Zhe Guo - Africa Agriculture Watch (AAgWa) Launch Event.pdf
1. Leveraging Artificial Intelligence (AI) & Satellite Remote
Sensing Data for Decision-making in the African
Agricultural Sector
AFRICA AGRICULTURE WATCH
(AAgWa)
• AI Trends in Remote Sensing
• Dr. Zhe Guo
Senior GIS Coordinator
International Food Policy Research Institute
2. Leveraging Artificial Intelligence (AI) & Satellite Remote Sensing Data for Decision-making in the African
Agricultural Sector
CGIAR (Consultative Group on International
Agricultural Research) Centers
3. Leveraging Artificial Intelligence (AI) & Satellite Remote Sensing Data for Decision-making in the African
Agricultural Sector
The Remote sensing have been successfully applied in the many fields
such as agriculture, environmental sciences, ecology, urban planning
especially in U.S. and Europe
• USDA’s Crop Data Layer
• ESA’s Copernicus projects
• Time series of land cover, cropland, and forest changes from
University of Maryland
Researchers are exploring the potential applications of remote
sensing in Africa, but they face challenges due to the spatial
heterogeneity of smallholder agriculture and the complexity
of farming practices in the region.
4. Leveraging Artificial Intelligence (AI) & Satellite Remote Sensing Data for Decision-making in the African
Agricultural Sector
The rapid evolution of remote sensing technology is introducing promising
solutions for remote sensing applications in Africa for agricultural studies,
especially in the following areas:
• Access to publicly available satellite datasets such as Landsat (30m, 16-day revisit) and Sentinel (10m,
5-day revisit) is becoming increasingly widespread.
• Free cloud computing platforms like Google Earth Engine provide a scalable and cost-effective way to
process large amounts of remote sensing data.
• Low/no cost machine learning packages and high-end computers allow for the development of
sophisticated algorithms for analyzing remote sensing data, including the ability to classify land cover
and detect changes over time.
• Satellite data can now be accessed in near real-time, enabling more timely and accurate monitoring of
agricultural systems and environmental conditions.
• The availability of more ground truth data is improving the accuracy of remote sensing analyses and
enabling the development of more robust models
7. Leveraging Artificial Intelligence (AI) & Satellite Remote Sensing Data for Decision-making in the African
Agricultural Sector
Space agencies on food security: $15 million
NASA Harvest project
8. Leveraging Artificial Intelligence (AI) & Satellite Remote Sensing Data for Decision-making in the African
Agricultural Sector
Descartes Lab
9. Spatial Data Analytics:
Smallholder crop type and yield estimation using satellite data and
machine learning approach
• New advances in AI offer promise for smallholder crop
area and yield estimation
o Publicly available satellite data - Sentinel(10m, 5-day revisit)
o Free cloud computing platform (Google Earth Engine)
o No/low cost of (deep) learning packages
• Crop type mapping in South Africa
o Deploy TensorFlow, an open source deep learning platform with time
series of sentinel data for crop type mapping.
o Limited ground truth samples (<500 samples per crop)
o Distinguished major crop types including inter-cropping and fallow land
in Free State with R-squared of 0.71
• Crop yield estimation in Ethiopia
o The deep learning neural networks outperform other machine learning
algorithms.
o vegetation index from Satellite + climate variables + soil give the best
model performance
o The maize yield estimate has R-squared of 0.62 across three AEZ zones.
For more information, please contact Zhe Guo (z.guo@cgiar.org)
Fallow
Maize
Pasture
SoyaBeans
Sunflower
Vegetables
WheatMaize
WheatSoya
Non crop area
Reference
Predicted
Predicted
yield
Reference yield
a a
10. Leveraging Artificial Intelligence (AI) & Satellite Remote Sensing Data for Decision-making in the African
Agricultural Sector
Hierarchy of Relevant Crop-related Statistics That
Might Be Improved Using Earth Science Data
• Cropland (areas)
Permanent (areas)
Arable (areas)
Rainfed/irrigated (system areas)
Growing seasons (periods)
Cropping patterns (areas)
Crop specific (areas and yields)
12. Spatial
Production
Allocation
Model
(SPAM)
SPAM model estimates spatial distribution of 42+ crop
types including area, production, and yield at pixel level by
disaggregating the data from coarser units, such as
countries and sub-national provinces, to finer units.
SPAM model uses:
o Entropy-based, data-fusion approach
o Combines a variety of inputs including tabular and
spatial raster data
o Assesses cropping system distribution and
performance of 42+ crops
o Spatial resolution: ~10km/1km
o Spatial Extent: Global/country
o Temporal resolution: every 5 years (2000, 2005,
2010, 2017, 2020)
13. SPAM Outputs
The outputs of the SPAM model produce maps of 42 crop types by 2 cropping systems
and 3 variables, a total of 42*6*6=1512 data layers.
o Variables (per crop and cropping system*)
• Harvest area (ha)
• Physical area (ha)
• Production (mt)
• Yield (kg/ha)
• Value of production (Int$)
• Value of production/harvested area (Int$/ha)
* Cropping systems: I, H, L, S, R(rainfed), A(sum of all)
14. Leveraging Artificial Intelligence (AI) & Satellite Remote Sensing Data for Decision-making in the African
Agricultural Sector
Nightlight density during Covid-19
in Nigeria
March 2019 March 2020
NO2 concentration during Covid19
in Nigeria
March 2019 March 2020
Agricultural Drought Monitoring
and Forecasting System
15. Leveraging Artificial Intelligence (AI) & Satellite Remote
Sensing Data for Decision-making in the African
Agricultural Sector
AFRICA AGRICULTURE WATCH
(AAgWa)
THANK YOU!
For more information, please contact Zhe Guo, z.guo@cgiar.org