Lepidoptera is a widely geographically distributed insect order that plays crucial ecological roles in forest and grass-shrub ecosystems as both pollinators and herbivores. This study compared Lepidoptera species composition and diversity between semi-arid forest and grass-shrub ecosystems in northern China's Pangquangou National Nature Reserve to provide scientific insights for biodiversity conservation and pest management. Nine representative plots (five forest and four grass-shrub) were established, and Lepidoptera communities were analyzed using an integrated approach combining DNA metabarcoding, DNA barcoding, and classical morphological methods. A total of 2944 individuals were collected, representing 453 species from 80 genera and 31 families, with an 82% cross-method species identification match. Despite being more time-consuming, metabarcoding offers a clear advantage for large-scale biodiversity monitoring due to its extremely low per-sample cost (up to 79.2% of costs saved compared to traditional morphological methods and 87.3% of costs saved compared to DNA barcoding). However, morphological validation remains necessary to ensure accurate identification. Dominant families showed higher relative abundance in forests (85.07%) than in grass-shrub areas (75.18%). Alpha diversity, assessed via the Shannon-Wiener index, exhibited no significant difference between ecosystems (p = 0.72). Beta diversity analysis also indicated comparable community compositions, though differences were not statistically significant (p = 0.332). Five dominant families (Noctuidae, Notodontidae, Geometridae, Erebidae, and Nymphalidae) were shared between ecosystems, while Lasiocampidae occurred exclusively in forests. Notably, metabarcoding data accurately reflected species composition and abundance, facilitated precise prediction of caterpillar pest outbreaks, and supported targeted control measures. This study confirms that DNA metabarcoding, when supplemented with morphological verification, is a robust tool for Lepidoptera biodiversity assessment. Our findings provide reliable solutions for ecosystem monitoring and pest management, underscoring the method's potential for large-scale biodiversity studies.