Deep-learning and process-based models for simulating carbon and water fluxes across scales (C04)

用于模拟跨尺度碳和水通量的深度学习和基于过程的模型（C04）

• 批准号: 505879376
• 金额: --
• 依托单位: Lehrstuhl für Biometrie und Umweltsystemanalyse
• 依托单位国家: 德国
• 项目类别: Collaborative Research Centres
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/505879376?language=en
• 关键词: Deep; learning; process; based; models

C4.1 Process-based model simulationsWe will develop process-based model simulations and deep learning tools for data analysis to interact with and optimize the sensor-based monitoring, as well as to deepen our understanding of impacts of spatio-temporal heterogeneity and dynamics for total ecosystem water and carbon exchange. An existing 2D process-based model will be extended, calibrated and run in a now- and forecasting 3D mode, covering spatio-temporal heterogeneity of small-scale processes and integrating new scaling laws for non-linear interactions.C4.2 Deep learningUsing deep-learning algorithms, the plethora of data will be efficiently evaluated to distinguish between important and redundant data. The aim is to provide sufficient spatio-temporal resolution and save sensor node energy and reduce redundant data accumulation. Thereby, deep learning and process simulations interact with the sensor network through i) data assimilation from the sensor network and ii) adjustment and optimization of the measuring design based on simulated outputs and predictions

C4.1基于过程的模型模拟我们将开发用于数据分析的基于过程的模型模拟和深度学习工具，以与基于传感器的监测进行交互并优化，并加深我们对时空异质性和动态对总生态系统水和碳交换的影响的理解。现有的基于过程的2D模型将被扩展、校准，并以现在和预测的3D模式运行，涵盖小规模过程的时空异质性，并为非线性相互作用集成新的缩放律。C4.2深度学习使用深度学习算法，将有效评估过多的数据，以区分重要数据和冗余数据。其目的是提供足够的时空分辨率和节省传感器节点的能量，减少冗余数据积累。因此，深度学习和过程模拟通过i）来自传感器网络的数据同化和ii）基于模拟输出和预测的测量设计的调整和优化与传感器网络交互