Estatística aplicada à aquicultura, desafios e métodos para a modelagem de crescimento do camarão cinza

Abstract

Aquaculture in Brazil has been standing out in agribusiness in recent years, despite being a relatively recent sector when compared to cattle and poultry. Aquaculture production in Brazil in 2020 was 629.3 thousand tons. The vast hydrographic network (75,000 km) in rivers, lakes, and ponds (about 167,000 km2), in addition to the extension of the coastline (7,367 km from Amapá to Rio Grande do Sul), combined with the favorable climate, given Brazil a country with great potential for the sector. To highlight aquaculture, in a competitive market (national and international), it is necessary to keep up the evolution of statistical methods and artificial intelligence in the search for more efficient production. Within the aquaculture business, revenue, costs, and profits are based on the animal weight. Therefore, the growth modeling of organisms is used as a production management tool. Some growth data in aquaculture have peculiar characteristics that generate consequences for analysis and modeling. They are usually incomplete or limited. This means that the data are restricted to a few observations and are often limited to observations below the inflection point of the sigmoid curve. This occurs due to the economic strategy of the farms or simply the demand of the consumer market. This limitation of the observed data presumably causes bias in the inference of nonlinear models. Results from simulations and comparisons between the growth of wild animals and fisheries supported this hypothesis. As a result, a method was proposed to correct this possible bias using a hierarchical Bayesian approach. Real data were used to compare it with the traditional frequentist approach used. The sensitivity in detecting the best treatment can make the new method a powerful management tool in animal production, including trials designed for scientific research. In the second chapter, based on the proposed Bayesian methodology, six non-linear hierarchical models were evaluated for modeling the growth of gray shrimp (Litopenaeus vannamei) (Morgan-Mercer- Flodin, Michaelis-Menten, Weibull, von Bertalanffy, Gompertz, and Logistics) and adjusted to real data from a production farm. The Weibull growth equation stood out. The final model was validated showing an accuracy of 95.76% and 85.71% in the hierarchical pond and production cycle levels, respectively. Finally, a sensitivity analysis was carried out to detect subtle differences between crops and we concluded that the new approach is very efficient for comparing treatments.