Invasive plants have significantly impacted the function and biodiversity of the global ecosystem. In the context of global climate change, the effective control of invasive plants is important for maintaining the stability of grassland ecosystems. The spectral differences between invasive plants and native dominant species during phenological stages provide an opportunity for remote sensing technology to monitor their spatiotemporal distribution. Previous studies have primarily focused on single-phase monitoring and the classification of plants, with relatively fewer studies on multitemporal continuous monitoring, particularly during early phenological stages. In this study, we focused on Pedicularis kansuensis, an invasive species of the Bayinbuluk grassland of Xinjiang, using UAV-based multispectral data and machine learning algorithms to extract spatial distribution data for P. kansuensis during key phenological stages in 2023 (emergence, initial flowering, peak flowering, and senescence stages) and the peak flowering stage in 2024. We examined the feasibility of extracting P. kansuensis at each phenological stage and analyzed changes in inter-annual spatial distribution. The results indicated that (1) The random forest algorithm slightly outperformed the support vector machine, with model accuracy varying with the growth stages of P. kansuensis; specifically, peak flowering stage (late July to late August) > initial flowering stage (late June to early July) > emergence stage (mid-June) > senescence stage (mid-September). Throughout the growth season, spatial distribution during the early growth stages (emergence and initial flowering) exhibited a high spatial overlap with the peak flowering stage, and the key features were consistent with those of the peak flowering stage. This suggests that the random forest algorithm can effectively map the distribution of P. kansuensis during the emergence stage, which provides important technical support for the early-stage monitoring of invasive plants; (2) The spatial distribution of P. kansuensis exhibited significant inter-annual variation, with less than 15% spatial overlap between the two years; (3) During the growth season, the most important feature for distinguishing P. kansuensis (excluding the senescence stage), from other co-occurring species was the normalized difference index, calculated from the 555 nm and 720 nm bands, followed by the green band. Because P. kansuensis had entered the senescence stage, there was a noticeable change in feature importance, with significant differences in various background environments. These results demonstrate the feasibility of using UAV-based multispectral remote sensing technology for monitoring the early phenological stages of P. kansuensis and offer technical support for early warning and control measures.