A joint analysis of crop production trends in Sahel using MODIS NDVI time series, crop modeling and statistic data

(1)

A joint analysis of crop

production trends in Sahel

using MODIS NDVI time

series, crop modeling and

statistic data

Leroux L1_{., Baron C}1_{., Zoungrana B}2_{., Bégué A}1_{., Lo Seen D}1 1_{CIRAD, UMR TETIS, Montpellier, France}

2_{AGRHYMET, Niamey, Niger}

(2)

Background of the study

Food security in West Africa

West Africa is characterized by:

- A strong climate variability in space and time

- High population growth rates

Impacts on agricultural production and on food

security

Challenge of West Africa : Enhance knowledge on crop production

dynamics both at regional and local scale

(3)

Agricultural monitoring in West Africa

Crop conditions and trends Crop area Crop type Crop production Surface harvested yield Yield prediction Biomasse prediction LAI … Remote Sensing Agricultural statistics Crop modeling

Crop area from MODIS land Cover

Potential millet yields for the rainy season 2012 (Agrhymet, August 2012)

(4)

Agricultural monitoring in West Africa

(5)

(6)

Question addressed:

Objectives of the study

(7)

By performing a joint analysis of :

Question addressed:

Objectives of the study

What are the crop production dynamics in Sahel?

1. NDVI trends from remote sensing within the crop domain

2. Yield and biomass output trends from a crop model

(8)

Data & Methods Regional scale Local scale

Normalised Difference Vegetation Index : NDVI

Data

• NDVI product : MODIS MOD13Q1

• Sensitive to vegetation and its conditions

• Correlated to LAI, FAPAR and vegetation primary production  NDVI = proxy for vegetation greeness and biomass production

(9)

Data & Methods

Regional scale

Normalised Difference Vegetation Index : NDVI

Data

Methods

1) Extraction of the crop domain : Application of an averaged crop mask (MODIS Land Cover Product) on annualy integrated NDVI time series

2)Trends analysis by OLS (Ordinary Least Square Regression) 3) Anomalies

Local scale

• NDVI product : MODIS MOD13Q1

• Sensitive to vegetation and its conditions

• Correlated to LAI, FAPAR and vegetation primary production  NDVI = proxy for vegetation greeness and biomass production

(10)

In situ observations from crop model and statistic data – Case study of the Kollo department (Niger)

Statistic data

(11)

Crop model : SARRA-H (Baron et al., 2005) Data & Methods Regional scale

• Suited for the analysis of climate impact on cereal growth and yield in dry environment

• Operating at daily time step

• Simulates attainable yields and biomass at the field scale under climatic constraint

Local scale

Source : sarra-h.teledetection.fr

Statistic data

(12)

Crop model : SARRA-H (Baron et al., 2005) Data & Methods Regional scale Local scale SARRAH Cultivated species and varieties Seeding date

Type of soils Level of fertilization

Rainfall

Other climatic variables (Humidity, temperature,

wind, insolation)

(13)

Crop model : SARRA-H (Baron et al., 2005) Data & Methods Regional scale Local scale SARRAH Cultivated species and varieties • Pearl Millet Hainy Kire

Seeding date Type of soils • Deep Sandy Level of fertilization Rainfall

• Rain gauge measurements

from AMMA-CATCH (7 stations, 2000-2010)

Other climatic variables (Humidity, temperature,

wind, insolation) • Climatic data from Agrhymet

(1 station, 2000 – 2010)

• Pearl Millet Hainy Kire

Rain gauge measurements AMMA-CATCH (7 stations, 2000-2010) Climatic data AGRHYMET (1 station, 2000 – 2010) • Deep Sandy Attainable Yield Day of year Attainable Biomass Day of year

• Automatically

(14)

What does vegetation monitoring by remote sensing tells us about crop production dynamics?

(15)

Data & Methods

Regional scale

NDVI anomalies between 2000 and 2012

NDVI anomaly 2002 NDVI anomaly 2006

NDVI anomaly 2010 NDVI anomaly 2012

Strong spatio-temporal variability of biomass production within the crop domain:  vegetation stress ≈ decrease in biomass production

Local scale

(16)

Data & Methods

NDVI trends between 2000 and 2010 in West Africa

NDVI MODIS trends – significant at 90%

Regional scale

Local scale

(17)

Data &

Methods NDVI MODIS trends – significant at 90%

Regional scale

Local scale

• Stable overall trend

• Identification of areas with:

- a significant increase in biomass production - a significant decrease in biomass production

(18)

Data &

Regional scale

Local scale

- a significant increase in biomass production - a significant decrease in biomass production • North/South effect

(19)

Data &

Regional scale

Local scale

- a significant increase in biomass production - a significant decrease in biomass production • North/South effect

(20)

Study area

• Centered on 13.5°N and 2.5°E • Area of 10,000 km²

• Rainy season from June to September – 200-400mm/year

– Sahelian climate • Rainfed agriculture dominated

by pearl millet (low input and low yield)

Site instrumented since 1990’s : availibility of climate and meteorogical daily time series Data & Methods Regional scale Local scale

(21)

(22)

NDVI trends between 2000 and 2010 in the Kollo department

(23)

(24)

(25)

Data & Methods

Regional

scale Are the SARRA-H simulated yields and biomass in agreement _{with estimation from remote sensing?}

Local scale

(26)

Data & Methods

Regional scale

At the site scale? Between 2000 and 2010

Are the SARRA-H simulated yields and biomass in agreement with estimation from remote sensing?

Local scale -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z -sc o res Kare Simulated Yield

(27)

Data & Methods

Regional scale

Local scale -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z -sc o res Kare Simulated Yield Simulated Biomass

(28)

Data & Methods

Regional scale

Local scale -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z -sc o res Kare Simulated Yield Simulated Biomass NDVI_Crop_mask

(29)

R=0.08

R

R=0.08

R=0.19

(30)

R=0.08

R

R=0.08 R=0.19 R=0.28 R=0.53 * R=0.42

(31)

R=0.08

R

R=0.08 R=0.19 R=0.28 R=0.53 * R=0.42 R=-0.26

(32)

R=0.08

R

R=0.08 R=0.19 R=0.28 R=0.53 * R=0.42 R=-0.26 Poor agreement between

simulated data and remote sensing at the

(33)

Data & Methods

At the Kollo department scale? Between 2000 and 2010

Data aggregation of 7 stations • No significant trends (pvalue>0.10) Regional scale Local scale -4.000 -3.000 -2.000 -1.000 0.000 1.000 2.000 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z-sco re s

Simulated Yield (pvalue=0.18) Simulated Biomass (pvalue=0.05*) NDVI_Crop_mask (pvalue=0.27)

(34)

Data & Methods

Data aggregation of 7 stations • No significant trends

(pvalue>0.10)

• Good overall correlation between NDVI vs simulated data Regional scale Local scale R=0.71 4.0 204.0 404.0 604.0 804.0 1004.0 1204.0 1404.0 4.0 4.5 5.0 5.5 Yi e ld (k g/ h a) Annual NDVI NDVI_crop_mask/Yield (pvalue=0.01**) -4.000 -3.000 -2.000 -1.000 0.000 1.000 2.000 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z-sco re s

(35)

Data & Methods

(pvalue>0.10)

• Good overall correlation between NDVI vs simulated data Regional scale Local scale R=0.71 R=0.50 0.0 500.0 1000.0 1500.0 2000.0 2500.0 3000.0 3500.0 4000.0 4.0 204.0 404.0 604.0 804.0 1004.0 1204.0 1404.0 4.0 4.5 5.0 5.5 B io m ass (k g/ h a) Yi e ld (k g/ h a) Annual NDVI NDVI_crop_mask/Yield (pvalue=0.01**) NDVI_crop_mask/Biomass (pvalue=0.11) -4.000 -3.000 -2.000 -1.000 0.000 1.000 2.000 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z-sco re s

(36)

Data & Methods

(pvalue>0.10)

• Good overall correlation between NDVI vs simulated data Regional scale Local scale R=0.71 R=0.50 0.0 500.0 1000.0 1500.0 2000.0 2500.0 3000.0 3500.0 4000.0 4.0 204.0 404.0 604.0 804.0 1004.0 1204.0 1404.0 4.0 4.5 5.0 5.5 B io m ass (k g/ h a) Yi e ld (k g/ h a) Annual NDVI NDVI_crop_mask/Yield (pvalue=0.01**) NDVI_crop_mask/Biomass (pvalue=0.11)

Both satellite and simulation observations show no significant

trends

Good overall agreement

-4.000 -3.000 -2.000 -1.000 0.000 1.000 2.000 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z-sco re s

(37)

What do the statistics data say? Comparison between NDVI, simulated data and yields from AGRHYMET

(38)

Data & Methods

Comparison at the Kollo department scale- Between 2000 and 2010

Regional scale

Local scale

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Y ie ld ( t/ h a ) Millet 0.0085 t/ha/year

(39)

• Statistics vs Satellite : Opposite trends

-3.000 -2.500 -2.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z-sco re s

(40)

-3.000 -2.500 -2.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z-sco re s

Statistic yield (pvalue=0.36) NDVI_crop_mask (pvalue=0.27)

(41)

• Statistics vs Simulation : Overall the same year-to-year variability but opposite

trends -3.500 -3.000 -2.500 -2.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Z-sco re s

Statistic yield (pvalue=0.36) NDVI_crop_mask (pvalue=0.27) Simulated Yield (pvalue=0.18)

(42)

• Moderate correlation between NDVI VS Yield statistics

4.0 4.2 4.4 4.6 4.8 5.0 5.2 5.4 5.6 4 504 1004 A n n u al N D VI Kg/ha NDVI_crop/Statistic (pvalue=0.35) R=0.30

(43)

• No correlation between simulated data VS Yield statistics

4.0 4.2 4.4 4.6 4.8 5.0 5.2 5.4 5.6 4 504 1004 A n n u al N D VI Kg/ha NDVI_crop/Statistic (pvalue=0.35) R=0.30 0.0 200.0 400.0 600.0 800.0 1000.0 1200.0 1400.0 0.0 1000.0 2000.0 3000.0 4000.0 5000.0 6000.0 7000.0 4 204 404 604 804 Yi e ld (k g/ h a) B io m ass (k g/ h a) Statistic (kg/ha) Biomass/Statistic (pvalue=0.82) Yield/Statistic (pvalue=0.66) R=0.14 R=-0.06

(44)

• No correlation between simulated data VS Yield statistics

4.0 4.2 4.4 4.6 4.8 5.0 5.2 5.4 5.6 4 504 1004 A n n u al N D VI Kg/ha NDVI_crop/Statistic (pvalue=0.35) R=0.30 0.0 200.0 400.0 600.0 800.0 1000.0 1200.0 1400.0 0.0 1000.0 2000.0 3000.0 4000.0 5000.0 6000.0 7000.0 4 204 404 604 804 Yi e ld (k g/ h a) B io m ass (k g/ h a) Statistic (kg/ha) Biomass/Statistic (pvalue=0.82) Yield/Statistic (pvalue=0.66) R=0.14 R=-0.06

Opposite trends but no significant

(45)

At regional scale

• Satellite observations show an overall stable dynamic of the crop vegetation between 2000 and 2010

Conclusion and Perspectives

 Limitations : Only 10 years of data

(46)

At local scale

• Poor agreement between satellite and simulated observations at the site scale BUT good agreement at the Kollo department scale.

(47)

At local scale

 Simulations for different types of soils + different farmer strategies  Limitations : Only 10 years of data

(48)

At local scale

 Simulations for different types of soils + different farmer strategies  Use a crop mask with a better spatial accuracy

(49)

At local scale

 Use spatial rain data rather than local data  Limitations : Only 10 years of data

(50)

At local scale

• Poor agreement with statistic data

 Use spatial rain data rather than local data

 Weakness of aggregated statistic data : lack of spatial representativeness  Limitations : Only 10 years of data

(51)

(52)