Machine Learning Specialist, AKADEMIYA2063 The Use of Earth Observation for Food Crop Production Systems Transformation. The Case of Crop mapping for Rwanda, and Senegal Dr. Jean Paul l. FAYE

#2023ReSAKSS #2023ATOR Outline I. Introduction II. Satellite Remote Sensing Data and Machine Learning for Crop mapping III. Maize Mapping: Case Nyagatare, Rwanda IV. Groundnut Mapping: Case Groundnut Basin, Senegal V. Conclusion

#2023ReSAKSS #2023ATOR Introduction Ø Agriculture plays a crucial role in Africa for several by contributing significantly to the continent's economy, food security, employment, and overall development. Ø However, agriculture is highly dependent on weather conditions, and many African countries are vulnerable to the impacts of climate change. Ø Developing resilient and sustainable agricultural practices is crucial for adapting to changing climate patterns and ensuring long-term food security. Thus, innovation approaches are needed for achieving sustainable and resilient agricultural systems in Africa. Ø The utilization of technologies such as machine learning and earth observation has gained a lot of attention and are very promising for powering the agriculture in the very near future.

#2023ReSAKSS #2023ATOR Satellite Remote Sensing Data and Machine Learning for Crop mapping

#2023ReSAKSS #2023ATOR • Satellite uses sensors that collect information about the Earth's surface without actively emitting any signals • Sensors passively record the sunlight reflected or emitted by the Earth's surface in various wavelengths • Satellite uses sensors that actively emit signals or energy towards the Earth's surface and measure the reflected or scattered signals Satellite Remote Sensing Data

#2023ReSAKSS #2023ATOR Spectral bands refer to the specific ranges of EM radiation that sensors capture Electromagnetic Spectral

#2023ReSAKSS #2023ATOR Spectral bands refer to the specific ranges of EM radiation that sensors capture Coastal Blue Green Red NIR: vegetation Panchromatic Shortwave IR Thermal IR Shortwave IR Spectral Bands

#2023ReSAKSS #2023ATOR Machine Learning: Supervised and Unsupervised Learning Machine Learning is a complex of algorithms and methods that address the problems of Classification, Clustering, and Forecasting. Supervised Learning • From training data set 𝑥! , 𝑦! , we want to learn 𝑓 such that 𝑦! = 𝑓 𝑥! . • We want the model to generalize to unseen inputs. 𝑓 𝑥! ∗ = 𝑦! ∗ for new data point 𝑥! ∗ Unsupervised Learning • From training data set 𝑥! , we want to learn the structure of the data. Ex. Clustering data in such that all data belonging to the same group have the same properties

#2023ReSAKSS #2023ATOR Random Forest Classifier Maize Soya Support Vector Machine Classifiers Machine Learning Models Used for Crop Mapping

#2023ReSAKSS #2023ATOR Dataset: Calculate the indices in different location NDVI NDVI NDVI NDVI NDVI NDVI NDVI NDVI NDVI NDVI Same for all the remaining indices Polygone of cropland

#2023ReSAKSS #2023ATOR Machine Learning for Maize Mapping: Case of Nyagatare, Rwanda

#2023ReSAKSS #2023ATOR Survey Data:

#2023ReSAKSS #2023ATOR Area of Interest and Data distribution

#2023ReSAKSS #2023ATOR Model Validation using the Test Dataset 20 percent of the generated data are chosen for the testing data No Maize Maize

#2023ReSAKSS #2023ATOR Using the Validated Model for pixel Classifications • Zoom in an image with the correct calculated band indices at the pixel level. • Pixel size is: 10𝑥10 = 100 𝑚! • This image is given to the trained model and the pixels are classified as Groundnut or not.

#2023ReSAKSS #2023ATOR Maize mapping

#2023ReSAKSS #2023ATOR Maize mapping Region of interest: circle of 25 km radius around Nyagatare

#2023ReSAKSS #2023ATOR Machine Learning for Groundnut Mapping: Case of Groundnut Basin, Senegal

#2023ReSAKSS #2023ATOR Data Collection Senegal and its groundnut basin Collected data distribution

#2023ReSAKSS #2023ATOR Groundnut mapping

#2023ReSAKSS #2023ATOR Survey vs Predicted Data distribution Collected data distribution Predicted Probabilities data distribution

#2023ReSAKSS #2023ATOR Future Direction: Signature Transfer Between Countries Ø Classify the signature of each specific crop into clusters for each country where data has been collected and the machine model trained. Ø The obtained database can be used for crop annotation in other countries: • Collect remote sensing data and compute the same indices at the area of interest • Do the clustering of indices • Comparer with the database for annotation and pixel crop classification

#2023ReSAKSS #2023ATOR Conclusion v We use an approach based on Satellite Remote Sensing Data and machine Learning techniques for crop mapping v Application of the model in different countries where data have been collected shows a clear map of crops v With more data collection in any country, we will be able, in any time of the year, to run the model that will do the crop mapping for the entire country and tells us the crops and the type of crops grown in that country v The Crop Mapping output is one output, but the impact is multidimensional

THANK YOU