Office at International Food Policy Research Institute (IFPRI)/ P.O. Box 24063 Dakar, Senegal
IFPRI OFFICE DAKAR
resaksschallenge@cgiar.org
Follow Us

Predictive Modeling Contest

Predictive Modeling Contest

Description

Cassava’s profound significance within the Democratic Republic of Congo goes well beyond its status as a staple crop. Its exceptional adaptability serves as a linchpin in addressing food security across diverse agro-ecological zones, furnishing a dependable carbohydrate source amidst climatic uncertainties. This multiplicity extends culturally, where cassava’s incorporation enriches culinary traditions and fosters a sense of communal identity, embedding it as a cultural emblem. Economically, untapped potential rests in harnessing cassava’s versatility for export and value-added processing, buoyed by the escalating global demand for cassava-derived products. This market trend positions the DRC favorably, paving the way for economic growth and diversification. At the heart of this significance lies the ability to accurately predict cassava production, an imperative that empowers strategic resource allocation, a pivotal tenet for ensuring not only food security but also economic resilience.

The challenge is about predicting – at a pixel level – Cassava production in Democratic Republic of Congo for the year 2023 using machine learning techniques. The model you will develop uses remote sensing data such as the Normalized Difference Vegetation Index (NDVI), daytime Land Surface Temperature (LST), and Rainfall (R) as explanatory variables, and historical cassava production values as response variable. Both parameters, explanatory and response, are shared in csv files with latitude and longitude coordinates. You are asked to predict cassava production for each pixel. A functional form of the task is to build a model F such as:

Cassava Production = F (NDVI, LST, R)

The choice of the machine learning technique should be the participants’ choices as well as the addition of explanatory variables available on the ReSAKSS, e-Atlas, and AAgWa platforms for better accuracy performances. However, any addition of explanatory variables should be relevant to the prediction task and explainable.

The provided datasets are on an annual basis aggregated for each year’s growing season. Your prediction, therefore, will be for the 2023 season.

Evaluation

The evaluation metrics is the Root Mean-Squared Error. Your submission file should contain predicted production values for cassava, for each pixel and for the year 2023. We isolated a part of the dataset for accuracy evaluation of your model’s outputs. You should use the same production unit as the one used in the provided dataset.  

Your submission should look like this:

Pixel ID Latitude Longitude
2023 cassava production (predicted values)
00001 20
00002
     

Information about the dataset parameters

  • LST stands for Land Surface Temperature and is in Celsius Degree for day and night-time. It is the temperature of the air at one meter above ground.
  • The rainfall dataset was retrieved from the Climate Hazards Group InfraRed Precipitations with Station data (CHIRPS) databases. For more information, please visit https://www.chc.ucsb.edu/data/chirps.
  • The evapotranspiration was retrieved from the MOD16A2 version 006 from the Land Processes Distributed Active Archive Center (LP DAAC) product. For more information, please visit https://lpdaac.usgs.gov/products/mod16a2v006/.

Rules

The solution must use publicly available and open-source packages. 

  • Only submission files containing prediction values for all pixels will be considered.
  • Individuals can form teams to participate in this contest. However, you cannot submit your work as an individual while you are a member of a participating team. 
  • Only one submission is allowed. However, you can request a second submission and the ReSAKSS Data Challenge team will review it. 
  • If you are the winner, the ReSAKSS Data Challenge team will request you to submit your model for copyright and predictive values verification.  
  • If two solutions have the same scoring, the date and time of submissions will be considered to assess the winner.
  • You agree to legally assign ownership of all rights of copyright of the winning solution code to the ReSAKSS Data Challenge while reserving the right to the solution code for non-commercial purposes while crediting the ReSAKSS Data Challenge. 
  • The ReSAKSS Data Challenge can change the rules at any time. 
  • Each candidate can only participate in one of the two IT contests.

Timeline

The IT Predictive Modeling category will open on August 24, 2023 and will close on October 29, 2023. Applications will be made through this dedicated ReSAKSS Challenge website.

Prize

Only 1st prize will be attributed to this IT Predictive Modeling contest. The Winner will receive $1500 USD, and will be awarded during the 2023 ReSAKSS Conference.

Download guidelines and/or Apply