1. IFPRI’s Africa Region (AFR)
Highlights of current projects
RISE 2019
November 18, 2019
2. Predictive Modeling Plans for
Agriculture Watch
Racine Ly
Research Coordinator
AFR Division – IFPRI
Washington, November 18th, 2019
3. Facts and Motivations
• More than half of global population growth between now and 2050 is expected to occur in Africa.
• Need to increase agricultural productivity, accessible, nutrients contents with respect to the environment.
• Need to mitigate and anticipate effects of climate change.
• A need to monitor crop lands and assess what is going to happen (most likely) in the near future.
Data Challenges
• In developing countries access to data can be challenging (Scarcity, not collected, ownership)
• Remote sensing (satellites images) can help to « partially » close that gap.
Provide agricultural features forecasting tools (and datasets) to policy makers and researchers’ communities
4. Products
Crop Stat. Forecast
• Predict Normalized
Difference Vegetation
Index from remote sensing
product (Satellite Images)
• High correlation with crop
growth and Yield (0.94 for
Millet in Burkina Faso,
Rasmussen, 1992)
• Inputs: Mean NDVI values
retrieved from satellite
images for each countries’
boroughs
• Outputs: Prediction of
mean NDVI values for the
next 8 days
• Time-series data learned
with a Deep LSTM NN
Climatic Forecast
• Forecast of land surface
temperature (Day and
Night) and Rainfall
• Growth condition, Drought
(with surface water bodies
status)
• Inputs: Day and Night
LST/borough from satellite
images
• Outputs: LST prediction
for the next 8 days
• Time-series data learned
with a Deep LSTM NN
LUCC
• Group crops’ type by
reflectance similarities for
ground validation.
• Crop type can be very
heterogeneous in Africa
(ex. Senegal, 20.9% of
farmlands < 1ha, 50.7% <
3ha, Bourgoin 2016)
• Inputs: Reflectance bands
for Multispectral satellite
images
• Outputs: Crop Clusters
and Crop type
Classification.
• Crop Clustering with k-
means and Crop
classification with logistic
Regression
Yield Prediction
• By building the direct
mathematical relationship
between satellite images
and yield.
• Inputs: NDVI, Climatic,
Yield / Crop growth model,
soil type mask
• Outputs: Yield forecast per
crop species.
• Deep Artificial Neural
Network
Economic Models (CGE) – Impact in Economy
5. First Results - Crop Status
Location: Sare Bidji – Kolda (Rural)
Peak2002
Location: Ngor – Dakar (Urban)
Peak2003
Peak2009
Peak2010
Peak2011
Peak2017
Peak2018
Peak2019
Fig2. Normalized Difference Vegetation Index (NDVI) seasonality pattern –
smooth values for rural area dominated by crop lands and noisy signal for
Urban areas.
Terra and Aqua Satellites with MODIS sensor
Spatial Resolution 250 m (1 pixel ~ 6.25 ha)
Temporal Resolution Every 16 days, 8days when Terra and Aqua merged
Temporal window 2002 – Now (for the 8 days cycle)
Dataset size ~ 800 satellite images
Fig1. Terra and Aqua merged datasets to reach 8 days temporal resolution for
Senegal – May 25th to July 4th, 2003
May 25th June 2nd June 10th June 18th
June 26th July 4th
6. First Results - Crop Status (Cont’d)
Actual Predicted Absolute Error
GrowingSeasonOff/SowingSeason
Fig3. Actual vs. predicted mean Normalized Difference Vegetation Index for Senegal – First row corresponds
to 2016 growing season (NDVI peak) - second raw corresponds to Off / Sowing season (Min. NDVI values) –
last column represent predictions absolute errors. All boroughs are trained simultaneously
Key Results
• Good prediction (comparison btw. actual
and predicted maps/values)
• Taking the mean NDVI can help to reduce
the dataset dimensionality.
• Relatively large errors for large areas due
to noisy NDVI signal, urban areas are
being considered into the NDVI averaging
• Training all borough at the same time can
help save time and enrich the dataset
Future Work
• Subtract urban areas from dataset to only
target farmlands for better accuracy and
reduce computation cost.
• Implementation of Blocks 2 and 3.
• Platform (v0) under construction.
9. OBJECTIVE OF BR: Evaluate country performance in achieving the CAADP-Malabo
goals and targets for agricultural growth and transformation in Africa by 2025
Submitted a report Did not submit a report
1st Biennial Review (2017) 2nd Biennial Review (2019)
Reporting Dynamics
New countries (2019)
• Eritrea
• Guinea-Bissau
• Somalia
• South Sudan
Did not report (2019)
• Algeria
• Comoros
• Libya
• Sahrawi
• Egypt**
• Sao Tome and
Principe**
** Reported in 2017
47
8
49
6
Member states
that submitted
Members states that
did not submit
7 thematic areas
23 performance categories
43 indicators
7 thematic areas
24 performance categories
47 indicators+4
+1
10. Country BR-support pilot activities
BMGF funding to improve data and quality for CAADP implementation in 5
pilot countries (Kenya, Malawi, Mozambique, Senegal, and Togo)
Submission
(eBR)
•July 15
•August 31
Assessment of
the inaugural
(2017) BR process
and report
BR Teams
(core, data
clusters, review)
Training
(general, gaps
from assessment)
•Data
collection and
compilation
•Report
preparation and
revision
Validation
(review team,
senior mg’t,
ASWG, all
stakeholders)
Effect 1: difference-in-difference in reporting rate: (DID-RR)
DID-RR = (RR2019 – RR2017)pilots – (RR2019 – RR2017)like-pilots
Effect 2: difference in quality of reporting or incidence of data issues (D-QR)
D-QR = (QR2019)pilots – (QR2019)like-pilots
Like-pilots = non-pilots within 1, 2, & 3 standard deviations of mean (RR2017)pilots
11. BR-support pilot activities: results and lessons
Compared to the like-pilot countries, BR-support activities in the pilots helped:
raise their reporting rate by 8 to 9 %pts on average (DID-RR)
reduce their data issues by 3 to 7 %pts on average (D-QR)
As approach used in pilots shared with all member states the additional
resources and hand-holding are critical
Key challenge is how to maintain data clusters: engage them in using BR
data to conduct policy analysis financial support and technical assistance
5
8 9
12
2
4
9
5 4
0
5
10
15
1sd 2sd 3sd KEN MWI MOZ SEN TGO
Pilots Like-pilots Pilots
70
80
90
100
1sd 2sd 3sd KEN MWI MOZ SEN TGO
Pilots Like-pilots Pilots
2017 2019
% of data parameters reported with issuesData parameters reported, % of total required
6
-3 -2 -2
4 3
1
11
12
12. IFPRI’s Africa Region (AFR)
Highlights of current projects
RISE 2019
November 18, 2019
14. The Program - 1/2
Based at IFPRI Dakar since January 2017
Co-facilitated by and with teams at IFPRI, University of Bonn and Imperial
College London
Funded by African Development Bank, BMZ, DfID
15. The Program - 2/2
Aim
Panel of 17 experts
Facilitate policy choices by African
governments to achieve AU
Agenda 2063 and SDGs
Focus on what works, why and
how
Analysis of policy and institutional
innovations
Outputs
2 reports a year (July and Dec)
Malabo Montpellier Forum
Bilateral meetings
Papers, blogs, op-eds
16. Key activities since last RISE – 1/3
Water-Wise: Smart Irrigation Strategies for Africa
(Dec 2018)
6 country case studies - Ethiopia, Kenya, Mali,
Morocco, Niger, South Africa
Selected based on their level of irrigation and
pace of expansion of irrigated areas
9 Policy recommendations
Launched at MaMo Forum in Rabat, Morocco
17. Key activities since last RISE – 2/3
Byte by Byte: Policy Innovation for Transforming
Africa’s Food System with Digital Technologies
(July 2019)
7 country case studies – Côte d’Ivoire, Ghana,
Kenya, Morocco, Nigeria, Rwanda, Senegal
Selected based on their performance on EBA
ICT Index and GSMA Mobile Connectivity
Index
9 Policy recommendations
Launch at Mamo Forum in Kigali, Rwanda
18. Key activities since last RISE – 3/3
Conferences, workshops, bilateral meetings - AGRF, World Food Prize,
Atlantic Dialogues, AAAE, FAO, Government of Togo, IsDB etc.
Social media campaigns - #MaMoFaces on Twitter, quarterly webinars
TV/radio interviews, op-eds, blogs – Deutsche Welle, Africa Renewal,
IFPRI blog, CNBC Africa, IPS News, SciDev
19. Activities until end of 2019
22 Nov: webinar on digitalization in the agriculture sector in Senegal
26 Nov: event at IFPRI DC Transforming Africa’s Food System with Digital
Technologies
12 Dec: side event at Atlantic Dialogues in Marrakesh on food-energy-
water nexus
17 Dec: report launch and MaMo Forum in Banjul, The Gambia
21. Welfare implications of migration:
results from a village model
Fleur Wouterse and Sunday Odjo
• Niger is one of the least developed and
poorest countries in the world
• Population strongly depends on
agriculture
• Yields rely on a single short rainy season
• Circular labor mobility across rainfall
gradients are a livelihood strategy to
maximize investments of time and other
resources
• Can migration be leveraged for
structural transformation and how?
22. Village CGE for Niger
Village CGE model: household model embedded in village CGE
SAMs constructed using 2019 data on 600 rural households in
Tillaberi and Maradi
Three household groups:
oLandless (Poverty rate 53%)
oCropping only (Poverty rate 46%)
oCropping and large livestock (Poverty 42%)
Migration rate 0.34 percent
Migration mainly international, Nigeria a prime destination
Remittance share of income 16 percent for landless
households.
23. Welfare effects of migration
Scenarios
one additional person migrating internationally
redirection of one migrant from a domestic to an
international destination
return of migrant with additional human capital
(extra year of formal education)
a 10 percent increase in remittances
-2.6 -2.7
-6.4
0.5 0.2 0.5
9.1
7.5
00.3 0.1 0.4
-8
-6
-4
-2
0
2
4
6
8
10
Cropping only Cropping and large
livestock
Landless
%changeinEV
24. IFPRI’s Africa Region (AFR)
Highlights of current projects
RISE 2019
November 18, 2019
25. Julie Collins
Africa Region, International Food Policy Research Institute
Support to National Agricultural Investment
Plan (NAIP) Formulation
Strengthening evidence-based agricultural planning
26. Background
2014 Malabo Declaration upheld key CAADP commitments and expanded
focus to new commitments
oAgricultural growth and expenditure
oPoverty reduction and hunger elimination
oExpanding regional trade; value chain development
oFood security and nutrition; gender
oResilience and climate-smart agriculture
oMutual Accountability
Many first-generation National Agricultural Investment Plans (NAIPs) came
to an end in 2015
Second-generation NAIPs must achieve goals and targets in multiple
areas
27. ReSAKSS Resources for 2nd-generation NAIPs
Toolkit
o Clarify metrics, data needs, tools and
methodologies
Experts Group
o ~200 local experts identified
o 55 days of training workshop
implemented
Task Force
o 20 international experts to provide
backstopping support to experts
group
28. NAIP Analytical Outputs
o Country Malabo Status Assessment and Profile Report
oReview recent changes and evaluate the country’s current situation with
respect to each of the thematic areas
o Country Malabo Goals and Milestones Report
oIdentify investment priorities and milestones for the county to achieve key
Malabo commitments
o Country Policy and Program Opportunities Report
oDefine key elements of successful policies and programs in the different
thematic areas and identify opportunities for the country to achieve best
practices in program design
29. Progress to date
• SAP reports for 32 countries
• MGM reports for 25 countries
• PPO reports for 8 countries
• Additional support for NAIP
design as needed
35. Three Nutrient Adequacy
Measures
“Enough production
of nutrient X?”
“Enough consumption
of nutrient X?”
“Enough of nutrient X
reach the market?”
Types:
A: Post-harvest losses
B: Production constraints
C: Production constraints, with market opportunities
D: Demand constraints (low income or lack of awareness)
TYPOLOGY2
NutritionSmartAgriculture
36. TYPOLOGY 2 – Kenya (food energy and nutrient adequacy)
NUTRIENT
DEFICIENCIES
LOW
PRODUCTION
FOOD
LOSSES
DEMAND
PROBLEM
40. The ReSAKSS website
http://www.resakss.org/
It provides easy access to data, tools,
analysis, knowledge products, and
resources on CAADP implementation
and other African agricultural and rural
development strategies.
41. ReSAKSS Country eAtlas (RCeA)
http://eatlas.resakss.org/
The RCeA is a GIS-based data
exploration platform designed to help
policy analysts and policymakers
access and use high quality and highly
disaggregated data on agricultural,
socio-economic and bio-physical
indicators to guide agricultural policy
and investment decisions.
43. Predictive Modeling Plans for
Agriculture Watch
Racine Ly
Research Coordinator
AFR Division – IFPRI
Washington, November 18th, 2019
44. Facts and Motivations
• More than half of global population growth between now and 2050 is expected to occur in Africa.
• Need to increase agricultural productivity, accessible, nutrients contents with respect to the environment.
• Need to mitigate and anticipate effects of climate change.
• A need to monitor crop lands and assess what is going to happen (most likely) in the near future.
Data Challenges
• In developing countries access to data can be challenging (Scarcity, not collected, ownership)
• Remote sensing (satellites images) can help to « partially » close that gap.
Provide agricultural features forecasting tools (and datasets) to policy makers and researchers’ communities
45. Products
Crop Stat. Forecast
• Predict Normalized
Difference Vegetation
Index from remote sensing
product (Satellite Images)
• High correlation with crop
growth and Yield (0.94 for
Millet in Burkina Faso,
Rasmussen, 1992)
• Inputs: Mean NDVI values
retrieved from satellite
images for each countries’
boroughs
• Outputs: Prediction of
mean NDVI values for the
next 8 days
• Time-series data learned
with a Deep LSTM NN
Climatic Forecast
• Forecast of land surface
temperature (Day and
Night) and Rainfall
• Growth condition, Drought
(with surface water bodies
status)
• Inputs: Day and Night
LST/borough from satellite
images
• Outputs: LST prediction
for the next 8 days
• Time-series data learned
with a Deep LSTM NN
LUCC
• Group crops’ type by
reflectance similarities for
ground validation.
• Crop type can be very
heterogeneous in Africa
(ex. Senegal, 20.9% of
farmlands < 1ha, 50.7% <
3ha, Bourgoin 2016)
• Inputs: Reflectance bands
for Multispectral satellite
images
• Outputs: Crop Clusters
and Crop type
Classification.
• Crop Clustering with k-
means and Crop
classification with logistic
Regression
Yield Prediction
• By building the direct
mathematical relationship
between satellite images
and yield.
• Inputs: NDVI, Climatic,
Yield / Crop growth model,
soil type mask
• Outputs: Yield forecast per
crop species.
• Deep Artificial Neural
Network
Economic Models (CGE) – Impact in Economy
46. First Results - Crop Status
Location: Sare Bidji – Kolda (Rural)
Peak2002
Location: Ngor – Dakar (Urban)
Peak2003
Peak2009
Peak2010
Peak2011
Peak2017
Peak2018
Peak2019
Fig2. Normalized Difference Vegetation Index (NDVI) seasonality pattern –
smooth values for rural area dominated by crop lands and noisy signal for
Urban areas.
Terra and Aqua Satellites with MODIS sensor
Spatial Resolution 250 m (1 pixel ~ 6.25 ha)
Temporal Resolution Every 16 days, 8days when Terra and Aqua merged
Temporal window 2002 – Now (for the 8 days cycle)
Dataset size ~ 800 satellite images
Fig1. Terra and Aqua merged datasets to reach 8 days temporal resolution for
Senegal – May 25th to July 4th, 2003
May 25th June 2nd June 10th June 18th
June 26th July 4th
47. First Results - Crop Status (Cont’d)
Actual Predicted Absolute Error
GrowingSeasonOff/SowingSeason
Fig3. Actual vs. predicted mean Normalized Difference Vegetation Index for Senegal – First row corresponds
to 2016 growing season (NDVI peak) - second raw corresponds to Off / Sowing season (Min. NDVI values) –
last column represent predictions absolute errors. All boroughs are trained simultaneously
Key Results
• Good prediction (comparison btw. actual
and predicted maps/values)
• Taking the mean NDVI can help to reduce
the dataset dimensionality.
• Relatively large errors for large areas due
to noisy NDVI signal, urban areas are
being considered into the NDVI averaging
• Training all borough at the same time can
help save time and enrich the dataset
Future Work
• Subtract urban areas from dataset to only
target farmlands for better accuracy and
reduce computation cost.
• Implementation of Blocks 2 and 3.
• Platform (v0) under construction.
48. IFPRI’s Africa Region (AFR)
Highlights of current projects
RISE 2019
November 18, 2019
51. OBJECTIVE OF BR: Evaluate country performance in achieving the CAADP-Malabo
goals and targets for agricultural growth and transformation in Africa by 2025
Submitted a report Did not submit a report
1st Biennial Review (2017) 2nd Biennial Review (2019)
Reporting Dynamics
New countries (2019)
• Eritrea
• Guinea-Bissau
• Somalia
• South Sudan
Did not report (2019)
• Algeria
• Comoros
• Libya
• Sahrawi
• Egypt**
• Sao Tome and
Principe**
** Reported in 2017
47
8
49
6
Member states
that submitted
Members states that
did not submit
7 thematic areas
23 performance categories
43 indicators
7 thematic areas
24 performance categories
47 indicators+4
+1
52. Country BR-support pilot activities
BMGF funding to improve data and quality for CAADP implementation in 5
pilot countries (Kenya, Malawi, Mozambique, Senegal, and Togo)
Submission
(eBR)
•July 15
•August 31
Assessment of
the inaugural
(2017) BR process
and report
BR Teams
(core, data
clusters, review)
Training
(general, gaps
from assessment)
•Data
collection and
compilation
•Report
preparation and
revision
Validation
(review team,
senior mg’t,
ASWG, all
stakeholders)
Effect 1: difference-in-difference in reporting rate: (DID-RR)
DID-RR = (RR2019 – RR2017)pilots – (RR2019 – RR2017)like-pilots
Effect 2: difference in quality of reporting or incidence of data issues (D-QR)
D-QR = (QR2019)pilots – (QR2019)like-pilots
Like-pilots = non-pilots within 1, 2, & 3 standard deviations of mean (RR2017)pilots
53. BR-support pilot activities: results and lessons
Compared to the like-pilot countries, BR-support activities in the pilots helped:
raise their reporting rate by 8 to 9 %pts on average (DID-RR)
reduce their data issues by 3 to 7 %pts on average (D-QR)
As approach used in pilots shared with all member states the additional
resources and hand-holding are critical
Key challenge is how to maintain data clusters: engage them in using BR
data to conduct policy analysis financial support and technical assistance
5
8 9
12
2
4
9
5 4
0
5
10
15
1sd 2sd 3sd KEN MWI MOZ SEN TGO
Pilots Like-pilots Pilots
70
80
90
100
1sd 2sd 3sd KEN MWI MOZ SEN TGO
Pilots Like-pilots Pilots
2017 2019
% of data parameters reported with issuesData parameters reported, % of total required
6
-3 -2 -2
4 3
1
11
12
55. The Program - 1/2
Based at IFPRI Dakar since January 2017
Co-facilitated by and with teams at IFPRI, University of Bonn and Imperial
College London
Funded by African Development Bank, BMZ, DfID
56. The Program - 2/2
Aim
Panel of 17 experts
Facilitate policy choices by African
governments to achieve AU
Agenda 2063 and SDGs
Focus on what works, why and
how
Analysis of policy and institutional
innovations
Outputs
2 reports a year (July and Dec)
Malabo Montpellier Forum
Bilateral meetings
Papers, blogs, op-eds
57. Key activities since last RISE – 1/3
Water-Wise: Smart Irrigation Strategies for Africa
(Dec 2018)
6 country case studies - Ethiopia, Kenya, Mali,
Morocco, Niger, South Africa
Selected based on their level of irrigation and
pace of expansion of irrigated areas
9 Policy recommendations
Launched at MaMo Forum in Rabat, Morocco
58. Key activities since last RISE – 2/3
Byte by Byte: Policy Innovation for Transforming
Africa’s Food System with Digital Technologies
(July 2019)
7 country case studies – Côte d’Ivoire, Ghana,
Kenya, Morocco, Nigeria, Rwanda, Senegal
Selected based on their performance on EBA
ICT Index and GSMA Mobile Connectivity
Index
9 Policy recommendations
Launch at Mamo Forum in Kigali, Rwanda
59. Key activities since last RISE – 3/3
Conferences, workshops, bilateral meetings - AGRF, World Food Prize,
Atlantic Dialogues, AAAE, FAO, Government of Togo, IsDB etc.
Social media campaigns - #MaMoFaces on Twitter, quarterly webinars
TV/radio interviews, op-eds, blogs – Deutsche Welle, Africa Renewal,
IFPRI blog, CNBC Africa, IPS News, SciDev
60. Activities until end of 2019
22 Nov: webinar on digitalization in the agriculture sector in Senegal
26 Nov: event at IFPRI DC Transforming Africa’s Food System with Digital
Technologies
12 Dec: side event at Atlantic Dialogues in Marrakesh on food-energy-
water nexus
17 Dec: report launch and MaMo Forum in Banjul, The Gambia
Agricultural growth and expenditure
Poverty reduction and hunger elimination
Expanding regional trade
Value chain development
Food security and nutrition
Resilience and climate-smart agriculture
Gender
Mutual Accountability
SAP: Angola, Botswana , Cameroon, Eswatini, Ethiopia, Gabon, Kenya, Lesotho, Malawi, Mozambique, Namibia, Rwanda, Seychelles, Tanzania, Uganda, Zambia, and Zimbabwe
MGM: 15 ECOWAS countries plus Kenya and 9 GIZ countries (Angola, Botswana, Cameroon, Eswatini, Gabon, Lesotho, Namibia, Zambia, Zimbabwe)
PPO: 8 GIZ countries (all except Cameroon)