%0 Conference Proceedings
%4 sid.inpe.br/marte2/2017/10.27.16.23.18
%2 sid.inpe.br/marte2/2017/10.27.16.23.19
%@isbn 978-85-17-00088-1
%F 59569
%T Una evaluación del sesgo de muestreo sobre el análisis ROC de modelos de nicho
%D 2017
%A Nicasio, Sergio,
%A Mas, Jean-François,
%A Hernández, Gabriela,
%@electronicmailaddress jfmas@ciga.unam.mx
%E Gherardi, Douglas Francisco Marcolino,
%E Aragão, Luiz Eduardo Oliveira e Cruz de,
%B Simpósio Brasileiro de Sensoriamento Remoto, 18 (SBSR)
%C Santos
%8 28-31 maio 2017
%I Instituto Nacional de Pesquisas Espaciais (INPE)
%J São José dos Campos
%P 7483-7488
%S Anais
%1 Instituto Nacional de Pesquisas Espaciais (INPE)
%X During the last decades, ecological niche modeling has become a very popular methodologyin the exploration and analysis of biodiversity data. ROC analysis is widely used to assess the modelsand high performance is often reported in the literature. However, datasets derived from opportunisticobservations often exhibit a strong geographic bias, mainly due to accessibility. This unequal coverageof a species distribution can strongly affect the quality of the model when important parts of theenvironmental space suitable to a specie are poorly represented in the survey dataset. This study aimsat assessing the performance of ROC analysis in evaluating niche models. We elaborated independantniche models for Romerolagus diazzi using MaxEnt and data obtained during different decadesseparately. Each decade based model was trained using 75% of the data and assessed using theremaining 25%. ROC analysis based on the 25% of test data presented high scores for all the models.However, ACP analysis and the comparison between the species distribution derived from the modelspresented important differences. These results suggest that ROC analysis based on a subset of the datatend to be optimistically biased because the test set is not independant from the training set and presentsoften the same bias.
%9 Modelagem espacial
%@language es
%3 59569.pdf