Journal Article

Knowledge‐Based Deep Learning to Predict Vegetation Carbon, Nitrogen and Phosphorus Densities in China’s Shrublands

Deng, Ying, Wang, Yang, Zhao, Changming, Zhou, Chenyang, Xiong, Gaoming, Tang, Zhiyao, Xie, Zongqiang, Li, Jiaxiang, Liu, Qing, Xu, Wenting

2024

Overview

Accurate estimations of carbon (C), nitrogen (N), and phosphorus (P) densities in shrublands are pivotal for assessing terrestrial ecosystem carbon sequestration. Combining in‐situ investigations and machine learning facilitates large‐scale patterns mapping, however, which often overlooks underlying ecological regulations. Here we utilize data from 1,122 survey plots across China's shrublands and develop a novel knowledge‐based deep learning framework that integrates a structural equation model (SEM) to elucidate mechanisms and construct an artificial neural network (ANN) based on these causal relationships. Results show that biomass allocation to different organs follows allometric regulations and that N and P concentrations maintain a degree of stoichiometric homeostasis following biological stoichiometry theory. This insight guides the construction of our ANN, which outperforms both SEM and other prevalent machine learning methods. By leveraging ecological theories to inform model construction, our framework not only enhances prediction accuracy and explainability but also provides a methodological blueprint for ecological research. Plain Language Summary China has set a goal to achieve carbon neutrality by 2060, and one way to achieve this is by utilizing terrestrial ecosystems, which can absorb CO2 from the atmosphere. The effectiveness of natural carbon sinks is often limited by the availability of essential nutrients such as nitrogen (N) and phosphorus (P). Shrublands are unique and contribute the most uncertainty in estimating China's carbon storage. Thus, accurately mapping shrubland vegetation C, N, and P densities is critical. Previous studies usually apply data‐driven methods to scale up site information to larger scales, often failing to consider underlying ecological regulations. Here, we advance this approach by integrating deep learning (DL) with causal understanding. We found that C, N, and P allocation to different organs is relatively consistent, and their ratios maintain generally stable. These relationships are then applied to the DL algorithm. The knowledge‐based DL model outperforms popular machine learning methods. Our framework not only improves the ability to predict nutrient distributions in shrublands but also serves as a blueprint for further ecological research, enhancing both the accuracy and the explainability of ecological models. Key Points Biomass and nutrient allocation follow allometry and biological stoichiometry theory Structural equation model (SEM) and artificial neural network (ANN) are combined to achieve casual interference and accurate prediction Prior knowledge‐based deep learning can advance ecological modeling

• Publisher: John Wiley & Sons, Inc,Wiley
• Subject: Accuracy / Algorithms / Allometry / Artificial neural networks / Biological effects / biological stoichiometry / Body organs / Carbon capture and storage / Carbon dioxide / Carbon sequestration / Carbon sinks / Construction / Deep learning / Ecological models / Ecological research / Essential nutrients / Homeostasis / Learning algorithms / Machine learning / Mapping / Neural networks / Nitrogen / nutrient allocation / Nutrient availability / Nutrients / Organs / Phosphorus / Regulations / Shrublands / Stoichiometry / Terrestrial ecosystems / Vegetation