17 and 33 to be same as the ideal temporal profile (“irrigated- surface-water-rice-double crop”). This is a
qualitative illustration of SMTs. For quantitative methods refer to Thenkabail et al. 2007a.
22
1
Figure 6.7a. Cropland mapping method illustrated here for a global scale (see Thenkabail et al.,
2
2009b, 2011). The flowchart demonstrates comprehensive global cropland mapping methods
3
using multi-sensor, multi date remote sensing, secondary, field plot, and very high resolution
4
imagery data.
5
23
1
Figure 6.7b. Cropland mapping methods illustrated for a global scale. Top half shows
2
automated cropland classification algorithm (see Thenkabail and Wu, 2012; Wu et al., 2014a)
3
and bottom half shows class identification and labeling process.
4
5
24
6.7 Remote sensing-based global cropland products: current state-of-the-art maps, their
1
strengths, and limitations
2
Remote sensing offers the best opportunity to map and characterize global croplands most
3
accurately, consistently, and repeatedly. Currently, there are 3 global cropland maps that have
4
been developed using remote sensing techniques. In addition, we also considered a recent
5
MODIS global land cover and land use map where croplands are included. We examined these
6
maps to identify their strengths and weaknesses, to see how well they compare with each other,
7
and to understand the knowledge gaps that need to be addressed. These maps were produced by:
8
1. Thenkabail et al. (Thenkabail et al., 2009b, Biradar et al., 2009, Thenkabail et al., 2011);
9
2. Pittman et al. (2010);
10
3. Yu et al., (2013); and
11
4. Friedl et al (2010)
12
13
Thenkabail et al. (2009b, 2011; Figure 6.8, Table 6.3) used a combination of AVHRR, SPOT
14
VGT, and numerous secondary (e.g., precipitation, temperature, and elevation) data to produce a
15
global irrigated area map (Thenkabail et al., 2009b, 2011) and a global map of rainfed cropland
16
areas (Biradar et al., 2009, Thenkabail et al., 2011; Figure 6.8, Table 6.3). Pittman et al. (2010;
17
Figure 6.9, Table 6.4) used MODIS 250 m data to map global cropland extent. More recently,
18
Yu et al. (2013; Figure 6.10, Table 6.5) produced a nominal 30 m resolution cropland extent of
19
the world. These three global cropland extent maps are the best available current state-of-the-art
20
products. Friedl et al. (2010; Figure 6.11, Table 6.6) used 500 m MODIS data in their global land
21
cover and land use product (MCD12Q1) where croplands were one of land cover classes. The
22
methods, approaches, data, and definitions used in each of these products differ extensively. As a
23
result, the cropland extents mapped by these products also vary significantly. The areas in Tables
24
6.3-6.6 only show the full pixel areas (FPAs) and not sub-pixel areas (SPAs). SPAs are actual
25
areas, which can be estimated by re-projecting these maps to appropriate projections and
26
calculating the areas. For the purpose of this chapter, we did not estimate SPAs. However, a
27
comparison of the FPAs of the 4 maps (Figure 6.8 to 6.11) show significant differences in the
28
cropland areas (Table 6.3 to 6.6) as well as significant differences in the precise locations of the
29
croplands (Figure 6.8 to 6.11), the reasons for which are discussed in the next section.
30
25
Figure 6.8. Global cropland product by Thenkabail et al., (2011, 2009b) using the method illustrated in Figure 6.7 and described
in section 6.1.1 (details in Thenkabail et al., 2011, 2009b). This includes irrigated and rainfed areas of the world. The product is
derived using remotely sensed data fusion (e.g., NOAA AVHRR, SPOT VGT, JERS SAR), secondary data (e.g., elevation,
temperature, and precipitation), and in-situ data. Total area of croplands is 2.3 billion hectares.
26
Figure 6.9. Global cropland extent map by Pittman et al. (2010) derived using MODIS 250 m data. There is only one cropland class,
which includes irrigated and rainfed areas of the world. There is no discrimination between rainfed and irrigated areas. Total area of
croplands is 0.9 billion hectares.
.
27
Figure 6.10. Global cropland extent map by Yu et al. (2013) derived at nominal 30m data. Total area of croplands is 2.2 billion
hectares. There is no discrimination between rainfed and irrigated areas.
28
Figure 6.11. Global cropland classes (Class12 and Class14) extracted from MODIS Global land use and land cover (GLC)
500m product MCD12Q2 by Friedl et al. (2010). Total area of croplands is 2.7 billion hectares. There is no discrimination
between rainfed and irrigated cropland areas.
29
Table 6.3. Global cropland extent at nominal 1-km based on Thenkabail et al. (2009b, 2011)
1,2
.
1
Class#
Class Description
Pixels
Percent
#
Names
1 km
%
1
Croplands, irrigated dominance
9359647
40%
2
Croplands, rainfed dominance
14273248
60%
3
Natural vegetation with minor cropland
fractions
5504037
4
Natural vegetation dominance with very
minor cropland fractions
44170083
23632895
100%
1
=total of approximately 2.3 billion hectares; Note that these are full pixel areas (FPAs). Actual area is =
sub-pixel area (SPA). The SPA is not estimated here. See Thenkabail et al. (2007b) for the methods for
calculating SPAs.
2
= % calculated based on class 1 and 2. Class 3 and 4 are very small cropland fragments
2
Table 6.4. Global cropland extent at nominal 250 m based on Pittman et al. (2010)
1,2
.
3
Class#
Class Description
Pixels
Percent
#
Names
1km
%
1
Croplands
8948507
100
1
=total of approximately 0.9 billion hectares. Note that these are full pixel areas (FPAs). Actual area is =
sub-pixel area (SPA). The SPA is not estimated here. See Thenkabail et al. (2007b) for the methods for
calculating SPAs.
2
= % calculated based on class 1
4
Table 6.5. Global cropland extent at nominal 30 m based on Yu et al. (2013)
1,2
.
5
Class#
Class Description
Pixels
Percent
#
Names
1km
%
1
Croplands (classes 10 to 14)
7750467
35
2
Bare-cropland(classes 94 and 24)
14531323
65
22281790
100
1
=total of approximately 2.2 billion hectares. Note that these are full pixel areas (FPAs). Actual area is =
sub-pixel area (SPA). The SPA is not estimated here. See Thenkabail et al. (2007b) for the methods for
calculating SPAs.
2
= % calculated based on class 1 and 2.
6
Table 6.6. Global cropland extent at nominal 500 m based on Friedl et al. (2010)
1
.
7
Class#
Class Description
Pixels
Percent
#
Names
1km
%
1
Global croplands (Class 12 and 14)
27046084
100
1
= approximately, total 2.7 billion hectares based on class12 and 14. Note that these are full pixel areas
(FPAs). Actual area is = sub-pixel area (SPA). The SPA is not estimated here. See Thenkabail et al.
(2007b) for the methods for calculating SPAs.
8
6.7.1 Global cropland extent at nominal 1-km resolution
9
We synthesized the above 4 global cropland products and produced a unified Global Cropland
10
Extent map GCE V1.0 at nominal 1 km (Table 6.7a; Figure 6.12a). The process involved
11
resampling each global cropland product to a common resolution of 1 km and then performing
12
GIS data overlays to determine where the cropland extents matched and where they differed.
13
30
1
Figure 6.12a shows the aggregated global cropland extent map with its statistics in Table 6.7a.
2
Class 1 in Figure 6.12a and Table 6.7a provides the global cropland extent included in all 4
3
maps. Actual area of this extent is not calculated yet, but it includes approximately 2.3 billion
4
full pixel areas (FPAs) (Table 6.7a). The spatial distribution of these 2.3 billion hectares is
5
demonstrated as class 1 in Figure 12a. Class 2 and 3 are areas with minor or very minor cropland
6
fractions. Class 2 and Class 3 are classes with large areas of natural vegetation and\or desert
7
lands and other lands.
8
9
Figure 6.12b and Table 6.7b demonstrate where and by how much the 4 products match with one
10
another. For example, 2,802,397 pixels (class 1, Table 6.7b, Figure 6.12b) are croplands that are
11
irrigated. Some of the products do not separately classify irrigated vs rainfed croplands, although
12
all 4 products show where croplands are. We first identified where all 4 products match as
13
croplands and then added irrigation status or other indicators (e.g., irrigation dominance, rainfed;
14
Table 6.7b) from the product by Thenkabail et al(2009b, 2011).
15
16
Table 6.7b and Figure 6.12b show 12 classes of which classes 1 and 2 are croplands with
17
irrigated agriculture, classes 3 and 4 are croplands with rainfed agriculture, classes 5 and 6 are
18
croplands where irrigated agriculture dominates, classes 7 and 8 are croplands where rainfed
19
agriculture dominates, and classes 9 to 12 are areas with minor or very minor cropland fractions.
20
Classes 9 to 12 are those with large areas of natural vegetation and\or desert lands and other
21
lands.
22
23
Interestingly, and surprisingly as well, only 20% (class 1 and 3; Table 6.7b, Figure 6.12b) of the
24
total cropland extent are matched precisely in all 4 products. Further, 49% (Class 1, 2, 3, 4, and
25
7; Table 6.7b, Figure 6.12b) of the total cropland areas match in at least 3 of the 4 products. This
26
implies that all the 4 products have considerable uncertainties in determining the precise location
27
of the croplands. The great degree of uncertainty in the cropland products can be attributed to
28
factors including:
29
A. Coarse resolution of the imagery used in the study;
30
B. Definition of mapping products of interest;
31
C. Methods and approaches adopted ; and
32
D. Limitations of the data.
33
34
Table 6.7c and Figure 6.12c show 5 classes of which classes 1 and 2 are croplands with irrigated
35
agriculture, classes 3 is croplands with rainfed agriculture, classes 4 and 5 have ONLY minor or
36
very minor cropland fractions. We recommend the use of this aggregated 5 class global cropland
37
map (Figure 12c and Table 6.7c) produced based on the 4 major cropland mapping efforts [i.e.,
38
Thenkabail et al. (2009a, 2011), Pittman et al. (2010), Yu et al. (2013), and Friedl et al. (2010)]
39
using remote sensing. This map (Figure 6.12c, Table 6.7c) provides clear consensus view on of 4
40
major studies on global:
41
Cropland extent location;
42
Cropland watering method (irrigation versus rainfed).
43
The product ((Figure 6.12c, Table 6.7c) does not show where the crop types are or even the crop
44
dominance. However, cropping intensity can be gathered using multi-temporal remote sensing
45
over these cropland areas.
46
31
Figure 6.12a. An aggregated three class global cropland extent map at nominal 1-km based on four major studies: Thenkabail et al.
(2009a, 2011), Pittman et al. (2010), Yu et al. (2013), and Friedl et al. (2010). Class 1 is total cropland extent; total cropland extent is
2.3 billion hectares (full pixel areas). Class 2 and Class 3 have ONLY minor fractions of croplands. Refer to Table 6.7a for cropland
statistics of this map.
32
Figure 6.12b. A disaggregated twelve class global cropland extent map derived at nominal 1-km based on four major studies:
Thenkabail et al. (2009a, 2011), Pittman et al. (2010), Yu et al. (2013), and Friedl et al. (2010). Class 1 to Class 9 are cropland classes,
that are dominated by irrigated and rainfed agriculture. Class 10 to and Class 12 have ONLY minor or very minor fractions of
croplands. Refer to Table 6.7b for cropland statistics of this map.
33
Figure 6.12c. A disaggregated five class global cropland extent map derived at nominal 1-km based on four major studies: Thenkabail
et al. (2009a, 2011), Pittman et al. (2010), Yu et al. (2013), and Friedl et al. (2010). Class 1 to Class 5 are cropland classes, that are
dominated by irrigated and rainfed agriculture. However, class 4 and Class 5 have ONLY minor or very minor fractions of croplands.
Refer to Table 6.7c for cropland statistics of this map. Note: Irrigation major: areas irrigated by large reservoirs created by large and
medium dams, barrages and even large ground water pumping. Irrigation minor: areas irrigated by small reservoirs, irrigation tanks,
open wells, and other minor irrigation. However, it is very hard to draw a strict boundary between major and minor irrigation and in
places there can be significant mixing. So, when major irrigated areas such as the Ganges basin, California’s central valley, Nile basin
etc. are clearly distinguishable as major irrigation, in other areas major and minor irrigation may inter-mix.
34
Table 6.7a. Global cropland extent at nominal 1-km based on four major studies: Thenkabail et
al. (2009b, 2011), Pittman et al. (2010), Yu et al. (2013), and Friedl et al.(2010). Three class
map
1,2,3
.
Table 6.7b. Global cropland extent at nominal 1-km based on four major studies: Thenkabail et
al. (2009b, 2011), Pittman et al. (2010), Yu et al. (2013), and Friedl et al. (2010). Twelve class
map
1,2,3,4
.
Class#
Class Description
Pixels
Percent
#
Names
1 km
%
1
Croplands all 4, irrigated
2802397
12
2
Croplands 3 of 4 , irrigated
289591
1
3
Croplands all 4, rainfed
1942333
8
4
Croplands 3 of 4, rainfed
427731
2
5
Croplands, 2 of 4, irrigation dominance
3220330
14
6
Croplands, 2 of 4, irrigation dominance
1590539
7
7
Croplands, 3 of 4, rainfed dominance
6206419
26
8
Croplands, 2 of 4, rainfed dominance
3156561
13
9
Croplands, minor fragments, 2 of 4
3858035
17
10
Croplands, very minor fragments, 2 of 4
6825290
11
Croplands, minor fragments, 1 of 4
6874886
12
Croplands, very minor fragments, 1 of 4
44662570
Class 1 to 9 total
23493936
100
1
= approximately 2.3 billion hectares (class 1 to 9) of cropland is estimated. But this is full pixel
area. Actual area is = sub-pixel area (SPA). The SPA is not estimated here. See Thenkabail et al.
(2007b) for the methods for calculating SPAs.
2
= % calculated based on class 1 to 9
3
=Class 10,11and 12 are minor cropland fragments
4
= all 4 means , all 4 studies agreed
Class#
Class Description
Pixels
Percent
#
Names
1 km
%
1
1. Croplands
23493936
100
2
2.Cropland minor fractions
13700176
3
3.Cropland very minor fractions
44662570
1
= approximately 2.3 billion hectares (class 1) of cropland is estimated. But this is full pixel
area. Actual area is = sub-pixel area (SPA). The SPA is not estimated here. See Thenkabail et al.
(2007b) for the methods for calculating SPAs.
2
= % calculated based on Class 1.
3
= Class 2 and 3are minor / very minor cropland fragments
35
Table 6.7c. Global cropland extent at nominal 1-km based on four major studies: Thenkabail et
al. (2009b, 2011), Pittman et al. (2010), Yu et al. (2013), and Friedl et al.(2010). Five class
map
1,2,3
.
Class#
Class Description
Pixels
Percent
#
Names
1 km
%
1
1.Croplands, irrigation major
3091988
13
2
2.Croplands, irrigation minor
4810869
21
3
3.Croplands, rainfed
11733044
50
4
4.Croplands, rainfed minor fragments
3858035
16
5
5.Croplands, rainfed very minor fragments
13700176
Class 1 to 4 total
23493936
100.0%
1
= approximately 2.3 billion hectares (class 1 to 4 ) of cropland is estimated. But this is full pixel
area. Actual area is = sub-pixel area (SPA). The SPA is not estimated here. See Thenkabail et al.
(2007b) for the methods for calculating SPAs.
2
= % calculated based on Class 1 to 4.
3
= Class 5 is very minor cropland fragments
36
6.8 Change Analysis: Once the croplands are mapped (Figure 6.13), we can use the time-series
1
historical data such as continuous global coverage of remote sensing data from NOAA Very
2
High Resolution Radiometer (VHRR) and Advanced VHRR (AVHRR), Global Inventory
3
Modeling and Mapping Studies (GIMMS; 1982-2000), MODIS time-series (2001-present) to
4
help build an inventory of historical agricultural development (e.g., Figure 6.13, 6.14). Such an
5
inventory will provide information including identifying areas that have switched from rainfed to
6
irrigated production (full or supplemental), and non-cropped to cropped (and vice versa). A
7
complete history will require systematic analysis of remotely sensed data as well as a systematic
8
compilation of all routinely populated cropland databases from the agricultural departments of all
9
countries throughout the world. The differences in pixel sizes in AVHRR versus MODIS will:
10
(a) influence class identification and labeling, and (b) cause different levels of uncertainties. We
11
will address these issues by determining sub-pixel areas and uncertainties involved in class
12
accuracies and uncertainties in areas at various spatial resolutions using methods detailed in
13
recent work of this team (Thenkabail et al. 2007b, Velpuri et al., 2009, and Ozdogan and
14
Woodcock 2006). Change analyses (Tomlinson, 2003) are conducted in order to investigate both
15
the spatial and temporal changes in croplands (e.g., Figure 6.13, 6.14) that will help establish: (a)
16
change in total cropland areas, (b) change in spatial location of cropland areas, (c) expansion on
17
croplands into natural vegetation, (d) expansion of irrigation, (e) change from croplands to bio-
18
fuels, and (f) change from croplands to urban. Massive reductions in cropland areas in certain
19
parts of the world will be detected, including cropland lost as a result of reductions in available
20
ground water supply due to overdraft (Wada et al., 2012, Rodell et al., 2010).
21
22
6.9 Uncertainties of existing cropland products: Currently, the main causes of uncertainties in
23
areas reported in various studies (Ramankutty et al., 2008 versus; Thenkabail et al., 2009a;
24
Thenkabail et al., 2009c) can be attributed to, but not limited to: (a) reluctance of national and
25
state agencies to furnish the census data on irrigated area and concerns of their institutional
26
interests in sharing of water and water data; (b) reporting of large volumes of census data with
27
inadequate statistical analysis; (c) subjectivity involved in the observation-based data collection
28
process; (d) inadequate accounting of irrigated areas, especially minor irrigation from
29
groundwater, in national statistics; (e) definitional issues involved in mapping using remote
30
sensing as well as national statistics; (f) difficulties in arriving at precise estimates of area
31
fractions (AFs) using remote sensing; (g) difficulties in separating irrigated from rainfed
32
croplands; and (h) imagery resolution in remote sensing. Other limitations include (Thenkabail et
33
al., 2009a, 2011):
34
A. Absence of precise spatial location of the cropland areas for training and
35
validation;
36
B. Uncertainties in differentiating irrigated areas from rainfed areas;
37
C. Absence of crop types and cropping intensities;
38
D. Inability to generate cropland maps and statistics, routinely; and
39
E. Absence of dedicated web\data portal for dissemination cropland products.
40
41
42
37
Figure 6.13. Center image of global cropland (irrigated and rainfed) areas @ 1 km for year 2000 produced by overlying the
remote sensing derived product of the International Water Management Institute (IWMI; Thenkabail et al., 2012, 2011, 2009a, 2009b;
http://www.iwmigiam.org) over 5 dominant crops (wheat, rice, maize, barley and soybeans) of the world produced by Ramankutty et
al. (2008). The 5 crops constitute about 60% of all global cropland areas. The IWMI remote sensing product is derived using remotely
sensed data fusion (e.g., NOAA AVHRR, SPOT VGT, JERS SAR), secondary data (e.g., elevation, temperature, and precipitation),
and in-situ data. Total area of croplands is 1.53 billion hectares of which 399 million hectares is total area available for irrigation
(without considering cropping intensity) and 467 million hectares is annualized irrigated areas (considering cropping intensity).
Surrounding NDVI images of irrigated areas: The January to December irrigated area NDVI dynamics is produced using NOAA
AVHRR NDVI. The irrigated areas were determined by Thenkabail et al. (2011, 2009a, b).
38
Figure 6.14. Global agricultural dynamics over 2 decades illustrated here for some of the most significant agricultural areas of the
World. Once we establish GCAD2010 and GCAD1990 at nominal 30 m resolution for the entire world, we will use AVHRR-MODIS
monthly MVC NDVI time-series from 1982 to 2017 to provide a continuous time history of global irrigated and rainfed croplands,
establish their spatial and temporal changes, and highlight the hot spots of change. The GCAD2010, GCAD1990, and GCAD four
decade’s data will be made available on USGS global cropland data portal (currently under construction):
http://powellcenter.usgs.gov/current_projects.php#GlobalCroplandsAbstract.
39
Further, the need to map accurately specific cropland characteristics such as crop types and
1
watering methods (e.g., irrigated vs. rainfed) is crucial in food security analysis. For example, the
2
importance of irrigation to global food security is highlighted in a recent study by Siebert and
3
Doll (2009) who show that without irrigation there would be a decrease in production of various
4
foods including dates (60%), rice (39%), cotton (38%), citrus (32%), and sugarcane (31%) from
5
their current levels. Globally, without irrigation cereal production would decrease by a massive
6
43%, with overall cereal production, from irrigated and rainfed croplands, decreasing by 20%.
7
8
These limitations are a major hindrance in accurate/reliable global, regional, and country-by-
9
country water use assessments that in turn support crop productivity (productivity per unit of
10
land; kg\m
2
) studies, water productivity (productivity per unit of water; kg\m
3
) studies, and food
11
security analyses. The higher degrees of uncertainty in coarser resolution data are a result of an
12
inability to capture fragmented, smaller patches of croplands accurately, and the homogenization
13
of both crop and non-crop land within areas of patchy land cover distribution. In either case,
14
there is a strong need for finer spatial resolution to resolve the confusion.
15
16
6.10 Way forward
17
Given the above issues with existing maps of global croplands, the way forward will be to
18
produce global cropland maps at finer spatial resolution and applying a suite of advanced
19
analysis methods. Previous research has shown that at finer spatial resolution the accuracy of
20
irrigated and rainfed area class delineations improve because at finer spatial resolution more
21
fragmented and smaller patches of irrigated and rainfed croplands can be delineated (Ozdogan
22
and Woodcock, 2006; Velpuri et al., 2009). Further, greater details of crop characteristics such as
23
crop types (e.g., Figure 6.15) can be determined at finer spatial resolutions. Crop type mapping
24
will involve use of advanced methods of analysis such as data fusion of higher spatial resolution
25
images from sensors such as Resourcesat\Landsat and AWiFS\MODIS (e.g., Table 6.2)
26
supported by extensive ground surveys and ideal spectral data bank (ISDB) (Thenkabail et al.,
27
2007a). Harmonic analysis is often adopted to identify crop types (Sakamoto et al., 2005) using
28
methods such as the conventional Fourier analysis and adopting a Fourier Filtered Cycle
29
Similarity (FFCS) method. Mixed classes are resolved using hierarchical crop mapping protocol
30
based on decision tree algorithm (Wardlow and Egbert, 2008). Irrigated versus rainfed croplands
31
will be distinguished using spectral libraries (Thenkabail et al., 2007) and ideal spectral data
32
banks (Thenkabail et al., 2009a, 2007a). Similar classes will be grouped by matching class
33
spectra with ideal spectra based on spectral matching techniques (SMTs; Thenkabail et al.,
34
2007a). Details such as crop types are crucial for determining crop water use, crop productivity,
35
and water productivity leading to providing crucial information needed for food security studies.
36
However, the high spatial resolution must be fused with high temporal resolution data in order to
37
obtain time-series spectra that are crucial for monitoring crop growth dynamics and cropping
38
intensity (e.g., single crop, double crop, and continuous year round crop). Numerous other
39
methods and approaches exist. But, the ultimate goal using multi-sensor remote sensing is to
40
produce croplands products such as:
41
1. Cropland extent\area,
42
2. Crop types (initially focused on 8 crops that occupy 70% of global croplands),
43
3. Irrigated vs. rainfed croplands,
44
4. Cropping intensities\phenology (single, double, triple, continuous cropping),
45
5. Cropped area computation; and
46
40
6. Cropland change over space and time
1
2
3
Figure 6.15. Rice map of south Asia produced using the method illustrated in Figure 6.6.
4
[Source: Gumma et al., 2011].
5
6
6.11 Conclusions
7
This chapter provides an overview of the importance of global cropland products in food security
8
analysis. It is obvious that only remote sensing from Earth Observing (EO) satellites provides
9
consistent, repeated, high quality data for characterizing and mapping key cropland parameters
10
for global food security analysis. Importance of definitions and class naming conventions in
11
cropland mapping has been re-iterated. Typical EO systems and their spectral, spatial, temporal,
12
and radiometric characteristics useful for cropland mapping have been highlighted. The chapter
13
provides a review of various cropland mapping methods used at global, regional, and local
14
41
levels. One of the remote sensing methods for global cropland mapping has been illustrated. The
1
current state-of-the-art provides four key global cropland products (listed below later in this
2
paragraph) derived from remote sensing, each produced by a different group. These products
3
have been produced using: (a) time-series of multi-sensor data and secondary data, (b) 250 m
4
MODIS time-series data, (c) 30 m Landsat data, and(d) a MODIS 500 m time-series derived
5
cropland classes from a land use\land cover product has been used. These four products were
6
synthesized, at nominal 1 km, to obtain a unified cropland mask of the world (global cropland
7
extent version 1.0 or GCE V1.0). It was demonstrated from these products that the uncertainty in
8
location of croplands in any one given product is quite high and no single product maps
9
croplands particularly well. Therefore, a synthesis identifies where some or all of these products
10
agree and where they disagree. This provides a starting point for the next level of more detailed
11
cropland mapping at 250 m and 30 m. The key cropland parameters identified to be derived from
12
remote sensing are: (1) cropland extent\areas, (2) cropping intensities, (3) watering method
13
(irrigated versus rainfed), (4) crop type, and (5) cropland change over time and space. From these
14
primary products one can derive crop productivity and water productivity. Such products have
15
great importance and relevance in global food security analysis.
16
Authors recommend the use of composite global cropland map (see Figure 6.12c, Table 6.7c)
17
that provides clear consensus view on of 4 major cropland studies on global:
18
Cropland extent location;
19
Cropland watering method (irrigation versus rainfed).
20
The product (Figure 6.12c, Table 6.7c) does not show where the crop types are or even the crop
21
dominance. However, cropping intensity can be gathered using multi-temporal remote sensing
22
over these cropland areas.
23
24
25
6.12 Acknowledgements
26
The authors would like to thank NASA Making Earth Science Data Records for Use in Research
27
Environments (MEaSUREs) solicitation for funding this research. Support by USGS Powell
28
Center for a working group on global croplands is much appreciated. We thank the global food
29
security support analysis data @ 30 m (GFSAD30) project team for inputs. Figure 6.1 and 6.2
30
were produced by Dr. Zhuoting Wu, Mendenhall Fellow, USGS. We thank her for it.
31
32
33
34
35
36
42
6.13 References
1
2
Biggs, T., Thenkabail, P.S., Krishna, M., GangadharaRao, P., & Turral, H. (2006). Vegetation
3
phenology and irrigated area mapping using combined MODIS time-series, ground surveys, and
4
agricultural census data in Krishna River Basin, India. International Journal of Remote Sensing,
5
27(19), 4245-4266.
6
7
Bindraban, P. S., Bulte, E. H., & Conijn, S. G. (2009). Can large-scale biofuel production be
8
sustained by 2020? Agricultural Systems, 101, 197-199.
9
10
Biradar, C. M., P. S. Thenkabail, P. Noojipady, Y. Li, V. Dheeravath, H. Turral, M. Velpuri, et
11
al. 2009. ‘‘A Global Map of Rainfed Cropland Areas (GMRCA) at the End of Last Millennium
12
Using Remote Sensing.’’ International Journal of Applied Earth Observation & Geoinformation
13
11 (2): 114_129. doi:10.1016/j.jag.2008.11.002.
14
15
Chander, G., Markham, B.L., and Helder,D.L. 2009. Summary of current radiometric calibration
16
coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors, Remote Sensing of
17
Environment, Volume 113, Issue 5, 15 May 2009, Pages 893-903, ISSN 0034-4257,
18
http://dx.doi.org/10.1016/j.rse.2009.01.007.
19
(http://www.sciencedirect.com/science/article/pii/S0034425709000169)
20
21
Cohen, W. B., & Goward, S. N. (2004). Landsat's role in ecological applications of remote
22
sensing. BioScience, 54, 535-545.
23
24
Congalton, R. 2009. Accuracy and Error Analysis of Global and Local Maps: Lessons Learned
25
and Future Considerations. IN: Remote Sensing of Global Croplands for Food Security. P.
26
Thenkabail, J. Lyon, H. Turral, and C. Biradar. (Editors). CRC/Taylor & Francis, Boca Raton,
27
FL pp. 441-458.
28
29
Congalton, R. and K. Green. 2009. Assessing the Accuracy of Remotely Sensed Data: Principles
30
and Practices. 2nd Edition. CRC/Taylor & Francis, Boca Raton, FL 183p.
31
32
Congalton, R. 1991. A review of assessing the accuracy of classifications of remotely sensed
33
data. Remote Sensing of Environment. Vol. 37, pp. 35-46.
34
35
Crist, E. P., & Cicone, R. C. (1984). Application of the tasseled cap concept to simulated
36
Thematic Mapper data. Photogrammetric Engineering and Remote Sensing, 50, 343-352.
37
38
DeFries, R., Hansen, M., Townsend, J. G. R., & Sohlberg, R. (1998). Global land cover
39
classifications at 8 km resolution: the use of training data derived from Landsat imagery in
40
decision tree classifiers. International Journal of Remote Sensing, 19, 3141-3168.
41
42
Dheeravath, V., Thenkabail, P.S., Chandrakantha, G, Noojipady, P., Biradar, C.B., Turral. H.,
43
Gumma, M.1, Reddy, G.P.O., Velpuri, M. 2010. Irrigated areas of India derived using MODIS
44
500m data for years 2001-2003. ISPRS Journal of Photogrammetry and Remote Sensing.
45
http://dx.doi.org/10.1016/j.isprsjprs.2009.08.004. 65(1): 42-59.
46
43
1
EPW (2008). Food Security Endangered: Structural changes in global grain markets threaten
2
India’s food security. Economic and Political Weekly, 43, 5.
3
4
FAO (2009). Food outlook: Outlook Global Market Analysis. Rome: FAO.
5
6
Foley, J. A., Monfreda, C., Ramankutty, N., & Zaks, D. (2007). Our share of the planetary pie.
7
PNAS, 104, 12585-12586.
8
9
Foley, J. A., DeFries, R, Asner, G.P,, Barford, C,, Bonan, G,, Carpenter, S,R,, Chapin, F.S., Coe,
10
M.T., Daily, G.C, Gibbs, H.K., Helkowski, J.H., Holloway, T., Howard, E.A., Kucharik, C.J.,
11
et al., 2011. Solutions for a cultivated planet, Nature, 478(7369): 337-342.
12
13
Falkenmark, M., & Rockström, J. (2006). The New Blue and Green Water Paradigm: Breaking
14
New Ground for Water Resources Planning and Management. Journal of Water Resource
15
Planning and Management, 132, 1-15.
16
17
Friedl, M. A., McIver, D. K., Hodges, J. C. F., Zhang, X. Y., Muchoney, D., & Strahler, A. H.
18
(2002). Global land cover mapping from MODIS: Algorithms and early results. Remote Sensing
19
of Environment, 83, 287-302.
20
21
Friedl, M.A., Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N.S., and Huang, X.M.
22
2010. MODIS Collection 5 global land cover: Algorithm refinements and characterization of
23
new datasets. REMOTE SENSING OF ENVIRONMENT, 114(1), 168-182.
24
25
Funk, C., & Brown, M. (2009). Declining global per capita agricultural production and warming
26
oceans threaten food security. Food Security, DOI: 10.1007/s12571-009-0026-y.
27
28
Gibbs, H. K., Johnston, M., Foley, J. A., Holloway, T., Monfreda, C., Ramankutty, N., & Zaks,
29
D. (2008). Carbon payback times for crop-based biofuel expansion in the tropics: the effects of
30
changing yield and technology. Environ. Res. Lett., 034001, doi: 10.1088/1748-
31
9326/3/3/034001.
32
33
Goodchild, M. and S. Gopal (eds.) 1989. The Accuracy of Spatial Databases. Taylor and Francis.
34
New York. 290 p.
35
36
Gordon, L. J., Finlayson, C. M., & Falkenmark, M. (2009). Managing water in agriculture for
37
food production and other ecosystem services. Agricultural Water Management, 97, doi: DOI:
38
10.1016/j.agwat.2009.03.017.
39
40
Gumma, M.K., Nelson, A., Thenkabail, P.S., and Singh, A.N. 2011. "Mapping rice areas of
41
South Asia using MODIS multi temporal data", J. Appl. Remote Sens. 5, 053547 (Sep 01, 2011);
42
doi:10.1117/1.3619838.
43
44
44
Gutman, G., Byrnes, R., Masek, J., Covington, S., Justice, C., Franks, S., and Headley, R. 2008.
1
Towards monitoring Land-cover and land-use changes at a global scale: the global land survey
2
2005. Photogrammetric Engineering and Remote Sensing. 74(1):6-10.
3
Goldewijk, K., A. Beusen, M. de Vos and G. van Drecht (2011). The HYDE 3.1 spatially
4
explicit database of human induced land use change over the past 12,000 years, Global Ecology
5
and Biogeography 20(1): 73-86.DOI: 10.1111/j.1466-8238.2010.00587.x.
6
Hansen, M. C., DeFries, R. S., Townshend, J. R. G., Sohlberg, R., Dimiceli, C., & Carroll, M.
7
(2002). Towards an operational MODIS continuous field of percent tree cover algorithm:
8
examples using AVHRR and MODIS data. Remote Sensing of Environment. 83, 303-319.
9
10
Hossain, M., Janaiah, A., & Otsuka, K. (2005). Is the productivity impact of the Green
11
Revolution in rice vanishing? Economic and Political Weekly, 5595-9600.
12
13
Jiang, Y. (2009). China's water scarcity. Journal of Environmental Management, 90, 3185-3196.
14
15
Johnson, D.M.,& Mueller, R., 2010. The 2009 cropland data layer. Photogrammetric
16
Engineering and Remote Sensing. 76 (11), 12011205.
17
18
Khan, S., & Hanjra, M. A. (2008). Sustainable land and water management policies and
19
practices: a pathway to environmental sustainability in large irrigation systems. Land
20
Degradation and Development, 19, 469487.
21
22
Kumar, M. D., & Singh, O. P. (2005). Virtual water in global food and water policy making: Is
23
there a need for rethinking? Water Resources Management, 19, 759 - 789.
24
25
Kurz, B and Seelan, S. K. 2007. Use of Remote Sensing to Map Irrigated Agriculture in Areas
26
Overlying the Ogallala Aquifer, United States. Book Chapter in Global Irrigated Areas. IWMI.
27
Under Review.
28
29
Lal, R., & Pimentel, D. (2009). Biofuels from crop residues. Soil and Tillage Research, 93, 237-
30
238.
31
32
Loveland, T. R., Reed, B. C., Brown, J. F., Ohlen, D. O., Zhu, Z., & Yang, L. (2000).
33
Development of global land cover characteristics database and IGBP DISCover from 1 km
34
AVHRR data. International Journal of Remote Sensing, 21, 1303-1330.
35
36
McIntyre, B. D. (2008). International assessment of agricultural knowledge, science and
37
technology for development (IAASTD) : global report., Includes bibliographical references and
38
index. ISBN 978-1-59726-538-6, Oxford, UK.
39
40
Masek, J. G., Huang, C., Wolfe, R., Cohen, W., Hall, F., Kutler, J., & Nelson, P. (2008). North
41
American forest disturbance mapped from a decadal Landsat record. Remote Sensing of
42
Environment, 112, 2914-2926.
43
44
45
Masek, J.G., E.F. Vermote, N. Saleous, R. Wolfe, F.G. Hall, Huemmrich, F.Gao, J.Kutler, and
1
T.K.Lim, 2006, A Landsat surface reflectance data set for North America, 1990-2000,
2
Geoscience and Remote Sensing Letters, 3, 68-72.
3
4
Monfreda, C., Patz, J, A., Prentice, I.C., Ramankutty, N., Snyder, P.K. Global consequences of
5
land use Science. 2005 Jul 22; 309(5734):570-4.
6
7
Monfreda, C., N. Ramankutty, and J. A. Foley, 2008. Farming the planet: 2. Geographic
8
distribution of crop areas, yields, physiological types, and net primary production in the year
9
2000, Global Biogeochem. Cycles, 22(1): GB1022.
10
11
Olofsson, P., Stehman, S.V., Woodcock, C.E., Sulla-Menasche, D., Sibley, A.M., Newell, J.D.,
12
Friedl, M.A., and Herold, M. 2011. A global land cover validation dataset, I: fundamental design
13
principles. International Journal of Remote Sensing. In Review.
14
15
16
Ozdogan, M., & Woodcock, C. E. (2006). Resolution dependent errors in remote sensing of
17
cultivated areas. Remote Sensing of Environment, 103, 203-217.
18
19
Ozdogan, M., and G. Gutman, 2008. A new methodology to map irrigated areas using multi-
20
temporal MODIS and ancillary data: An application example in the continental US, Remote
21
Sensing of Environment, 112(9): 3520-3537.
22
23
Pennisi, E. 2008. The Blue revolution, drop by drop, gene by gene. Science 320 (5873):171-173
24
doi: 10.1126/science.320.5873.171
25
26
Pittman K., Hansen M.C., Becker-Reshef I., Potapov P.V., Justice C.O. Estimating Global
27
Cropland Extent with Multi-year MODIS Data. Remote Sensing. 2010; 2(7):1844-1863.
28
29
30
Portmann, F., S. Siebert, and P. Döll, 2008. MIRCA2000 Global monthly irrigated and rainfed
31
crop areas around the year 2000: a new high-resolution data set for agricultural and hydrological
32
modelling, Global Biogeochemical Cycles, GB0003435.
33
34
Ramankutty, N., Evan, A. T., Monfreda, C., & Foley, J. A. (2008). Farming the planet: 1.
35
Geographic distribution of global agricultural lands in the year 2000. Global Biogeochem.
36
Cycles, 22, doi:10.1029/2007GB002952.
37
38
Ramankutty, N., and J. A. Foley, 1998. Characterizing patterns of global land use: An analysis of
39
global croplands data, Global Biogeochemical Cycles, 12(4): 667-685.
40
41
Rodell, M., Velicogna, I., Famiglietti, J. S., 2009, Nature 460, 999-1002,
42
doi:10.1038/nature08238.
43
44
46
Searchinger, T., Heimlich, R., Houghton, R. A., Dong, F., Elobeid, A., Fabiosa, J., Tokgoz, S.,
1
Hayes, D., & Yu, T. H. (2008). Use of U.S. Croplands for Biofuels Increases Greenhouse Gases
2
Through Emissions from Land-Use Change. Science, 319, 1238-1240.
3
4
Siebert, S., & Döll, P. (2009). Quantifying blue and green virtual water contents in global crop
5
production as well as potential production losses without irrigation. Journal of Hydrology,
6
doi:10.1016/j.jhydrol.2009.07.031.
7
8
Siebert, S., Hoogeveen, J., & Frenken, K. (2006). Irrigation in Africa, Europe and Latin America
9
- Update of the Digital Global Map of Irrigation Areas to Version 4, Frankfurt Hydrology Paper
10
05, 134 pp. Institute of Physical Geography, University of Frankfurt, Frankfurt am Main,
11
Germany and Rome, Italy.
12
13
Thenkabail, P.S., Mariotto, I., Gumma, M.K.,, Middleton, E.M., Landis, and D.R., Huemmrich,
14
F.K., 2013. Selection of hyperspectral narrowbands (HNBs) and composition of hyperspectral
15
twoband vegetation indices (HVIs) for biophysical characterization and discrimination of crop
16
types using field reflectance and Hyperion/EO-1 data. IEEE JOURNAL OF SELECTED
17
TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, Pp. 427-439,
18
VOL. 6, NO. 2, APRIL 2013.doi: 10.1109/JSTARS.2013.2252601
19
20
Thenkabail P.S., Knox
J.W., Ozdogan, M., Gumma, M.K., Congalton, R.G., Wu, Z., Milesi, C.,
21
Finkral, A., Marshall, M., Mariotto, I., You, S. Giri, C. and Nagler, P. 2012. Assessing future
22
risks to agricultural productivity, water resources and food security: how can remote sensing
23
help?. Photogrammetric Engineering and Remote Sensing, August 2012 Special Issue on Global
24
Croplands: Highlight Article.
25
26
Thenkabail P.S., Wu Z. An Automated Cropland Classification Algorithm (ACCA) for
27
Tajikistan by Combining Landsat, MODIS, and Secondary Data. Remote Sensing. 2012;
28
4(10):2890-2918.
29
30
Thenkabail, P.S., Hanjra, M.A., Dheeravath, V., Gumma, M. 2011. Book Chapter # 16: Global
31
Croplands and Their Water Use Remote Sensing and Non-Remote Sensing Perspectives. In the
32
Book entitled: Advances in Environmental Remote Sensing: Sensors, Algorithms, and
33
Applications”. Taylor and Francis Edited by Dr. Qihao Weng. Pp. 383-419.
34
35
Thenkabail P.S., Hanjra M.A., Dheeravath V., Gumma M. A. 2010. A Holistic View of Global
36
Croplands and Their Water Use for Ensuring Global Food Security in the 21st Century through
37
Advanced Remote Sensing and Non-remote Sensing Approaches. Journal Remote Sensing.
38
2(1):211-261. doi:10.3390/rs2010211. http://www.mdpi.com/2072-4292/2/1/211.
39
40
Thenkabail. P., Lyon, G.J., Turral, H., and Biradar, C.M. 2009a. Book entitled: “Remote Sensing
41
of Global Croplands for Food Security” (CRC Press- Taylor and Francis group, Boca Raton,
42
London, New York. Pp. 556 (48 pages in color). Published in June, 2009.
43
44
Thenkabail, P.S., Biradar C.M., Noojipady, P., Dheeravath, V., Li, Y.J., Velpuri, M., Gumma,
45
M., Reddy, G.P.O., Turral, H., Cai, X. L., Vithanage, J., Schull, M., and Dutta, R. 2009b. Global
46
47
irrigated area map (GIAM), derived from remote sensing, for the end of the last millennium.
1
International Journal of Remote Sensing. 30(14): 3679-3733. July, 20, 2009.
2
Thenkabail, P. S., Lyon, G. J., Turral, H., & Biradar, C. M. (2009c). Remote Sensing of Global
3
Croplands for Food Security. Boca Raton, London, New York: CRC Press- Taylor and Francis
4
Group, Published in June,.
5
6
Thenkabail, P.S., GangadharaRao, P., Biggs, T., Krishna, M., and Turral, H., 2007a. Spectral
7
Matching Techniques to Determine Historical Land use/Land cover (LULC) and Irrigated Areas
8
using Time-series AVHRR Pathfinder Datasets in the Krishna River Basin, India.
9
Photogrammetric Engineering and Remote Sensing. 73(9): 1029-1040. (Second Place Recipients
10
of the 2008 John I. Davidson ASPRS President’s Award for Practical papers).
11
12
Thenkabail, P.S., Biradar C.M., Noojipady, P., Cai, X.L., Dheeravath, V., Li, Y.J., Velpuri, M.,
13
Gumma, M., Pandey., S. 2007b. Sub-pixel irrigated area calculation methods. Sensors Journal
14
(special issue: Remote Sensing of Natural Resources and the Environment (Remote Sensing
15
Sensors Edited by Assefa M. Melesse). 7:2519-2538.
16
http://www.mdpi.org/sensors/papers/s7112519.pdf.
17
18
Thenkabail, P.S., Schull, M., Turral, H. 2005. Ganges and Indus River Basin Land Use/Land
19
Cover (LULC) and Irrigated Area Mapping using Continuous Streams of MODIS Data. Remote
20
Sensing of Environment. Remote Sensing of Environment, 95(3): 317-341.
21
22
Thenkabail, P.S., Enclona, E.A., Ashton, M.S., Legg, C., Jean De Dieu, M., 2004. Hyperion,
23
IKONOS, ALI, and ETM+ sensors in the study of African rainforests. Remote Sensing of
24
Environment, 90:23-43.
25
26
Tillman, D., C. Balzer, J. Hill, and B. L. Befort, 2011. Global food demand and the sustainable
27
intensification of agriculture, Proceedings of the National Academy of Sciences of the United
28
States of America, November 21, doi: 10.1073/pnas.1116437108.
29
30
Turral, H., Svendsen, M., & Faures, J. (2009). Investing in irrigation: Reviewing the past and
31
looking to the future. Agricultural Water Management, doi:101.1016/j.agwt.2009.07.012.
32
33
UNDP (2012). Human Development Report 2012: Overcoming Barriers: Human Mobility and
34
Development. New York: United Nations.
35
36
Velpuri, M., Thenkabail, P.S., Gumma, M.K., Biradar, C.B., Dheeravath, V., Noojipady, P.,
37
Yuanjie, L., 2009. Influence of Resolution or Scale in Irrigated Area Mapping and Area
38
Estimations. Photogrammetric Engineering and Remote Sensing (PE&RS). 75(12): December
39
2009 issue.
40
41
Vinnari, M., & Tapio, P. (2009). Future images of meat consumption in 2030. Futures,
42
doi:10.1016/j.futures.2008.11.014.
43
48
Wada, Y., L. P. H. van Beek, and M. F. P. Bierkens (2012), Nonsustainable groundwater
1
sustaining irrigation: A global assessment, Water Resour. Res., 48, W00L06,
2
doi:10.1029/2011WR010562.
3
Wardlow, B. D., & Egbert, S. L. (2008). Large-area crop mapping using time-series MODIS 250
4
m NDVI data: An assessment for the U.S. Central Great Plains. Remote Sensing of Environment,
5
112, 1096-1116.
6
7
Wardlow, B. D., Egbert, S. L., & Kastens, J. H. (2007). Analysis of time-series MODIS 250 m
8
vegetation index data for crop classification in the U.S. Central Great Plains. Remote Sensing of
9
Environment, 108, 290-310.
10
11
Wardlow, B. D., Kastens, J. H., & Egbert, S. L. (2006). Using USDA crop progress data for the
12
evaluation of greenup onset date calculated from MODIS 250-meter data. Photogrammetric
13
Engineering and Remote Sensing, 72, 1225-1234.
14
15
Wu, Z., Thenkabail, P.S., Zakzeski, A., Mueller, R., Melton, F., Rosevelt, C., Dwyer, J.,
16
Johnson, J., Verdin, J. P., 2014a. Seasonal cultivated and fallow cropland mapping using modis-
17
based automated cropland classification algorithm. J. Appl. Remote Sens. 0001;8(1):083685.
18
doi:10.1117/1.JRS.8.083685.
19
20
Wu, Z., Thenkabail, P.S., and Verdin, J. 2014b. An automated cropland classification algorithm
21
(ACCA) using Landsat and MODIS data combination for California. Photogrammetric
22
Engineering and Remote Sensing. Vol. 80(1): 81-90.
23
24
Yu, L., Wang, J., Clinton, N., Xin, Q., Zhong, L., Chen, Y., and Gong, P. 2013. International
25
Journal of Digital Earth. FROM-GC: 30 m global cropland extent derived through multisource
26
data integration, International Journal of Digital Earth ,DOI:10.1080/17538947.2013.822574.
27
28
29
30
31
32
33
34
35
36
... A cropland extent map is a vital component of land system studies (Verburg et al. 2013). Besides, crop maps with high reliability are the basis of accurate agricultural statistics estimation (Gallego et al. 2012;Gallego, Carfagna and Baruth 2010;Gallego et al. 2014), stratification purposes (Boryan and Yang 2013), crop yield prediction (Kolotii et al. 2015;, drought risk assessment Kussul et al. 2010;Kussul, Shelestov and Skakun 2011) and food security studies (Foley et al. 2011;Teluguntla et al. 2015;Thenkabail et al. 2010;Matejicek and Kopackova 2010). ...
... These products provide a fundamental understanding of the spatial distribution pattern and characteristics of croplands. However, the coarse spatial resolution intensifies the mixed pixel effect, that different land cover types mess up within a pixel, leading to significant errors in estimating crop areas and ignoring agriculture fields smaller than a cell size (Teluguntla et al. 2015;Thenkabail et al. 2010). ...
... Temporary layer Processing step Judgment conditions Output file Content courtesy of Springer Nature, terms of use apply. Rights reserved Global Mapping of Irrigation Areas produced by Meier et al. 47 with a spatial resolution of 1 km and a strong correlation with statistical data (r = 0.84); GRIPC, generated using a supervised classification method with remote sensing, climate, and agricultural inventory data at 500 m resolution and had an overall accuracy of 69% 3 ; GFSAD (Global Food Security-support Analysis Data) which was a NASA-funded project to provide high-resolution global cropland data and their water use to sustain global food security 48 They were generated through a human-computer interaction method with Landsat imagery and had high overall accuracies of over 90% 50,51 . In the NLCD, paddy fields (class code: 11) which indicated rice and other paddy crops like lotus root were considered as irrigated croplands in this study. ...
Article
Full-text available
As a routine agricultural practice, irrigation is fundamental to protect crops from water scarcity and ensure food security in China. However, consistent and reliable maps about the spatial distribution and extent of irrigated croplands are still unavailable, impeding water resource management and agricultural planning. Here, we produced annual 500-m irrigated cropland maps across China for 2000–2019, using a two-step strategy that integrated statistics, remote sensing, and existing irrigation products into a hybrid irrigation dataset. First, we generated intermediate irrigation maps (MIrAD-GI) by fusing the MODIS-derived greenness index and statistical data. Second, we collected all existing available irrigation maps over China and integrated them with MIrAD-GI into an improved series of annual irrigation maps, using constrained statistics and a synergy mapping method. The resultant maps had moderate overall accuracies (0.732~0.819) based on nationwide reference ground samples and outperformed existing irrigation products by inter-comparison. As the first of this kind in China, the annual maps delineated the spatiotemporal pattern of irrigated croplands and could contribute to sustainable water use and agricultural development.
... The analysis is based on a combination of remotely sensed and atmospheric data detailed in Table 2 Engine platform [82] using the GFSAD1000 gridded cropland product [83]; then, the mean value of each variable is extracted for each day of the study period over SEA. ...
Article
Full-text available
In a world where climate change, population growth, and global diseases threaten economic access to food, policies and contingency plans can strongly benefit from reliable forecasts of agricultural vegetation health. To inform decisions, it is also crucial to quantify the forecasting uncertainty and prove its relevance for food security. Yet, in previous studies both these aspects have been largely overlooked. This paper develops a methodology to anticipate the agricultural Vegetation Health Index (VHI) while making the underlying prediction uncertainty explicit. To achieve this aim, a probabilistic machine learning framework modelling weather and climate determinants is introduced and implemented through Quantile Random Forests. In a second step, a statistical link between VHI forecasts and monthly food price variations is established. As a pilot implementation, the framework is applied to nine countries of South-East Asia (SEA) with consideration of national monthly rice prices. Model benchmarks show satisfactory accuracy metrics, suggesting that the probabilistic VHI predictions can provide decision-makers with reliable information about future cropland health and its impact on food price variation weeks or even months ahead, albeit with increasing uncertainty as the forecasting horizon grows. These results - ultimately allowing to anticipate the impact of weather shocks on household food expenditure - contribute to advancing the multidisciplinary literature linking vegetation health, probabilistic forecasting models, and food security policy.
... In light green are the locations of agricultural crop areas identified by satellite imagery at 30-m resolution, with the map obtained from https:// cropl ands. org/ app/ map 30,31 . Dashed grey line separate western and eastern collection sites. ...
Article
Full-text available
The fall armyworm, Spodoptera frugiperda (J.E. Smith) is native to the Americas and a major pest of corn and several other crops of economic importance. The species has characteristics that make it of particular concern as an invasive pest, including broad host range, long-distance migration behavior, and a propensity for field-evolved pesticide resistance. The discovery of fall armyworm in western Africa in 2016 was followed by what was apparently a remarkably rapid spread throughout sub-Saharan Africa by 2018, causing economic damage estimated in the tens of billions USD and threatening the food security of the continent. Understanding the history of the fall armyworm invasion of Africa and the genetic composition of the African populations is critical to assessing the risk posed to different crop types, the development of effective mitigation strategies, and to make Africa less vulnerable to future invasions of migratory moth pests. This paper tested and expanded on previous studies by combining data from 22 sub-Saharan nations during the period from 2016 to 2019. The results support initial descriptions of the fall armyworm invasion, including the near absence of the strain that prefers rice, millet, and pasture grasses, while providing additional evidence that the magnitude and extent of FAW natural migration on the continent is more limited than expected. The results also show that a second entry of fall armyworm likely occurred in western Africa from a source different than that of the original introduction. These findings indicate that western Africa continues to be at high risk of future introductions of FAW, which could complicate mitigation efforts.
Article
Full-text available
Map of croplands is one of the information layers required in the efficient management of these lands. Having such maps makes it possible to monitor agricultural fields during the growing season continuously. In this study, a solution to produce a map of Shahrekord’s agricultural lands in two agricultural and non-agricultural classes is presented using the time series of different extracted indices from Sentinel-2 images. Since the use of large data sources is one of the obstacles to the development of methods based on the time series of satellite images, the Google Earth engine processing platform has been used in this study. The proposed method is based on integrating supervised pixel-based classification results with segmentation results. First, training data for supervised classification is provided in a rigorous refining process without the need for collecting data from field surveys or interpretation of high-resolution satellite images. Then, by calculating the separability of the two target classes in the time series of each index, the optimal indices are selected. Finally, by combining the results of segmentation and classification methods based on the votes obtained from the classification results, the agricultural or non-agricultural class is assigned to each of the image segments. In addition to incorporating spatial information including edges and spatial proximity, this method has been able to improve the noise and porous results of pixel-based classification and has increased the overall accuracy of the final map from 90.7% to 96.05%. Also, user accuracy of both agricultural and non-agricultural classes shows an improvement of 3.27 and 7.97%, respectively.
Chapter
There is a wide variety of global thematic Land Use Cover (LUC) datasets characterizing agricultural covers. Most of them focus on cropland areas, providing information on their extent or the percentage of cropland cover on the ground. In some cases, the focus is more specific and they provide information on cropland irrigation practices. In other cases, specific maps charting the extension of different crops are also available. In this chapter, we review 8 different datasets with a spatial resolution of at least 1 km. There are many other datasets characterizing agricultural covers at coarser resolutions, such as the Historic Croplands Dataset, GMRCA or GIAM. Their coarse resolution hampers their potential application in practice, which is why they are not described in detail in this chapter. Nor do we analyse FROM-GC, a dataset mapping the extent of global cropland at 30 m, because it is not currently accessible. GFSAD30 has the highest resolution of all the datasets reviewed (30 m). It also provides some of the most up-to-date information (2015). However, it only charts the extent of cropland. As part of an associated project, GFSAD1KCD and GFSAD1KCM characterize cropland areas in 9 and 7 categories respectively at 1 km for 2010. They provide information on the irrigation status of the crops. GFSAD1KCD and GFSAD1KCM were obtained from data fusion. This method is commonly used in the production of many of the cropland datasets reviewed: IIASA-IFPRI cropland map, Global Synergy Cropland Map, Unified Cropland Layer (UCL) and ASAP Land Cover Masks. The IIASA-IFPRI (2005) and ASAP maps provide information on the proportion of cropland at a spatial resolution of 1 km. ASAP also includes a map on rangeland covers, and as such is the only dataset described in this chapter that maps a cover other than croplands. The Global Synergy Cropland Map (2010) and the Unified Cropland Layer (2014) also map cropland proportions, although they have been produced at higher spatial resolutions: 500 and 250 m respectively. The Global Cropland Extent product maps the extent of cropland at 250 m based on imagery from 2000-2008. Although thematically limited, this dataset is less affected by time variability, as it is based on imagery taken over a long period (8 years). Finally, GRIPC maps the extent of three types of cropland area (irrigated, rainfed and paddy crops) at 500 m for 2005.
Article
Full-text available
Land use and land cover (LULC) mapping are of paramount importance to monitor and understand the structure and dynamics of the Earth system. One of the most promising ways to create accurate global LULC maps is by building good quality state-of-the-art machine learning models. Building such models requires large and global datasets of annotated time series of satellite images, which are not available yet. This paper presents TimeSpec4LULC (https://doi.org/10.5281/zenodo.5913554; Khaldi et al., 2022), a smart open-source global dataset of multispectral time series for 29 LULC classes ready to train machine learning models. TimeSpec4LULC was built based on the seven spectral bands of the MODIS sensors at 500 m resolution, from 2000 to 2021, and was annotated using spatial–temporal agreement across the 15 global LULC products available in Google Earth Engine (GEE). The 22-year monthly time series of the seven bands were created globally by (1) applying different spatial–temporal quality assessment filters on MODIS Terra and Aqua satellites; (2) aggregating their original 8 d temporal granularity into monthly composites; (3) merging Terra + Aqua data into a combined time series; and (4) extracting, at the pixel level, 6 076 531 time series of size 262 for the seven bands along with a set of metadata: geographic coordinates, country and departmental divisions, spatial–temporal consistency across LULC products, temporal data availability, and the global human modification index. A balanced subset of the original dataset was also provided by selecting 1000 evenly distributed samples from each class such that they are representative of the entire globe. To assess the annotation quality of the dataset, a sample of pixels, evenly distributed around the world from each LULC class, was selected and validated by experts using very high resolution images from both Google Earth and Bing Maps imagery. This smartly, pre-processed, and annotated dataset is targeted towards scientific users interested in developing various machine learning models, including deep learning networks, to perform global LULC mapping.
Article
The crop detection and classification using satellite imagery can be helpful in crop management. Therefore, it is needed to be explored for effective crop detection and classification. In this study, the phenological metrics derived from dense Landsat Normalized Difference Vegetation Index (NDVI) time-series from January 2017 to June 2018 were used to classify sugarcane crop type. The Savitzkey–Golay method was employed as curve fitting method for smoothening and filtering of the pre-processed NDVI time-series data. Thereafter, the phenological metrics extracted from TIMESAT were used to classify planted and ratoon sugarcane using the decision-tree approach. The overall accuracy was 84.5%, and the kappa coefficient was 0.7. The results showed that the classification method was consistent and robust which has the potential to increase the classification accuracy by the use of better resolution satellite imagery time-series. Moreover, the phenological metrics have the potential to explain the spatial distribution of agricultural pattern, use of crop varieties and management practices that can be useful in better understanding of agricultural practices in the region along with crop diversification and crop yield modelling.
Article
Full-text available
Irrigation is widely implemented in China to enhance grain production and ensure food security. Spatiotemporal information on irrigation is critically important, but the existing global and regional irrigation products have a coarse spatial resolution, low accuracy, and short temporal coverage, which is a knowledge gap to be filled. Generating explicit and accurate information on the spatial and temporal extent of irrigation is essential to underpin and facilitate water resource allocation and management, as well as to understand catchment hydrology, and evaluate irrigation infrastructure investment. We proposed a two-step strategy to map annual irrigated areas at 500 m resolution in China from 2000 to 2019. We first generated initial irrigation maps using MODIS data and statistical data (MIrAD-GI). Then we combined MIrAD-GI, the existing irrigation maps, and land use/cover products with irrigation information, into an improved series of annual irrigation maps (IrriMap_Syn) with constrained statistics. Afterwards, pixel-wise accuracy assessment for IrriMap_Syn was conducted in four representative regions in three years. The resultant irrigation dataset performed well with a mean overall accuracy of 0.89 and a kappa coefficient of 0.82. According to our annual irrigation maps, the total irrigated area in China increased by 24.8%, from 52.8 million ha in 2000–65.9 million ha in 2019, at a rate of 690,000 ha per year. Spatially, irrigated croplands increased substantially in North China, especially in Xinjiang and Heilongjiang. Land reclamation largely contributed to the increase in irrigation in North China. A growing population and improved irrigation infrastructure also promoted the increase in irrigated areas. As the first of its kind in the country, our spatially explicit maps of irrigated croplands advance our understanding of the spatiotemporal pattern of irrigation dynamics in China and are expected to contribute to sustainable water resource management and irrigation strategies in the face of climate change.
Article
Full-text available
Data on irrigation patterns and trends at field-level detail across broad extents are vital for assessing and managing limited water resources. Until recently, there has been a scarcity of comprehensive, consistent, and frequent irrigation maps for the US. Here we present the new Landsat-based Irrigation Dataset (LANID), which is comprised of 30 m resolution annual irrigation maps covering the conterminous US (CONUS) for the period of 1997–2017. The main dataset identifies the annual extent of irrigated croplands, pastureland, and hay for each year in the study period. Derivative maps include layers on maximum irrigated extent, irrigation frequency and trends, and identification of formerly irrigated areas and intermittently irrigated lands. Temporal analysis reveals that 38.5×106 ha of croplands and pasture–hay has been irrigated, among which the yearly active area ranged from ∼22.6 to 24.7×106 ha. The LANID products provide several improvements over other irrigation data including field-level details on irrigation change and frequency, an annual time step, and a collection of ∼10 000 visually interpreted ground reference locations for the eastern US where such data have been lacking. Our maps demonstrated overall accuracy above 90 % across all years and regions, including in the more humid and challenging-to-map eastern US, marking a significant advancement over other products, whose accuracies ranged from 50 % to 80 %. In terms of change detection, our maps yield per-pixel transition accuracy of 81 % and show good agreement with US Department of Agriculture reports at both county and state levels. The described annual maps, derivative layers, and ground reference data provide users with unique opportunities to study local to nationwide trends, driving forces, and consequences of irrigation and encourage the further development and assessment of new approaches for improved mapping of irrigation, especially in challenging areas like the eastern US. The annual LANID maps, derivative products, and ground reference data are available through https://doi.org/10.5281/zenodo.5548555 (Xie and Lark, 2021a).
Article
Full-text available
Identification of the onset of vegetation greenup is a key-factor in characterizing and monitoring vegetation dynamics over large areas. However, the relationship between greenup onset dates estimated from satellite imagery and the actual growth stage of vegetation is often unclear. Herein, we present an approach for comparing pixel-level dnset dates to regional planting and emergence information for agricultural crops, with the goal of drawing reliable conclusions regarding the physical growth stage of the vegetation of interest at the time of greenup onset. To accomplish this, we calculated onset of greenup using modis 250 m, 16-day composite NDVl time series data for Kansas for 2001 and a recently proposed methodology for greenup detection. We then evaluated the estimated greenup dates using the locations of 1,417 large field sites that were planted to corn, soybeans, or sorghum in 2001, in conjunction with United States Department of Agriculture (usda) weekly crop progress reports containing crop planting and emergence percentage estimates. Average greenup onset dates calculated for the three summer crops showed that the dates were consistent with the relative planting order of corn, sorghum, and soybeans across the state. However, the influence of pre-crop vegetation (weeds and "volunteer" crops) introduced an early bias for the greenup onset dates calculated for many field sites. This pre-crop vegetation signal was most pronounced for the later planted summer crops (soybeans and sorghum) and in areas of Kansas that receive higher annual precipitation. The most reliable results were obtained for corn in semi-arid western Kansas, where pre-crop vegetation had considerably less influence on the greenup onset date calculations. The greenup onset date calculated for corn in western Kansas was found to occur 23 days after 50 percent of the crop had emerged. Corn’s greenup onset was detected, on average, at the agronomic stage where plants are 15 to 45 cm (6 to 18 inches) tall and the crop begins its rapid growth.
Article
Full-text available
Increasing drought occurrences and growing populations demand accurate, routine, and consistent cultivated and fallow cropland products to enable water and food security analy- sis. The overarching goal of this research was to develop and test automated cropland classi- fication algorithm (ACCA) that provide accurate, consistent, and repeatable information on seasonal cultivated as well as seasonal fallow cropland extents and areas based on the Moderate Resolution Imaging Spectroradiometer remote sensing data. Seasonal ACCA develop- ment process involves writing series of iterative decision tree codes to separate cultivated and fallow croplands from noncroplands, aiming to accurately mirror reliable reference data sources. A pixel-by-pixel accuracy assessment when compared with the U.S. Department of Agriculture (USDA) cropland data showed, on average, a producer’s accuracy of 93% and a user’s accuracy of 85% across all months. Further, ACCA-derived cropland maps agreed well with the USDA Farm Service Agency crop acreage-reported data for both cultivated and fallow croplands with R-square values over 0.7 and field surveys with an accuracy of ≥95% for cultivated croplands and ≥76% for fallow croplands. Our results demonstrated the ability of ACCA to generate cropland products, such as cultivated and fallow cropland extents and areas, accurately, auto- matically, and repeatedly throughout the growing season.
Article
Full-text available
Until recently, advanced very high-resolution radiometer (AVHRR) observations were the only viable source of data for global land cover mapping. While many useful insights have been gained from analyses based on AVHRR data, the availability of moderate resolution imaging spectroradiometer (MODIS) data with greatly improved spectral, spatial, geometric, and radiometric attributes provides significant new opportunities and challenges for remote sensing-based land cover mapping research. In this paper, we describe the algorithms and databases being used to produce the MODIS global land cover product. This product provides maps of global land cover at 1-km spatial resolution using several classification systems, principally that of the IGBP. To generate these maps, a supervised classification methodology is used that exploits a global database of training sites interpreted from high-resolution imagery in association with ancillary data. In addition to the IGBP class at each pixel, the MODIS land cover product provides several other parameters including estimates for the classification confidence associated with the IGBP label, a prediction for the most likely alternative class, and class labels for several other classification schemes that are used by the global modeling community. Initial results based on 5 months of MODIS data are encouraging. At global scales, the distribution of vegetation and land cover types is qualitatively realistic. At regional scales, comparisons among heritage AVHRR products, Landsat TM data, and results from MODIS show that the algorithm is performing well. As a longer time series of data is added to the processing stream and the representation of global land cover in the site database is refined, the quality of the MODIS land cover product will improve accordingly.