Bipartite network with heterogeneous features for modelling stingless bee preferred habitat

Graphs are useful in abstracting the real-world systems. The habitat suitability studies employing graph-theoretic approaches mostly focus on the general characteristics of these systems. Most studies seldom account for the heterogeneous features present within individual nodes, thereby limiting the contribution of heterogeneity to the distinct characteristics of each node. Existing network model on bees, including stingless bees, focus on the network metrics such as modularity, robustness, and network diversity that do not incorporate the characteristics of individual bees and the habitat they are found. Bipartite modelling for stingless bee habitat is scarce in Southeast Asia, despite agricultural demand and pollinator mapping alternatives. Thus, the bipartite network approach is employed in this study to formulate the preferred habitat of stingless bees, helping to mitigate under-functioning preferred habitat locations. This research aims to characterise the graph structure of habitat preference for stingless bees, model the preferred habitat of stingless bees through bipartite network approach by incorporating the physical characteristics and environmental properties of every node and evaluate the formulated bipartite preferred habitat network model of stingless bees by verification and validation. The study uses Malaysian Agricultural Research and Development Institute (MARDI) data, on the abundance and diversity of stingless bees in Malaysia. The methodology employed in this research comprises three stages, adapted from the bipartite network modelling methodology framework. The location node data include weather data and physical features of the location, while the meliponine node data include the bee morphometrics. Parameter quantification uses machine learning algorithms. Consequently, the Diversity Abundance Bipartite Network (DivABNet) Model, comprising 12 location nodes, 12 meliponine genus-level nodes, and 107 edges linking the bipartite nodes, is formulated. The three top-ranked locations, based on Habitat Preference Indicator (HPI), are L3 (Lata Kekabu Recreational Forest, Perak), L2 (Titi Hayun Recreational Forest, Kedah), and L7 (Pulau Tekak, Tasik Kenyir, Terengganu), which are preferred by stingless bees are high in abundance and diversity. The three top-ranked bee genera are M2 (Geniotrigona), M1 (Heterotrigona) and M8 (Tetrogonilla). Evaluation of the DivABNet Model through benchmark verification shows that the Root Mean Square Error (RMSE) values for both node types have fulfilled the threshold value of no greater than 0.05. As for analytical verification using the Spearman Rank Correlation Coefficient (SRCC), both node types have passed the threshold value of 0.70. Validation of the model using previous study results also reports an SRCC that has passed the threshold value of 0.70. A limitation of the study is the use of genus rather than species for bee grouping. Thus, future studies are recommended to consider species level modelling to enhance taxonomic and ecological resolution.