CONTRIBUTION OF MODIS LAND COVER PRODUCT TO
THE ANALYSIS OF THE AGRICULTURAL DOMAIN BETWEEN 2001 AND 2011 IN
WEST AFRICA
Leroux L1., Jolivot A1., Zoungrana B2., Bégué A1., Lo Seen D1
1 CIRAD, UMR TETIS, Montpellier, France 2 AGRHYMET, Niamey, Niger
© Leroux, 2013
March 19-21,2014 Berlin, Germany
Background of the study
Food security in West Africa
West Africa is characterized by: A strong climate variability in space and time
Impacts on agricultural production and on food security
Challenge of West Africa : Enhance knowledge on crop production dynamics both at regional and local scale
Boyd et al., 2013, NCC
High population growth rates
Background of the study
Crop production and crop area in West Africa
Production = Yield x Area
Remote sensing and the monitoring of crop areas
208,256,000 ha
38,764,012 ha (20%)
Cultivable area Cultivated area
FAO, 2005
• Global land products: improve
our understanding of the extent and distribution of major land cover types at a global scale
• Maps locating the most important cropping areas 0
10
20
30
40
50
60
70
80
90
100
0
1
2
3
4
5
6
7
8
9
10
1991
1993
1994
1995
1997
1998
1999
2000
2001
2002
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
Cum
ula
tive
(%)
Fre
quency
Frequency Cumulative (%)
Number of papers on global land products (Web Of Science)
Total = 80
Objectives of the study
Question addressed:
What is the reliability of Global Land Cover products for crop areas characterisation and mapping in West Africa? Case study of the MODIS Land Cover Product v5.1 (MCD12Q1.V51)
Outline:
Assessment of cultivated areas
1
Agricultural statistic databases
Spatial distribution of
cultivated areas At local scale
2
Crop areas classification from high resolution
images
Examine the spatial distribution of the
precision at regional scale
3
To what extent the MODIS Land Cover Product is able to estimate cultivated areas?
Assessment of the MODIS LCP reliability for crop area estimations
Data
• Agricultural Statistic databases
FAOSTAT - Agricultural Land (2001 – 2011)
Regional & National scales
AGRHYMET - Surfaces harvested for major crops (2001 – 2011)
Burkina Faso : Subnational L-1 & Subnational L-2 scales
• MODIS Land Cover Product
V51 – 500 m - Yearly (2001-2011) 2 Cropland classes (IGBP): Cropland (100%)
Cropland/Natural Vegetation Mosaic (50%)
𝐶𝑟𝑜𝑝𝑎𝑟𝑒𝑎 = 𝑛𝑏𝑠𝑝𝑖𝑥𝑒𝑙𝑠 × 5002
𝐶𝑟𝑜𝑝𝑎𝑟𝑒𝑎_𝑚𝑖𝑥𝑡𝑒 = 𝑛𝑏𝑠𝑝𝑖𝑥𝑒𝑙𝑠 × 5002 × 0.5
Regional Scale – FAOSTAT/MODIS
Average values between 2001 - 2011
16 % 80.4Mha
18% 92.6Mha
Cropland FAOSTAT Cropland MODIS
Assessment of the MODIS LCP reliability for crop area estimations
Dynamics between 2001 - 2011
60
65
70
75
80
85
90
95
100
105
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Cro
pla
nd
are
a (
Mh
a)
Cropland area FAOSTAT (pvalue=4x10^(-6)***)
Cropland area MODIS (pvalue=0.002**)
1.3 MHa/year
1.5 MHa/year
Good estimation at regional scale
Assessment of the MODIS LCP reliability for crop area estimations
National Scale – FAOSTAT/MODIS
Average values between 2001 - 2011 Dynamics between 2001 - 2011
MODIS
FAOSTAT
Niger
Overall good estimation at national scale Some discrepancies in terms of dynamics
Assessment of the MODIS LCP reliability for crop area estimations
Subnational L-1 Scale (Region)– Burkina Faso – AGRHYMET/MODIS
Average values between 2001 - 2011
East AGRHYMET
Dynamics between 2001 - 2011
MODIS Moderate estimation at subnational scale Overestimations by MODIS LCP Some discrepancies in terms of dynamics - East
Assessment of the MODIS LCP reliability for crop area estimations
Subnational L-2 Scale (Province) – Burkina Faso – AGRHYMET/MODIS
Average values between 2001 - 2011
Tapoa Gourma
Yatenga
Dynamics between 2001 - 2011
AGRHYMET
MODIS
Poor estimation at subnational scale Overestimations by MODIS LCP Opposite dynamics
What about the MODIS Land Cover Product accuracy concerning the spatial distribution of crop areas?
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Data
• MODIS Land Cover Product
Cropland and mixte cropland classes for 2011
• High resolution classifications of the crop domain 4 classifications from SPOT (2.5m) - Supervised method 2007-Mali (Vintrou et al., 2012, RSE)
30 classifications from Google Earth by photo-interpretation Square of 25km²
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Methods
Pareto Boundary
MODIS
HRS
Omission error
Comission error
and F-Scores
Error Matrix
Low resolution cell
Crop class in HS data
75%-100%
50%-75% <25%
25%-50%
% of crop class in each low resolution cell
Site size = 5km x 5km (100 MODIS Pixels)
Boschetti et al., 2004 (RSE)
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Methods
Pareto Boundary
MODIS
HRS
Omission error
Comission error
Confusion Matrix
Co
mm
issi
on
Err
or
Omission Error
1
0
0 1
Low-resolution biais
Performance of the classification algorithm
Pareto Boundary
Boschetti et al., 2004 (RSE)
Site size = 5km x 5km (100 MODIS Pixels)
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Crop areas estimation – MODIS vs HRS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
5 9 16 15 19 4 29 27 25 18 8 28 2 13 24 3 30 17 20 10 14 31 23 21 6 26 1 7 22
% C
rop
Site
% Crop HRS % Crop MODIS
MODIS Over-estimation MODIS Under-estimation
Over-estimation < ~40%< Under-estimation
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Crop areas accuracy – MODIS vs HRS
Zone 1 : North Senegal % of crop class in each MODIS cell
Homogeneous site
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Crop areas accuracy – MODIS vs HRS
Zone 1 : North Senegal The Pareto Boundary
Low unreacheable region High performance of the MODIS classification
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Crop areas accuracy – MODIS vs HRS
Zone 2 : Gambia % of crop class in each MODIS cell
Fragmented site
Large patches of crop
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Crop areas accuracy – MODIS vs HRS
The Pareto Boundary
Moderate unreachable region Moderate performance of the MODIS classification
Zone 2 : Gambia
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Crop areas accuracy – MODIS vs HRS
Zone 18 : South Nigeria % of crop class in each MODIS cell
Highly fragmented site Little patches of crop
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Crop areas accuracy – MODIS vs HRS
The Pareto Boundary
Very High unreachable region Moderate performance of the MODIS classification
Zone 18 : South Nigeria
Can we map the crop areas MODIS LCP accuracy considering the landscape spatial heterogeneity?
Map of the MODIS LCP uncertainties for crop areas
Methods
MODIS Indicators of Landscape fragmentation at regional scale
(grid of 5km x 5km) Matheron Compactness Crop area
Crop perimeter Nbs of patches Mean Focal Diversity
Multiple Linear Regression at the site scale
ActualAccuracy Indicators at site scale
Estimated Accuracy Indicators at regional scale
Map of the MODIS LCP uncertainties for crop areas
Variables Selection and Model Fitting
𝑂𝑚𝑖𝑠𝑠𝑖𝑜𝑛 = 0.38 + 0.36𝑀𝑎𝑡ℎ𝑒𝑟𝑜𝑛 − 0.01𝐶𝑟𝑜𝑝𝑃𝑒𝑟𝑖𝑚𝑒𝑡𝑒𝑟 R²=0.60*** RMSE=0.15%
Map of the MODIS LCP uncertainties for crop areas
Map of the users accuracy estimate
Maximum user accuracy estimate : 70% Better accuracy estimate in the West
Conclusion
Crop area estimations
Regional-National-SubNational-1-SubNational-2
Definition of Fallows (Statistic) Definition of the Mosaic class (MODIS) Weakness of aggregated statistic data: lack of spatial representativeness
MODIS LCP accuracy for crop area spatial distribution
Over-estimation for crop fraction below 40% Fragmented landscape = impact on the low resolution map accuracy Notion of Mixel = Mixed Pixel
Map of the MODIS LCP uncertainties for crop areas
Omission error = 𝐹 𝑀𝑎𝑡ℎ𝑒𝑟𝑜𝑛, 𝐶𝑟𝑜𝑝 𝑃𝑒𝑟𝑖𝑚𝑒𝑡𝑒𝑟 Maximum user accuracy estimate = 70% for crop Increase the number of test sites to improve our results Analyse the spatial distribution of accuracy in accordance with agro-ecological
climatic zonations and slopes.
Conclusion
Key ideas
Original approach based on Remote Sensing data and statistic data to assess the reliability of the MODIS LCP for crop areas estimation and localisation in West Africa
An assessment of the Global Land Product accuracy allows to anticipate product limitations for users and leads to appropriate map use
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Mo
nth
ly N
DV
I
Cropping system with 1 cycle
Cropping system with 2 cycles
Benin Mali
Assessment of the MODIS LCP accuracy for crop area spatial distribution
Map of the MODIS LCP uncertainties for crop areas
Model validation
