Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.


Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  2021 ASABE Annual International Virtual Meeting  2100807.(doi:10.13031/aim.202100807)
Authors:   Matheus Alves Sasso, Paula Dornhofer Paro Costa
Keywords:   agroecology, artificial-intelligence, big-data,species-distribution-model

ABSTRACT. Despite playing an essential role in the Brazilian economy, traditional agriculture generally leads to several environmental problems because it stimulates deforestation, pollutes rivers and soils by using pesticides and fertilizers, and promotes social inequality between large and small producers. As an alternative to traditional agriculture, agroecology studies how polyculture creates beneficial interactions with the environment. It can reduce the ecological impact, promote the greater equality of competition, and expand the Land Equivalent Ratio (LER). However, creating an ecologically balanced system using multiple species with different growing conditions becomes a complicated task. Thus, this project uses artificial intelligence algorithms combined with georeferenced programming techniques to create Species Distribution Models (SDMs) applied to highly used plant species in agroecology inside Brazilian territory. The data required for this task is a combination of geospatial occurrences of species found in Global Biodiversity Information Facility (GBIF) with geo-environmental data made available by the WORLDCLIM and ENVIREM databases. The statistical technique used to create the models was the One Class Support Vector Machine (OCSVM), which can differentiate regions where geo-climatic conditions offered favorable conditions for planting a given species from unfavorable regions. The metric used for evaluating the model was the Area Under the Curve (AUC), which refers to the area under the curve Receiver Operating Characteristic Curve (ROC), demonstrating a binary classifier performance. The metric results ranged from 84% to 99% for the studied species. As a final result, we built a data-driven model that is capable of providing a practical guide for small and medium farmers regarding the crops that are suitable for the desired region through adaptability scores on Brazil's map.

(Download PDF)    (Export to EndNotes)