The effect of flowering stage on distribution modelling performance: A case study of Acacia dealbata using maximum entropy modelling and RPA images

Abstract

Aim of study: To classify and validate the coverage of Acacia dealbata by stratifying its area into three different flowering stages using remotely piloted aircraft (RPA)-derived image orthomosaics. Area of study: We selected three sites in the west of Ourense province (Galicia, Spain). This area is the eastern cluster of A. dealbata populations in Galicia. Material and methods: We used a multirotor RPA equipped with an RGB and a multispectral camera. The flights were carried out on 10th and 11th March 2020. We performed a visual interpretation of the RGB orthomosaics to identify the patches of A. dealbata in three different flowering stages. We then used a maximum entropy (MaxEnt) programme to estimate the probability of A. dealbata presence in each study site at each of the three flowering stages. Main results: The performance of the MaxEnt models for the three flowering stages in each of the three study sites were acceptable in terms of ROC area under the curve (AUC) analyses the values of which ranged from 0.74 to 0.91, although in most cases was greater than 0.80, this being an improvement on the classification without stratification (AUC from 0.73 to 0.86). Research highlights: Our approach has proven to be a valid procedure to identify patterns of species distributions at local scale. In general, the performance of the models improves when stratification into flowering stages is considered. Overall accuracy of the presence prediction maps ranged from 0.76 to 0.91, highlighting the suitability of this approach for monitoring the expansion of A. dealbata.