Algorithms for forecasting cotton yield based on climatic parameters in Brazil

ABSTRACT

Accurate forecasts of cotton yield are of great interest for the development of the market, increasing the sustainability of the sector worldwide. Thus, the objectives of this study were: 1) to evaluate the influence of climatic elements on cotton yield in Brazil, 2) to predict cotton yield using machine learning algorithms based on climatic elements, 3) to calibrate and test machine learning models to forecast cotton yield based on climate data, and 4) to interpolate the estimated cotton yield of the most accurate model. The cotton yield forecast as a function of climatic elements was performed using machine learning algorithms with four parameters adjusted by ordinary least squares. The models show that cotton yield has a sigmoid trend due to the accumulation of P, PET, STO, and EXC during the cycle. It is possible to forecast cotton yield for the main producing regions of Brazil using Machine learning algorithms. Extra-trees regressor models performed better in forecasting cotton yield using climatic data from planting to flowering. Therefore, it is possible to have average anticipation of around 80 days, allowing the producer time to plan his activities such as harvest and sales strategies.

KEYWORDS:

• Artificial intelligence
• random forest
• cop modelling
• water balance
• deep learning
• bigdata

Additional information

Funding

This work was supported by the Coordination for the Improvement of Higher Education Personnel - Brazil (CAPES) - Financing Code 001.