To what extend can satellite rainfall estimates be used for crop yield prediction

in West Africa?

Ramarohetra J.

Sultan B., Baron C., Gaiser T., Gosset M., Alcoba M.

95% Rainfed

Population growth

Agriculture

1/15

Introduction Context and challenges

Cereal production /hab. ↘

70% of the population

95% Rainfed

70% of the population

Crop yield increase =

a challenge for tomorrow

Agriculture

1/15

Introduction Context and challenges

Population growth

Cereal production /hab. ↘

Agriculture

High variability

1/15

Introduction Context and challenges

Agriculture

Limitation

1/15

Introduction Context and challenges

Agriculture

Limitation

Adaptation

EWS 1/15

Introduction Context and challenges

Agriculture

Limitation

Adaptation

EWS

Translate climate variations in terms of agricultural relevant

variables

1/15

Introduction Context and challenges

Crop yield

Crop models

Climate inputs

Where can we find rainfall data?

2/15

Introduction How to make crop yield predictions?

Crop yield

Crop models

Climate inputs

Where can we find rainfall data?

Long historical time series Insufficient network density Missing data Delay in reporting time

2/15

Introduction How to make crop yield predictions?

Long historical time series Insufficient network density Missing data Delay in reporting time

2/15

Crop yield

Crop models

Climate inputs

Where can we find rainfall data?

Introduction How to make crop yield predictions?

Long historical time series Insufficient network density Missing data Delay in reporting time

Past, present and future climate Big biases

2/15

Crop yield

Crop models

Climate inputs

Where can we find rainfall data?

Introduction How to make crop yield predictions?

Long historical time series Insufficient network density Missing data Delay in reporting time

Past, present and future climate Big biases

2/15

… and ?

Crop yield

Crop models

Climate inputs

Where can we find rainfall data?

Introduction How to make crop yield predictions?

Long historical time series Insufficient network density Missing data Delay in reporting time

Past, present and future climate Quite bad spatial resolution

Quasi global coverage Better and better resolution Increasingly available Accessible in near real time

2/15

Crop yield

Crop models

Climate inputs

Where can we find rainfall data?

Introduction How to make crop yield predictions?

Long historical time series Insufficient network density Missing data Delay in reporting time

Past, present and future climate Quite bad spatial resolution

Quasi global coverage Better and better resolution Increasingly available Accessible in near real time

2/15

Crop yield

Crop models

Climate inputs

Where can we find rainfall data?

Introduction How to make crop yield predictions?

Crop yield

To what extend can satellite rainfall estimates be used for crop

prediction ? 3/15

?

Introduction How to make crop yield predictions?

Crop models

Yield to assess

Crop models

4/15

?

Material and Method Satellite rainfall products assessment.

Control yield

COMPARISON

To what extend can satellite rainfall estimates be used for crop

prediction ?

Material and Method Satellite products to assess

Satellite Product Type PERSIANN

Uncalibrated CMORPH

TRMM 3B42-RT GSMAP-MKV+

GPCP-1dd Calibrated TRMM 3B42v6

RFEv2

Daily, 1°, 2003-2009

5/15

Near real time Satellite data only

Satellite and in situ data

Niamey & Djougou AMMA-CATCH observing

system

Precipitation:

Block kriging performed from

- 54 to 56 raingauges in Niger

- 54 raingauges in Benin

6/15

Material and Methods Area of study and groundbased data

2003-2009 mean annual rainfall

Maize, low fertilization (5kg of N /year).

2 Mechanistic crop models Field scale

Daily time step

First aim: quantify soil erosion impacts on soil productivity Nutrient uptake taken into account

3 pearl millet: - Hainy Kire (early) - Somno (late) - Souna 3 (early)

7/15

Based on water balance No account for nutrient

Material and Methods Crop modelling

Niamey Djougou

8/15

Near real time Global non calibrated

Global calibrated Regional

calibrated

(Niger)

(Bénin)

(Niger)

(Niger)

Results The biases in crop yield prediction

8/15

Near real time Global non calibrated

Global calibrated Regional

calibrated

(Niger)

(Bénin)

(Niger)

(Niger)

Results The biases in crop yield prediction

+

-

Near real time Global non calibrated

Global calibrated Regional

calibrated

(Niger)

(Bénin)

(Niger)

(Niger)

Results The biases in crop yield prediction

8/15

+

-

Near real time Global non calibrated

Global calibrated Regional

calibrated

(Niger)

(Bénin)

(Niger)

(Niger)

Results The biases in crop yield prediction

8/15

+

-

Near real time Global non calibrated

Global calibrated Regional

calibrated

(Niger)

(Bénin)

(Niger)

(Niger)

Results The biases in crop yield prediction

8/15

+

-

What are these biases due to ?

Near real time Global non calibrated

Global calibrated Regional

calibrated

Results The biases in crop yield prediction

9/15

10/15

Results What drives maize yields in Benin?

Maize yield vs. Cycle cumulative rainfall

Maize yield vs. Rainy days number

10/15

Results What drives maize yields in Benin?

R=0.59

Maize yield vs. Cycle cumulative

rainfall < 800 mm

Maize yield vs. Rainy days number

All Cum. Rain. < 800 mm

10/15

Results What drives maize yields in Benin?

Limitant rainfall

70% 30%

R=0.59

Maize yield vs. Cycle cumulative

rainfall < 800 mm

Maize yield vs. Rainy days number

All Cum. Rain. < 800 mm

All Cum. Rain. < 800 mm

All Cum. Rain. > 800 mm

10/15

Results What drives maize yields in Benin?

The number of rainy days is crucial for crop yields in Benin

Maize yield vs. Cycle cumulative

rainfall < 800 mm

Maize yield vs. Rainy days number (cum.

rainfall > 800 mm)

For non limitant rainfall Limitant

rainfall R=0.59

70% 30%

11/15

Cumulative rainfall bias Rainy days number bias Maize yield bias

Bia

s %

Near real time Global non calibrated Global calibrated

Regional calibrated

Results The rainfall biases in Benin

Maize yield bias

→ Systematic bias

11/15

Cumulative rainfall bias Rainy days number bias Maize yield bias

Bia

s %

Near real time Global non calibrated Global calibrated

Regional calibrated

Results The rainfall biases in Benin

Maize yield bias and cumulative rainfall bias

→ No systematic bias

11/15

Cumulative rainfall bias Rainy days number bias Maize yield bias

Bia

s %

Near real time Global non calibrated Global calibrated

Regional calibrated

An underestimation of the number of rainy days in Benin

Results The rainfall biases in Benin

Maize yield bias and rainy days number bias

12/15

Results What drives millet (souna) yields?

Millet Souna yield vs. Cycle cumulative rainfall

12/15

Results What drives millet (souna) yields?

The total rainfall amount is crucial for crop yields in Niger

Millet Souna yield vs. Cycle cumulative rainfall < 600 mm

83% 17%

Limitant rainfall

12/15

Results What drives millet (souna) yields?

An underestimation of the total rainfall amount leads to an underestimation of millet yields

Millet Souna yield bias vs. Cycle cumulative rainfall bias

83% 17%

Limitant rainfall

Millet Souna yield vs. Cycle cumulative rainfall < 600 mm

12/15

Calibrated Uncalibrated

Results What drives millet (souna) yields?

An underestimation of the total rainfall amount leads to an underestimation of millet yields

Millet Souna yield bias vs. Cycle cumulative rainfall bias

83% 17%

Limitant rainfall

Millet Souna yield vs. Cycle cumulative rainfall < 600 mm

13/15

Near real time Global non calibrated

Global calibrated Regional

calibrated

Results Biases attribution

Millet in Niger

Maize in

Benin

Raw SRFE

13/15

Near real time Global non calibrated

Global calibrated

Results The biases in Niger are reduced by correcting the total rainfall

Regional calibrated

Millet in Niger

Maize in Benin

Cumulative rainfall bias corrected SRFE

13/15

Near real time Global non calibrated

Global calibrated Regional

calibrated

Results The biases in Benin are reduced by correcting the rainfall distribution

Millet in Niger

Maize in Benin

Distribution bias corrected SRFE

Conclusion

• Satellite rainfall estimates may introduce large biases in crop yield prediction.

• Rainfall estimates bias propagation to yield depends on the crop and region considered.

• Crop yield biases are not simple extrapolation of rainfall cumulative amount biases, rainfall distribution is crucial as well.

• Uncalibrated products give the highest biases.

Prospects

• Compare simulated yield variations with yield observation

• What happens if radiation is taken from satellites as well ?

• Would a statistical bias correction improve near real time products performance for an Early Warning System application?

Thank you for your attention !

Material and Method: Bias attribution What are crop yield biases due to ? What are crop yield biases due to ?

Bias = +50% R = 0.83

113 rainy days

Bias = 0 R = 1

121 rainy days

Material and Method: Bias attribution What are crop yield biases due to ?

1. Adjusted cumulative rainfall

Bias = +50% R = 0.83

113 rainy days

Bias = 0 R = 1

121 rainy days

Bias = 0 R = 0.83

113 rainy days

Material and Method: Bias attribution What are crop yield biases due to ?

1. Adjusted cumulative rainfall

2. Adjusted distribution

Bias = +50% R=1

121 rainy days

Bias = +50% R = 0.83

113 rainy days

Bias = 0 R = 1

121 rainy days

Bias = 0 R = 0.83

113 rainy days

Maize, low fertilization (5kg of N /year).

Material and Method: Crop models

Based on water balance

First aim: quantify erosion effect on soil productivity

3 pearl millet: - Hainy Kire (local, early) - Somno (MTDO) (local, late) - Souna 3 (improved, early)

Satellite Product

Type Temporal coverage

Spat. Res. Temp. Res.

PERSIANN Near real time 03/2000 → 0.25° 3h

CMORPH Near real time 12/2002 → 0.072° (0.25°)

30min

TRMM 3B42-RT Near real time 2002 → 0.25° 3h

GSMAP-MKV+ Global

uncalibrated 01/2003-12/2008 0.1° 1h

GPCP-1dd Global calibrated 1997 → 1° daily TRMM 3B42v6 Global calibrated 01/1998 → 0.25° 3h

RFEv2 Régional

calibrated 01/2001 → 0.1° daily