IITA News

ACAI combines machine learning, field trial data, and crop models to optimize fertilizer recommendations

ACAI combines machine learning, field trial data, and crop models to optimize fertilizer recommendations

The African Cassava Agronomy Initiative (ACAI) site-specific fertilizer recommendation (FR) tool is built to provide optimized and economically beneficial recommendations to cassava growers. The tool considers local soil data, weather conditions, prices of available fertilizers and cassava root produce, planting and harvest dates, and the investment capacity of the farmer.

ACAI combines machine learning, field trial data, and crop models to optimize fertilizer recommendations

Onsite data gathering by members of the ACAI team.

ACAI has been conducting nutrient omission trials (NOTs) in Nigeria and Tanzania in collaboration with national research and development partners to find out how cassava responds to nutrients. Results show a large variation in nutrient response indicating the need for site-specific fertilizer recommendations.

To provide site-specific recommendations, ACAI is developing an integrated system using machine learning techniques coupled with process-based crop models. The ACAI team is combining the Light Interception and Utilization model (LINTUL), Quantitative Evaluation of the Fertility of Tropical Soils model (QUEFTS), and economic optimizer algorithms to calibrate the recommendations. The mechanism put in place determines the soil nutrient supply capacity, yield potential, nutrient-limited yield, and fertilizer rates required to acquire a target yield maximizing net revenue by combining observations from field trials, available GIS data, weather data, and farmers’ ability to invest in fertilizer.

Using the QUEFTS model, the soil NPK supply was accurately predicted using the observed yield response in the NOTs. At these locations, the relationship between apparent soil nutrient supply and soil properties obtained from GIS layers from the International Soil Reference and Information Centre (ISRIC) was modeled using machine learning techniques. These models, in turn, were used to predict the soil NPK for the entire target intervention area. These soil properties can sufficiently explain the regional level of soil variation. To explain soil variation at a short range, however, the GIS layers need to be complemented with a local scale soil fertility indicator.

The use of common local soil fertility indicators, such as local soil name, soil depth/color, cropping history, perception of soil fertility, cropping history, manure/fertilizer use, etc., are not sufficiently generic as their predictive ability depends on the local context. Such indicators are therefore challenging to use in a standardized way. Within ACAI, the current yield was found to be the best generic fertility indicator to adjust the soil nutrient supply at a regional scale to local soil conditions. The resulting FR tool is currently providing site-specific recommendations packaged as an open data kit (ODK) form and can be applied offline in the field on a mobile device. Progress is being made to develop generally accessible versions using IVR, USSD, a mobile app, and printable maps and guides.

One of the major challenges to improve the accuracy of the FR recommendation is the quality of the price data both for the fertilizers and the cassava roots. ACAI is exploring partnerships with various organizations providing digital market information as well as price mapping to provide meaningful default values.

Next steps include validating the FR tool both functionally, verifying whether the recommendations outperform current practices in the field, and architecturally, evaluating the user-friendliness and how the tool can best fit within the dissemination strategy of development partners.

ACAIfertilizerIITA News no. 2486soil fertility

Communications • 23rd May 2019


Previous Post

Next Post

Leave a Reply

Your email address will not be published / Required fields are marked *