a knowledge-based automated cropland mapping … · a knowledge-based automated cropland mapping...

U.S. Geological Survey

U.S. Department of Interior

Knowledge Based Cropland Layer Algorithm Derived Cropland Layer

Prasad S. Thenkabail1 and Zhouting Wu2

1, 3, 4 = U.S. Geological Survey (USGS), USA; 2 = Northern Arizona University (NAU), USA

Landsat Science Meeting. March 1-3, 2011, held at Phoenix, AZ, USA

125

145

165

185

205

1 2 3 4 5 6 7 8 9 10 11 12

MO

DIS

Sc

ale

d N

DV

I

Month

Irrigated_algorithm

Rainfed-Algorithm

Others-Algorithm125

145

165

185

205

1 2 3 4 5 6 7 8 9 10 11 12

MO

DIS

Sc

ale

d N

DV

I

Month

Irrigated-Knowledge

Rainfed-Knowledge

Others-Knowledge

A Knowledge-based Automated Cropland Mapping Algorithm

using Advanced Remote Sensing Methods and Approaches

U.S. Department of the Interior


Importance, Need, and Scope



A Knowledge-based Automated Cropland Mapping Algorithm Importance, Need, and Scope for an Automated Cropland Algorithm

1. Croplands:

A. Rapid and repeatable (e.g., year after year, season after season) mapping of

croplands requires an automated algorithm;

Accuracy of the algorithm will depend on an accurate knowledge layer of croplands;

B. Currently absence of such accurate algorithms make cropland mapping tedious

and\or inaccurate;

C. Need for advance methods (e.g., automated algorithms), and approaches (e.g.,

multiple remote sensing data fusion) is critical;

D. Accurate cropland mapping using advanced remote sensing enables: (a) frequent (e.g.,

yearly, seasonal) update of agricultural statistics, (b) spatial view of croplands distribution

and their changes over space and time;

2. Water use by croplands:

Helps us determine blue water use and green water use accurately;

3. Food security:

If we can achieve 1 and 2 we can make significant contribution towards a food secure world

Blue water use: water used

by irrigated areas coming

from reservoirs, lakes, rivers,

deep aquifer water, and even

direct rainfall over irrigated

areas

Green water use: water

used by rainfed cropland

areas coming from rainfed

croplands and soil

moisture stored in

unsaturated zoneCroplands: 1.53 billion hectares (yr: 2000)



1. Resolution or Scale: Resolution too coarse;

2. Area fractions: uncertainties in determining fractional

areas;

3. Definition issues (e.g., does supplemental irrigation fall

into irrigated areas?);

4. Type of data used (e.g., traditional vs. remote sensing);

5. Methodologies: complexity or simplicity of methods;

6. Data sharing issues: reluctance to share data due to

vested interests;

7. Inadequate accounting of minor irrigation (e.g., ground

water, small reservoirs, tanks);

Uncertainties in Existing Cropland Data Causes of Uncertainties

CBIP

CBIP

CB

IP



30m

Landsat

Data

highlights

small

water

bodies

such as

tanks and

small

reservoirs

Uncertainties in Existing Cropland Data Minor irrigation not properly Accounted in Coarse Resolution



Finer resolution (e.g., Landsat) in combination with temporal resolution (e.g.,

MODIS), secondary data, and in-situ data will bring: (a) crop type\dominance; (b)

precise location of crops, (c) watering methods (e.g., rainfed, irrigated), (d)

cropping intensity……..leading to an advanced agricultural monitoring

system…..that allows precise water use assessments.

Overcoming Uncertainties in Existing Cropland Data Is Landsat Data Panacea?



Overarching Goal




Overarching Goal

The Overarching Goal

of this study is to develop

an Automated Algorithm, that is Knowledge-Based,

to Map Irrigated and Rainfed Cropland Areas

of a Country, Continent, and The World using Advanced Remote Sensing Methods and Approaches

This Presentation will Focus on a Country and

Demonstrate the Working of such an Algorithm



A New Strategy for Mapping Croplands

Accurately and Rapidly




Strategy in Developing the Algorithm1. Establish crop calendars: Divide the world into distinct zones cropping

calendars. Develop algorithm for each zone separately;

2. Multi-data Fusion: Develop mega-file data cube that makes use of multiple

sources of remotely sensed data and numerous other data such as (a)

Landsat 30m, (b) MODIS NDVI MVC monthly composites, (c) Suite of

Secondary data (e.g., elevation, slope, temperature, rainfall), (d) higher

resolution imagery as ground-truth, and (e) in-situ data;

3. Knowledge Layer of Croplands: Obtain and\or generate an accurate

knowledge layer of irrigated croplands, rainfed croplands, and other LCLU

classes;

4. Develop automated algorithm for a country and\or a region;

5. Develop automated cropland algorithm for the entire world by integrating

suite of cropland algorithm of different countries and\or regions;

6. Test the algorithm on independent datasets; and

7. Establish accuracies and errors.

………….today, we will illustrate the development, working, and testing of this

cropland algorithm for one Central Asian Country.



0.4 billion Hectares of irrigated croplands at the end of the last millennium

Global Irrigated Areas: Spatial DistributionDevelop Algorithm for Automated Cropland Mapping based on

countries or regions or continents and then put them all together

Our focus area

Global Irrigated Areas + Rainfed AreasSpatial Distribution of Global Croplands

1.53 billion Hectares of total croplands at the end of the last millennium



Our focus area

Pakistan

India

USA

China

Irrigated AreasDominant Countries

28.5% of the

Global AIA

3.5% of the

Global AIA

32.5% of the

Global AIA

5.2% of the

Global AIA

Note: AIA = annualized irrigated areas.



125

135

145

155

165

175

185

195

1 2 3 4 5 6 7 8 9 10 11 12

Scale

d N

DV

I

Month

Irrigated-Knowledge (MODIS 2009)

Rainfed-Knowledge (MODIS 2009)

138

143

148

153

158

163

168

173

178

1 2 3 4 5 6 7 8 9 10 11 12

AV

HR

R S

cale

d N

DV

I 1999

Month

Irrigated (dominant)

Rainfed (dominant)


Coarse vs. Fused Multi-Resolution

1 km

1 km

30 m

250 m

MODIS 2010

MODIS 2010



0

25

50

75

0 10 20

HY675

HY

91

0

barley

wheat

0

10

20

30

40

50

400 460 520 580 640 700 760 820 880 940 1000

Wavelength (nm)

Ref

lect

an

ce (

per

cen

t)

Y. sec. Forest

P. forest

Slash&Burn

Raphia palm

Bamboo

P. Africana

0

0.1

0.2

0.3

0.4

0.5

300 400 500 600 700 800 900 1000

wavelength (nanometers)

refl

ecta

nce f

acto

r

wheat (64)

barley (44)

fallow (9)

higher reflectance of barley throughout visible spectrum

as a result of pigmentation. Barley greenish

brown/seafoam color compared to deep green of wheat.

peak NIR reflectance around

910 nanometers.

absorption maxima around

680 nanometers

moisture sensitive and biomass related

trough centered around 980

nanometers

erectophile (65 degrees) structure results in steep slopes

in NIR reflectance from 740-nm to 940-nm

50

75

100

125

10 30 50

TM3

TM

4

barley

wheat

A Knowledge-based Automated Cropland Mapping Algorithm Landsat compared to hyperspectral Narrow-bands in Discriminating Wheat vs. Barley

Hyperion

Data Cube

0

10

20

30

40

50

400 500 600 700 800 900 1000Wavelength (nm)

Refl

ecta

nce (

percen

t)

Y. sec. Forest

P. forest

Slash&Burn

Raphia palm

Bamboo

P. Africana

0

10

20

30

40

50

400 900 1400 1900 2400

Wavelength (nm)

Refl

ecta

nce (

percen

t)

Y. sec. Forest

P. forest

Slash&Burn

Raphia palm

Bamboo

P. Africana

0

10

20

30

40

50

400 900 1400 1900 2400Wavelength (nm)

Refl

ecta

nce (

percen

t)

Y. sec. Forest

P. forest

Slash&Burn

Raphia palm

Bamboo

P. Africana 0

10

20

30

40

50

400 900 1400 1900 2400Wavelength (nm)

Refl

ecta

nce (

percen

t)

Y. sec. Forest

P. forest

Slash&Burn

Raphia palm

Bamboo

P. Africana

IKONOS: Feb. 5, 2002 (hyper-spatial)

ALI: Feb. 5, 2002 (multi-spectral)

ETM+: March 18, 2001 (multi-spectral)

Hyperion: March 21, 2002 (hyper-spectral)

Comparison of Hyperspectral Data with Data from Other Advanced Sensors

Hyperspectral, Hyperspatial, and Advanced Multi-spectral Data



0

0.2

0.4

0.6

0.8

1

1 49 97 145

193

241

289

337 17 65 11

316

120

925

730

535

3

Julian Date

ND

VI V

alu

e

Rabi, 2001 Summer, 2001 Khariff, 2001

Rabi, 2002 Summer, 2002 Khariff, 2002

Irrigated area

class #21

Need for Temporal Data in Cropland Mapping

MODIS Time-Series for Ganges River basin



0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Irrigated cropland

-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Months

Tundra

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Rainfed cropland

0

0.1

0.2

0.3

0.4

0.5

0.6

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Months

Forest

HMFDC(Mega-file data-cube)

LandsatModis Jan

Modis Feb

Modis Dec.

…………

Lossless compression

Typical spectral profile

illustrations of different land

cover classes derived from

MODIS 250m NDVI time-series

A Knowledge-based Automated Cropland Mapping Algorithm Mega-file Data Cube (MFDC) showing Fusion of Landsat and MODIS Datasets



Datasets used to Generate

Knowledge Layer (year 2005)



A Knowledge-based Automated Cropland Mapping Algorithm Datasets used to produce Knowledge Layer on Croplands for Tajikistan

1. Landsat 30 m Data: Landsat GLS2005 (nominal 2005) dataset;

2. MODIS NDVI MVC, monthly: MODIS NDVI MVC monthly data for 2005;

3. Very high resolution data as “groundtruth”: sub-meter to 4 meter (very

high resolution data) for nominal 2005 as “groundtruth”;

4. In-Situ data as “groundtruth”: in-situ data gathered from various sources

as “groundtruth”.

0

0.2

0.4

0.6

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Months

Irrigated cropland

Wheat

Cotton



Croplands versus non-cropland identification based

on Tassel cap bi-spectral plots and Very High

Resolution Imagery

Very high resolution

imagery: sub-meter to

6m (e.g., quickbird,

IKONOS, Geoeye,

Rapideye)

WaterWetland

Greenness

Brightness

Wetness

Class# 21

Class# 37

Class# 26

Class# 15

Class# 188

Class# 164Barren

SnowCropland

Forest

Very high resolution

imagery: sub-meter to

6m (e.g., quickbird,

IKONOS, Geoeye,

Rapideye)

Tassel cap bi-spectral plot

A Knowledge-based Data Layer Derived by Classifying the MFDC Identifying and Labeling the 199 Information Classes: Bispectral Plots

Multiple measures used in

algorithms to identify and label

agricultural cropland classes

-0.1

-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Months

Class146Class139Class124Class180Class151Class171Class167

Tundra

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Months

Class37

Class39

Class43

Irrigated

cropland,

double crop

-0.9

-0.8

-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Months

Class13Class15Class16

Water body

Snow

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

1 2 3 4 5 6 7 8 9 10 11 12

ND

VI

Months

Class49

Class65

Class69

Irrigated

cropland,

single crop

23 out of 199 classes identified and illustrated here

A Knowledge-based Data Layer Derived by Classifying the MFDC Identifying and Labeling the 199 Information Classes: MODIS Temporal NDVI Plots for Certain Classes

01 Irrigated, conjunctive use, cotton-wheat-rice-dominant, large scale, double crop

02 Rainfed, wheat-barley-dominant, large scale, single crop

03 Shrub/rangeland dominate with rainfed croplands

04 Shrublands-grasslands

05 Mixed, shrublands, grasslands, urban built-up

06 Forest

07 Tundra

08 Wetlands

09 Barren or sparcely vegetated

10 Snow

11 Waterbodies

0

5

10

15

20

25

30

35

40

45

50

0 5 10 15 20 25 30 35 40 45

La

nd

sa

t b

4 (

Re

fle

cta

nc

e %

)

Landsat b3 (Reflectance %)

Wetlands



WheatRice Cotton

Snow

Shrublands

Forest

Tundra

Waterbodies

Grasslands

Barren or sparcely

vegetated

A Knowledge-based Data Layer Derived by Classifying the MFDC Identifying and Labeling the 199 Information Classes: Final 11 Classes with typical “groundtruth” view



1 irrigated class

1 rainfed class.

Note: there is also highly

fragmented rainfed class

mixed with other LCLUC.

A Knowledge-based Data Layer Derived by Classifying the MFDC

Final 11 unique knowledge-based classes of Tajikistan

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

1 2 3 4 5 6 7 8 9 10 11 12

Mo

dis

ND

VI

Months

Rainfed, wheat-dominant (e.g. class #56)

Irrigated, cotton-dominant (e.g. class #37)

Cropping intensities

Single crop Double crop



Irrigated, double crop, cotton

Rainfed, single crop, wheat dominant

A Knowledge-based Data Layer Derived by Classifying the MFDC Identifying and Labeling 199 Information Classes: “Groundtruth” Data used in Conjunction with MODIS NDVI Time-series



Generating Algorithm Derived Cropland Layer

using same (2005) Landsat and MODIS Datasets

as Knowledge Layer



Landsat false color composite of Band 4, 3, 5: terrain view

An Automated Cropland Mapping Algorithm Applied to Same Dataset used to Derive Knowledge Layer

Datasets used: Landsat 2005

Landsat scaled NDVI: biomass, LAI

Landsat band 3 reflectance (%): chlorophyll absorption Landsat band 5 reflectance (%): moisture sensitivity




Datasets used: Landsat Thermal Data

Landsat thermal emissivity



Jan Feb Mar Apr

May Jun Jul Aug

Sep Oct Nov Dec


Datasets used: MODIS NDVI monthly MVC. Note: irrigated areas most discernable during July-November

Irrigated areas

begin to be distinct

Irrigated areas

become very distinct




Datasets used: SRTM 90 m Elevation

Note: most irrigated areas in elevation < 550 m




Datasets used: SRTM derived Slopes

Note: most irrigated areas in slopes < 3%



Algorithm for Irrigated Areas

using Data for the Years Same (2005)as Knowledge Layer



Automated Mapping of Irrigated Areas of Tajikistan

Algorithm 1a

Slope, SRTM derived MODIS Yearly Total NDVI

≤1.5 % >1.5 % <1920 ≥1920

Irrigated Others



Slope, SRTM

derived

MODIS Yearly Total NDVI

>1.5

and

≤2.5 %

>2.5 % <1950 ≥1950

Irrigated Others

Elevation,

from SRTM

≤510 m >510 m


Algorithm 1b



Slope, SRTM

derived

MODIS Yearly Total NDVI

>2.5

and

≤7.0 %

>7.0 % <2000 ≥2000

Irrigated Others

Elevation

from SRTM

≤510 m >510 m


Algorithm 1c



Landsat B5 (moisture

sensitivity)Landsat B3 (chlorophyll

absorption)

>15 and ≤25 %

reflectance

≤ 15 or >25 %

reflectance≤16 %

reflectance

>16 %

reflectance

Irrigated Others

Elevation,

SRTM derived

≤900 m >900 m


Algorithm 2a



Landsat NDVI Landsat B3 (chlorophyll

absorption)

≤160 >160 ≤16 % reflectance >16 %

reflectance

Irrigated Others

Elevation, SRTM

derived

≤900 m >900 m


Algorithm 2b



Landsat B6 (thermal

temperature)Landsat B3 (chlorophyll

absorption)

>140

and

≤160 in

degree

Celsius

≤140 or

>160 in

degree

Celsius

≤16 %

reflectance

>16 %

reflectance

Irrigated Others

Elevation,

SRTM derived

≤900 m >900 m


Algorithm 2c



MODIS Dec NDVI

≤130 >130

Irrigated Others

Elevation, SRTM

≤1300 m >1300 m


Algorithm 3



Irrigated Areas of Tajikistan for year 2005: Whole Country View

Comparison: Knowledge Layer vs. Algorithm Derived Layer

Note: Once you have the algorithm, it takes only a few minutes to derive irrigated areas.

30 m spatial resolution 30 m spatial resolution

Knowledge Layer

(truth) for year 2005Algorithm derived layer

also for year 2005



Knowledge Layer

(truth) for year 2005Algorithm derived layer

also for year 2005

Irrigated Areas of Tajikistan for year 2005: Zoom in View





Algorithm for rainfed Areas




Landsat B3 (chlorophyll

absorption)MODIS Feb NDVI

>18 and ≤30

%

reflectance

≤18 or >30 %

reflectance≤130 >130

Rainfed Others

Elevation, SRTM

derived

≤1300 m >1300 m

Automated Mapping of Rainfed Areas of Tajikistan

Algorithm 4a



Landsat B5 (moisture

sensitivity)MODIS Feb NDVI

>31 and ≤40 %

reflectance

≤31 or >40 %

reflectance≤130 >130

Rainfed Others

Elevation from

SRTM

≤1300 m >1300 m


Algorithm 4b



Landsat NDVI MODIS Feb NDVI

>130

and

≤160

≤130

or

>160

≤130 >130

Rainfed Others

Elevation from

SRTM

≤1300 m >1300 m


Algorithm 4c



Landsat B6 (thermal

temperature)MODIS Feb NDVI

>155

and

≤185 in

degree

Celsius

≤155 or

>185 in

degree

Celsius

≤130 >130

Rainfed Others

Elevation

from SRTM

≤1300 m >1300 m


Algorithm 4d



Rainfed Others

Elevation from SRTM

slope

≤1500 m >1500 m


Algorithm 4e



Knowledge Layer (truth) for year 2005

Rainfed Areas of Tajikistan for year 2005: Full Country View


Note: Once you have the algorithm, it takes only a few minutes to derive rainfed areas.


Algorithm derived layer also for year 2005

Algorithm derived layer also for year 2005Knowledge Layer (truth) for year 2005



Rainfed Areas of Tajikistan: Zoom in view





Algorithm for irrigated +rainfed Areas




Irrigated + Rainfed Areas of Tajikistan: Full Country View


Note: Once you have the algorithm, it takes only a few minutes to derive irrigated and rainfed areas.






Irrigated + Rainfed Areas of Tajikistan: Zoom in View





Accuracy Assessment

Error matrix



Irrigated, Rainfed, and Other Areas of Tajikistan

Error Matrix: Knowledge Layer vs. Algorithm Derived

Algorithm Derived Data

Irrigated AreasRainfed

areas

All other LCLU

classesRow total

Producer's

accuracy

Errors of

Omissions

Irrigated Areas 7398009 152145 30263 7580417 97.6 2.4

Rainfed areas 143604 2520206 252235 2916045 86.4 13.6

All other LCLU

classes33029 129403 142088064 142250496 99.9 0.1

Column total 7574642 2801754 142370562 152006279

User's accuracy 97.7 90.0 99.8 152746958

Errors of

Commission2.3 10.0 0.2

Overall

accuracy99.5

Khat 0.96

Ref

eren

ce D

ata



Algorithm Applied for

Independent Data Layer (year 2010)

Showing Irrigated Areas



Irrigated Areas of Tajikistan 2010: Whole Country View


Knowledge Layer (truth) of

irrigated areasAlgorithm derived irrigated areas for

independent data set of year 2010




irrigated areas

Algorithm derived irrigated areas for




Irrigated Areas of Tajikistan 2010: Zoom in View







Showing Rainfed Areas



Rainfed Areas of Tajikistan 2010: Full Country View





rainfed areas

Algorithm derived rainfed areas of


Algorithm derived rainfed areas of



rainfed areas



Rainfed Areas of Tajikistan 2010: Zoom in view







Showing Rainfed Areas



Irrigated + Rainfed Areas of Tajikistan 2010: Full Country View





irrigated + rainfed areas

Algorithm derived irrigated + rainfed areas

of independent data set of year 2010

Algorithm derived irrigated + rainfed areas

of independent data set of year 2010Knowledge Layer (truth) of

irrigated + rainfed areas



Irrigated + Rainfed Areas of Tajikistan 2010: Zoom in View




irrigated + rainfed areasAlgorithm derived irrigated + rainfed areas

of independent data set of year 2010



Irrigated + Rainfed Areas of Tajikistan 2010: Zoom in View




Accuracy Assessment

Error matrix



Irrigated, Rainfed, and Other Areas of Tajikistan

Error Matrix: Knowledge Layer vs. Algorithm Derived

(2010)Algorithm Derived Data

Irrigated AreasRainfed

areas

All other LCLU

classesRow total

Producer's

accuracy

Errors of

Omissions

Irrigated Areas 6514801 1035399 30217 7580417 85.9 14.1

Rainfed areas 206960 2456860 252225 2916045 84.3 15.7

All other LCLU

classes40163 127180 142083136 142250479 99.9 0.1

Column total 6761924 3619439 142365578 151054797

User's accuracy 96.3 67.9 99.8 152746941

Errors of

Commission3.7 12.1 0.2

Overall

accuracy98.9

Khat 0.91

Ref

eren

ce D

ata



Way Forward



Irrigated and Rainfed Areas of India MODIS 500 m time-series data

Uncertainties still

exist since Landsat

was not used in

this study

0.00

0.20

0.40

0.60

0.80

1.00

Jun-00 Sep-00 Dec-00 Mar-01

Date

ND

VI

01. Irrigated 100% - Rice/Rice 03. Irrigated 100% - Rice 12. Water bodies

Irrigated Rice Areas of South AsiaMODIS 500 m time-series data

Use of

Landsat, in

addition to

MODIS, will

help establish

crop type

and\or

dominance



Irrigated and Rainfed Areas of ChinaLandsat + MODIS data Fusion

Large sample of field plot data for

entire China from precise locations

in addition to Landsat and MODIS

data



Irrigated cropland areas of Africa. The global annualized irrigated area (AIA) in the African continent is only about 2%

compared to 14% of the global population. There is a real opportunity to expand irrigated areas in Africa to facilitate

green and blue revolutions.

Irrigated Areas of Africa Based on Multi-resolution Data

Fragmented irrigated

cropland areas of Africa

certainly needs fine spatial

resolution……probably even

IKONOS\Quickbird\Rapideye

etc.



Distribution of rain-fed croplands in the African continent. The total global rain-fed cropland area (Global: 1.13 Bha) in

Africa is 17% and its productivity can be improved through irrigation and better land management practices to help

move towards a green revolution.

Rainfed Areas of Africa Based on Multi-resolution Data

Fragmented irrigated

cropland areas of Africa

certainly needs fine spatial

resolution……probably even

IKONOS\Quickbird\Rapideye

etc.

Global Irrigated Areas + Rainfed AreasSpatial Distribution of Global Croplands

1.53 billion Hectares of total croplands at the end of the last millennium



Combine algorithms to derive global croplands



PE&RS Special Issue on Remote Sensing of Global Croplands and Their Water Use: Call for Papers

a knowledge-based automated cropland mapping … · a knowledge-based automated cropland mapping...

Documents