Predicting Lokta (Daphne bholua) distribution using machine learning algorithms: Implications for sustainable management

Machine learning has markedly improved the accuracy of species distribution modelling, yet its use in the management of Non-Timber Forest Products remains limited. This gap is especially evident in Nepal for Lokta (Daphne bholua). Although Lokta has long been valued for traditional use and trade, its spatial distribution and production ecology have received little scientific attention, limiting efforts to manage it sustainably. Against this background, this study provides the first machine-learning-based prediction of Lokta distribution in Nepal and examines how climatic, edaphic, and topographic factors shape its occurrence. Data were collected from 396 randomly established 100 m 2 plots in Dolakha district, Bagmati Province, Nepal, of which 161 recorded the presence of Lokta. Four machine learning models were used to predict its distribution and generate habitat suitability maps based on climatic, edaphic, and topographic variables. Among the tested models, Extreme Gradient Boosting (XGB) delivered the best predictive performance (AUC = 0.93; accuracy = 86%), outperforming Random Forest, Artificial Neural Networks, and K-Nearest Neighbours. The model estimated that about 43% of the 100,311 ha of forest area is suitable for Lokta growth. Precipitation during the driest quarter was the strongest predictor, followed by elevation and soil texture, indicating that seasonal moisture, terrain, and soil conditions jointly shape Lokta habitat suitability. These findings provide a stronger scientific basis for targeting Lokta cultivation and enrichment planting in ecologically suitable areas, while also informing soil and water conservation measures to support natural regeneration and production in marginal habitats. More broadly, the study shows the value of machine learning, particularly XGB, for identifying suitable Lokta habitats and supporting conservation planning, sustainable harvesting, and spatially informed management at community forest, municipal, and national levels.