The use of multisource spatial data for determining the proliferation of stingless bees in Kenya

Abstract

Stingless/meliponine bees are eusocial insects whose polylactic nature enables interaction with a wide variety of wild plants and crops that enhance pollination and, hence, support ecosystem services. However, their true potential regarding pollination services and honey production is yet to be fully recognized. Worldwide, there are over 800 species of meliponine bees, with over 20 species documented on the African continent. Out of these, only 12 species have been well documented in Kenya. Moreover, interest on meliponine bees has increased amid climate change, agricultural intensification, and other anthropogenic effects. Generally, stingless bees are under-researched, with no previous documented evidence of their ecological niche (EN) distribution in most African countries. Hence, this study sought to establish the influence of bioclimatic, topographic, and vegetation phenology on their spatial distribution and change patterns. Stingless response variables from 490 sample points were collected and used in conjunction with 11 non-conflating features to build stingless ecological niche models. Six machine learning-based EN models were used to predict the distribution of seven stingless bees’ species combined. The results from the EN models showed that annual precipitation was the most influential variable to stingless bee distribution (contributing 43.09% logit), while potential evapotranspiration and temperature seasonality contributed 21.18% of the information needed to predict the spatial distribution of stingless bees. Vegetation phenology (21.36%) and topography (14.36%) had moderate effect on stingless bees’ distribution. On the other hand, high seasonality in precipitation and temperature indicated high stingless niche variability in the future (i.e. 2055). The performance of six EN algorithms used to predict distribution of stingless bees was found to be “excellent” for random forest (true skills statistics (TSS) = 0.91) and ranger (TSS = 0.90) and “good” for generalized additive models (TSS = 0.87), multivariate adaptive regression spline (TSS = 0.80), and boosted regression trees (TSS = 0.80), while they were “fair” for recursive portioning and regression trees (TSS = 0.79). These EN models could be utilized to inform stingless bee farming and insects pollinated crops by highlighting regions that provide highly suitable conditions for stingless bees. Additionally, the findings could be harnessed to increase both bee and agricultural productivity and forest conservation efforts through supplementary pollination services.

Keywords

• stingless bees
• climate change
• vegetation phenology
• artificial intelligence
• machine learning
• species distribution modeling (sdm)