Modelos de distribuição histórica de peixes migradores nas bacias hidrográficas do Rio Grande do Sul

• Ano de defesa: 2019
• Autor(a) principal: Maciel, Felipe Coutinho
• Orientador(a): Fontoura, Nelson Ferreira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós Graduação em Ecologia e Evolução da Biodiversidade
• Departamento: Escola de Ciências
• País: Brasil
• Palavras-chave em Português: Modelo Estatístico; LOGIT; MaxEnt; Mapa de Distribuição; Peixes Migradores; Distribution Map; Migratory Fish
• Palavras-chave em Inglês: Statistical Model
• Área do conhecimento CNPq: CIENCIAS BIOLOGICAS::ZOOLOGIA
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/8822

Statistic modeling is an important mathematical tool and has been widely used to create species distribution models. This essay utilizes the information provided by three projects, performed at the watersheds of Uruguai, Jacuí and Camaquã to generate predictive maps of occurrence for three species of large fish, such as Salminus brasiliensis (Dourado), Megaleporinus obtusidens (Piava) and Prochilodus lineatus (Grumatã). The models utilized the independent variables basin areas and altitude, with logarithmized values, obtained by a Digital Terrain Model (DTM), in this way, one can identify the dynamics of occurrence of the migratory species and analyze their distribution patterns. Using data from the interviews conducted in the previous works, with 376 available sampling points, and additional 52 points of presence collected in databases. Logistic regression was performed through the software SPSS Statistics 17.0, represented by the formula: P = e(b0+b1.x1+b2.x2). (1+e(b0+b1.x1+b2.x2))-1, where: P is probability of occurrence of the species (0-1), x1, x2 are the environmental describers already logarithmized, altitude and basin are, respectively; b0, b1 e b2 are model calibration coefficients, obtaining results that showed a good adhesion percentage, around 80% for the three species. After estimating descriptors and calculating the coefficients, presence probability maps for the three species were constructed in IDRISI Andes software, applying the regression over the elevation layers (ln) and watershed area (ln) as independent variables. To create the predicted distribution maps of each species by means of maximum entropy approach (MaxEnt program), we used only the presence of points of the sample matrix with the addition of the museum survey. By using the same variables, we stipulated 15 replicates in each case, with random test percentage of 25% for internal validation. Obtained results for area under the curve (AUC) were all above 0.9 for the three species. Comparing sensitivity and specificity obtained from both models, logit models showed higher sensitivity whereas the models generated by MaxEnt had a higher specificity. In general, the overall accuracy was high for both models for all species. The two models obtained for the three species followed the pattern already established, indicating increased probability of occurrence of these species at higher basin areas and lower altitudes, suggesting that the parameters are suitable for estimating the distribution pattern for migratory species.