[Characterization and Prediction Analysis of Spatial and Temporal Distribution of Carbon Stock in Erhai Basin Based...

Highland lake basins， with their high ecological sensitivity， fragmented land types， and significant human intervention， are typical areas for revealing the interaction between carbon storage spatial patterns and land use conflicts. Understanding the mechanisms underlying the evolution of carbon storage patterns is crucial for comprehending the carbon sink response of high-altitude lakes and supporting the implementation of the "dual carbon" strategy. The study took the Erhai Basin as a typical representative region， constructed an ANN（MLP）-InVEST coupled model， analyzed the characteristics of carbon storage evolution from 1992 to 2022， and simulated the distribution patterns of carbon storage under different land use scenarios in 2032 and 2042. The results show that： ① The ANN（MLP）-InVEST model simulated the land use and carbon storage changes with a good performance， Kappa coefficient of 0.89， and overall accuracy of 0.91， which verified the reliability of its spatial simulation and carbon storage prediction. ② From 1992 to 2022， the overall carbon stock in the Erhai Basin showed a spatial pattern of "high value at the edge and low value in the center，" with the carbon stock increasing from 250.49×10⁵ t to 288.7×10⁵ t. ③ Mountainous areas， with stable growth in carbon storage， are the core area for carbon sink enhancement； dams and lakes， with weak carbon storage functions， urgently need to strengthen ecological restoration and land use structure optimization. ④ The carbon storage structure of Erhai Basin from 2032 to 2042 will be stable as a whole， but the spatial heterogeneity will be intensified， in which the carbon storage in the mountainous area will maintain a high value， and it is the main support area for carbon sinks. The carbon storage in the dam area will continue to decline， the wetland degradation in the lake area will be significant， and the function of the regional carbon sinks will show a trend of differentiation. The coupled model framework of "deep learning + ecological simulation" also provides scientific reference for the assessment of carbon storage dynamics and land use decision-making in highland lake basins.