大气对流过程影响着地球系统的能量和水循环。陆气之间通量交换的水平非均一性又显著影响着对流过程。全球气候模式中次网格对流过程无法显式解析需参数化，这就给模拟带来了极大的不确定性。基于此，超级参数化被提出和应用，即在传统大气环流模式每个网格内嵌入一定数目的云分辨模式来显式解析次网格对流过程。但是，其依然采用网格平均量来进行陆气之间通量交换，这给边界层和对流的模拟带来了很大的误差。本项目将改进超级参数化，使其在次网格尺度直接显式进行陆气通量交换。通过改进后的模式，研究全球不同区域不同时间尺度陆气通量交换非均一性对对流影响。改进后的模式模拟数据将被用于机器学习提高训练样本质量，改善其对大气次网格过程的参数化。本项目有望突破采用高分辨过程模式研究陆气通量交换非均一性的区域和时间限制，解决基于标准超级参数化模拟机器学习的大气次网格参数化方案不包含陆气通量交换非均一性问题，推动我国模式发展。

Convection impacts the energy and water cycles of the Earth system while Land-atmosphere flux heterogeneities influence convection. Current global climate models (GCMs) have to parameterize subgrid convection because it cannot be resolved, which introduces large uncertainties in climate simulation. Therefore, superparameterization, in which a large number of cloud-resolving models (CRMs) are embedded in the GCM, is proposed and widely used to explicitly represent subgrid processes (e.g., convection). However, in this approach, the land-atmosphere fluxes are still grid-mean averaged which results in large biases of the simulation of the boundary layer and convection processes. This proposal aims to improve the superparameterization by explicitly representing subgrid land-atmosphere fluxes. With the improved superparameterization, the impact of land-atmosphere flux heterogeneities on convection for different time slices on a global scale will be investigated. The corresponding simulation data will be further used in machine learning to parameterize atmospheric subgrid processes. This project is expected to eliminate the spatiotemporal limit of high-resolution process models in the investigation of the impact of land-atmosphere flux heterogeneities on convection. Also, it will introduce the information of subgrid land-atmosphere fluxes in trials used for the machine learning to parameterize atmospheric subgrid processes, which will advance the model development in China.