Afrotropical Diptera entomofauna conservation: technology application for Diptera identification in Afro-subtropical savannah ecosystems, Zimbabwe

Abstract

Flies in the insect order Diptera are important to humans as providers of food, medicines, and for ecosystem services such as pollination and disease vectors, among others. Patterns of Diptera diversity, distribution, and conservation status remain less explored in the Afrotropical region (AR). For some areas, Diptera inventories are incomplete, some keys for certain families are outdated, under revision, or incomplete, making the work for taxonomists in this field extremely difficult. Despite advancements in Artificial Intelligence (AI) and machine learning (ML), there has been relatively limited focus on the application of these tools for the identification of Afrotropical Diptera (AD). This work focused on exploring fly diversity and distribution in a typical protected area in Zimbabwe, Matopos National Park (MNP). This study involved sampling dipterans in three habitats (i.e., mountains, woodlands, and grasslands) and across seasons (i.e., cool-dry, hot-dry, and hot-wet seasons) in MNP, between December 2021 – April 2025, except the year 2023. Sampling was done using pan traps of three colours (yellow, white, and blue) mixed with a drop of liquid soap to break surface water tension. A total of 1838 individuals were collected across habitats and seasons. Of these, 58 dipterans were identified to species level, and 94 genera belonging to 29 Diptera families. The GLM (Generalized linear model) results revealed that species abundance was highest in GL (Grasslands), and Mt (Mountains), but least in the WL (Woodlands). Furthermore, 2022 had the highest species abundance, but 2025 had the least abundance. Species richness was highest in GL, while lowest in WL. However, Hw (Hot-wet) had the highest richness, while Hd (Hot-dry) had the least richness. Similarly, 2022 had the highest species richness, while 2025 had the least richness. Regarding Shannon diversity, GL had the highest diversity, while Mt had the least diversity. Hd had the highest diversity, while Cd (Cooldry) had significantly lower species diversity. The year 2022 had the highest diversity, on the contrary, species diversity was significantly low in 2024. The logistic regression models revealed that the probability of encountering Anthomyiidae, Muscidae, and Platystomatidae were high during the Cd compared to other seasons, whereas the probability of Asilidae, Rhiniidae, and Tabanidae was high in Hw. However, the probability of encountering the family Lauxaniidae were high during the Hd. A model for species identification was trained on Google Teachable Machine 2.0, exported as a TensorFlow Lite, and integrated into a mobile application to identify 30 species. The 90 images (i.e., 3 images from each morpho-species) and the confusion matrix were used to vi | P a g e

test for the accuracy of the application. The application retained an accuracy score of 100% which makes it promising for enhanced species identification. This study concluded that Diptera sampling is still insignificant in the AR, particularly in Zimbabwe, where world-heritage sites like MNP lack inventories and checklists of AD. The success of the mobile application offers a tremendous opportunity in automated species identification, especially in resource-limited countries like Zimbabwe. More trainings must be done to raise awareness of dipteran pollination in ecosystems.