This study is designed to identify strategies for substituting or reducing the use of sulfur dioxide (SO₂) while improving the quality of semi-sweet wines. Using the ‘Petit Manseng’ semi-sweet wine as the research object, the effects of different concentrations and combinations of glutathione (GSH), mannan (Man), vitamin C (VC), chitosan oligosaccharide (COS), and SO₂ were investigated. The results show that under VC treatment, the DPPH free radical scavenging rate of the wine is higher than that of other antioxidant treatments, with the V-1 treatment being 16.9% higher than the CK treatment. The wine treated with 30 mg/L of SO₂ has the highest reducing power. After COS treatment, the wine color deepens, making it unsuitable for sweet white wines. GSH and Man can delay the increase in OD240 and the decrease in total phenolic. Man can increase the content of sweet amino acids such as serine. The treatments with man and GSH result in a greater variety of aromatic compounds and higher content of epigallocatechin. The combination treatment of GSH, Man, and SO₂ resulted in semi-sweet wines with a bright color, rich aroma, and full-bodied, balanced taste, achieving the best quality indicators and the highest sensory evaluation. This wine exhibits a light yellow color, full body, and a complex, balanced aroma, which are improvements in quality compared to semi-sweet wines with 30 mg/L of SO₂ added, and it reduces the use of SO₂ by 33.3%. This study shows that adding 5 mg/L of GSH, 5 mg/L of Man, and 20 mg/L of SO₂ can significantly improve the antioxidant properties and quality of semi-sweet wines.

Phenolic compounds are critical components that determine the processing quality of wine grapes, and their accumulation is jointly affected by environmental conditions and genotype. However, the respective contributions of production site and cultivar, as well as the response differences in various phenolic components to environmental changes, remain unclear in northern China. In this study, three major red wine grape cultivars (Vitis vinifera L. cvs. Marselan, Cabernet Franc, Cabernet Sauvignon) were sampled from three typical producing areas, including Turpan, the Hexi Corridor, and Huailai in 2024. Botanical characteristics, quantitative agronomic traits, physicochemical indexes, and five phenolic components in berry skins and seeds were determined with three biological replicates. A general linear model (GLM) was used to quantify the effects of production site and genotype on berry quality, and principal component analysis (PCA) and Pearson correlation analysis were performed to explore the relationships between quality indicators. Our results indicate that clear quality differences were observed among the three studied production sites, which were the main source of variation in this dataset, accounting for 65.7% to 82.1% of total variance across most traits. Three distinct regional patterns were observed: grapes from Xinjiang had the highest sugar content, those from Gansu had the highest phenolic levels, and those from Zhangjiakou were richest in anthocyanins. The responsiveness of individual phenolic components to ecological conditions varied markedly, with anthocyanins being the most sensitive, followed by proanthocyanidins, tannins, total phenolics, and flavonoids. Marselan maintained a stable genotypic advantage in proanthocyanidin accumulation across all three sites, with skin levels 12.5–230.5% higher than other cultivars under the same growing conditions. Additionally, a potential tissue-specific association between cluster weight and phenolic components was preliminarily observed: higher cluster weight was associated with lower concentrations of late-accumulating anthocyanins and proanthocyanidins, but did not negatively affect total phenolics and tannins that form in early fruit development. Based on the above results, preliminary suggestions for cultivar–site matching were provided for reference. This study provides preliminary baseline data for cultivar selection and viticultural management in northern China, pending further multi-year validation.