Permafrost soils store a large amount of nitrogen (N) which could be activated under the continuous climate warming. However, compared with carbon (C) stock, little is known about the size and spatial distribution of permafrost N stock. By combining measurements from 519 pedons with two machine learning models (supporting vector machine (SVM) and random forest (RF)), we estimated the size and spatial distribution of N stock across the Tibetan alpine permafrost region. We then compared these spatially-explicit N estimates with simulated N stocks from the Community Land Model (CLM). We found that N density (N amount per area) in the top three meters was 1.58 kg N m-2 (interquartile range: 1.40-1.76) across the study area, constituting a total of 1802 Tg N (interquartile range: 1605-2008), decreasing from the southeast to the northwest of the plateau. N stored below 1 m accounted for 48% of the total N stock in the top three meters. CLM4.5 significantly underestimated the N stock on the Tibetan Plateau, primarily in areas with arid/semi-arid climate. The process of biological N fixation played a key role in the underestimation of N stock prediction. Overall, our study highlights that it is imperative to improve the simulation of N processes and permafrost N stocks in land models to better predict ecological consequences induced by rapid and widespread permafrost degradation.
Keywords: Climate warming; Community Land Model; Machine learning; Nitrogen cycle; Permafrost; Tibetan Plateau.
Copyright © 2018. Published by Elsevier B.V.