Abstract: Prevention and control of soil erosion can be represented as one of the most important protocols in the ecological environment. Particularly, the soil and water conservation can be subjected to significant land-use variation in the ecologically fragile regions. Among them, the red soil hilly region of southern China is highly susceptible to soil erosion. The rapid expansion of cash crops (like citrus orchards) has introduced new dynamics to soil erosion and carbon cycling. This study aims to explore the spatiotemporal pattern and driving factors of soil erosion, conservation and carbon sequestration capacity against the citrus orchard expansion. Pingjiang River Basin was also selected as the representative targets in the critical zone of red soil erosion. The data was collected from the multi-temporal remote sensing images, land use/cover data, rainfall records, normalized difference vegetation index (NDVI), topographic data, and soil properties in the years 1995, 2005, 2015, and 2023. The object-based image analysis within eCognition software was utilized to combine with the spatial analysis in ArcGIS. Firstly, the spatiotemporal distribution of the citrus orchard areas was precisely delineated over the basin. Subsequently, the revised universal soil loss equation (RUSLE) model was applied to quantitatively assess the soil erosion dynamics. The key parameters were calculated from the datasets for each time period. According to the erosion assessment, the different land use types were evaluated for the soil conservation and carbon sequestration capacity. Finally, the geographical detector was employed to identify the primary driving factors of the spatiotemporal variations in soil conservation and carbon sequestration over these land use types. The results demonstrate that: 1) The soil erosion in the Pingjiang River Basin exhibited a significant overall decreasing trend over the study period (1995–2023). Specifically, the soil erosion modulus decreased markedly from 904.88 t/(km²·a) in 1995 to 395.30 t/(km²·a) in 2023. The year 2015 was recorded as the lowest average erosion modulus (251.28 t/(km²·a). The micro- and slight erosion grades collectively covered 94.61% of the basin area. There was a dominant shift towards the lower erosion intensities. The severe and very severe erosion grades also displayed a distinct point-like or patchy distribution pattern. Specifically for the citrus orchards, the average erosion modulus also decreased from 781.96 t/(km²·a) in 1995 to 381.54 t/(km²·a) in 2023. It also peaked sharply at 1070.6 t/(km²·a) in 2005 during the initial expansion. However, the localized intensification of the moderate and higher intensity erosion occurred in some orchard areas post-2015. This trend was attributed primarily to the citrus Huanglongbing (HLB) disease outbreaks and anthropogenic disturbances, like orchard abandonment or replanting. 2) Significant disparities existed in the soil conservation and carbon sequestration potential among different land use types. Forest land consistently exhibited the highest total carbon sequestration capacity. But there was the outstanding trend: increasing from 1995 to 2015, followed by a decrease from 2015 to 2023. Cultivated land was ranked second in total sequestration. But there was an inverse temporal trend, compared with the forest land. Furthermore, the shrubland, grassland, water bodies, construction land, and unused land generally demonstrated lower and relatively stable soil carbon content. Overall, the basin was functioned as a net carbon sink in the study periods, with the carbon sequestration in most land types; Nevertheless, the grassland and water bodies shared the minor net carbon losses. Crucially, the citrus orchards displayed a clear temporal evolution in their carbon budget: A net soil carbon loss of 3.8 ×10⁴ t in the establishment phase (1995–2005), followed by significant carbon sequestration of 7.5 × 10⁴ t (2005–2015) and 6.8 ×10⁴ t (2015–2023), as the orchards developed. 3) Factor detector analysis revealed that the vegetation coverage was the dominant explanatory factor for the variations in the soil conservation and carbon sequestration in citrus orchards. Conversely, the slope gradient was emerged as the primary explanatory factor for the variations in the forest, cultivated land, and grassland. The interaction detector indicated that all paired factors exhibited either bi-enhancement or non-linear enhancement interactions. Importantly, the interaction between slope gradient and other factors (especially, vegetation coverage and rainfall erosivity) consistently demonstrated the strongest explanatory power (highest q-values) for the driving changes in the soil conservation and carbon sequestration across all major land use types. Therefore, the future soil and water conservation strategies should prioritize the synergistic effects of the slope conditions, anthropogenic disturbances (like those induced by HLB), and vegetation cover dynamics. Particularly, those aimed enhancing the carbon sink potential. The findings can provide practical references to formulate the targeted strategies for the soil and water loss control, especially for the ecosystem carbon sequestration capacity in the ecologically sensitive red soil hilly basins under agricultural intensification. Effective decision-making on the landscape can also be made on the land-use dynamics and the key driver interactions, especially slope-vegetation coupling.