Long-term observations uncover sustained carbon dioxide emissions from lakes following aquaculture retreat.

Aquaculture has emerged as one of the dominant anthropogenic drivers altering the carbon dioxide (CO2) balance of lake ecosystems. Although aquaculture retreat (pen removal) has been widely implemented in lakes, its long-term effects on CO2 variability remain poorly constrained due to insufficient field measurements, thereby hindering our understanding of anthropogenic carbon cycling. To address this knowledge gap, we investigated the partial pressure of carbon dioxide (pCO2) and CO2 flux in two typical aquaculture-affected lakes in eastern China (Lake Yangcheng, LYC; Dongtaihu Bay of Lake Taihu, DTH) based on long-term (2000-2015) field measurements. Results revealed that both systems were net CO2 sources, yet LYC with high nutrient loadings exhibited significantly higher emissions (108 ± 56 g C m-2 yr-1; pCO2: 1180 ± 390 μatm) than DTH (37 ± 24 g C m-2 yr-1; pCO2: 716 ± 185 μatm). Crucially, CO2 emissions in LYC exhibited a counter-intuitive "V-shaped" trajectory in response to management interventions: declining by 54 % during the active removal phase, but rebounding by 56 % in the post-restoration phase. This unexpected rebound was driven not merely by nutrient legacy, but fundamentally by the loss of submerged vegetation, thereby severely compromising the ecosystem's CO2 uptake capacity. In contrast, DTH maintained stable emissions under a moderate retreat strategy mostly due to the stabilizing effect of macrophyte retention. The Random Forest (RF) model further showed that aquaculture area functions as a non-linear regulator of emission intensity, challenging linear scaling assumptions. Our findings demonstrate that successful carbon mitigation in aquaculture systems relies not merely on nutrient load reduction but fundamentally on the structural recovery and functional integrity of the autotrophic carbon sink.