A Non-Destructive Time Series Model for the Estimation of Cherry Coffee Production
2022
Pablo Rodr韌uez, Jhonn | Camilo Corrales, David | Griol, David | Callejas, Zoraida | Carlos Corrales, Juan | Universidad del Cauca [Popayán] | AGroécologie, Innovations, teRritoires (AGIR) ; Institut National Polytechnique (Toulouse) (Toulouse INP) ; Université de Toulouse (UT)-Université de Toulouse (UT)-Ecole d'Ingénieurs de Purpan (INP - PURPAN) ; Institut National Polytechnique (Toulouse) (Toulouse INP) ; Université de Toulouse (UT)-Université de Toulouse (UT)-Institut National de Recherche pour l’Agriculture, l’Alimentation et l’Environnement (INRAE) | Carlos III University of Madrid | Dept. of Languages and Computer Systems ; Universidad de Granada = University of Granada (UGR) | Departamento Administrativo de Ciencia, Tecnologia e Innovacion Colciencias | Innovaccion-Cauca (SGR-Colombia) 4633, 04C-2018
International audience
显示更多 [+] 显示较少 [-]英语. Coffee plays a key role in the generation of rural employment in Colombia. More than 785,000 workers are directly employed in this activity, which represents the 26% of all jobs in the agricultural sector. Colombian coffee growers estimate the production of cherry coffee with the main aim of planning the required activities, and resources (number of workers, required infrastructures), anticipating negotiations, estimating, price, and foreseeing losses of coffee production in a specific territory. These important processes can be affected by several factors that are not easy to predict (e.g., weather variability, diseases, or plagues.). In this paper, we propose a non-destructive time series model, based on weather and crop management information, that estimate coffee production allowing coffee growers to improve their management of agricultural activities such as flowering calendars, harvesting seasons, definition of irrigation methods, nutrition calendars, and programming the times of concentration of production to define the amount of personnel needed for harvesting. The combination of time series and machine learning algorithms based on regression trees (XGBOOST, TR and RF) provides very positive results for the test dataset collected in real conditions for more than a year. The best results were obtained by the XGBOOST model (MAE = 0.03; RMSE = 0.01), and a difference of approximately 0.57% absolute to the main harvest of 2018.
显示更多 [+] 显示较少 [-]