基于模型数据融合的长白山阔叶红松林碳循环模拟

 充分、有效地利用各种陆地生态系统碳观测数据改善陆地生态系统模型, 是当前我国陆地生态系统碳循环研究领域亟待解决的重要问题之一。该研究以2003~2005年长白山阔叶红松林的6组生物计量观测数据和涡度相关技术测定的碳通量数据为基础, 利用马尔可夫链-蒙特卡罗方法对陆地生态系统模型的关键参数(即碳滞留时间)进行了反演, 进而预测了长白山阔叶红松林生态系统碳库、碳通量及其不确定性。反演结果表明, 长白山阔叶红松林叶凋落物和微生物碳的平均滞留时间最短, 为2~6个月; 其次是叶和细根生物量碳, 二者的平均滞留时间为1~2 a; 慢性土壤有机碳的平均滞留时间为8~16 a; 碳在木质生物量和惰性土壤有机质库中的滞留时间最长, 平均滞留时间分别为77~109 a和409~1 879 a。模拟结果显示, 碳库和累积碳通量模拟值的不确定性将随着模拟时间的延长而增大。当气温升高10%和20%时, 长白山阔叶红松林总初级生产力年总量将分别增加6.5%和9.9%, 净生态系统生产力(NEP)年总量的变化取决于土壤温度的变化。若土壤温度保持不变, NEP年总量将分别增加11.4%~21.9%和17.6%~33.1%; 若土壤温度也相应升高10%和20%, NEP年总量的增幅反而下降甚至低于原来的水平。假设气候和植被保持在2003~2005年的状态, 2020年长白山阔叶红松林NEP年总量为(163±12) g C·m^–2·a^–1, 土壤呼吸年总量为(721±14) g C·m^–2·a^–1。马尔可夫链-蒙特卡罗方法是反演模型参数、优化模拟结果和评估模拟结果不确定性的有效方法, 但今后仍需在惰性土壤碳滞留时间的估计、驱动数据和模型结构的不确定性分析、模型数据融合方法方面进行深入研究, 以进一步提高碳循环模拟的准确性。

CARBON CYCLE MODELING OF A BROAD-LEAVED KOREAN PINE FOREST IN CHANGBAI MOUNTAIN OF CHINA USING THE MODEL-DATA FUSION APPROACH

Aims Our objective was to use multiple terrestrial carbon observations to improve existing terrestrial ecosystem models. Methods We conducted a Bayesian probabilistic inversion to estimate the key parameter (i.e., carbon residence time) of a terrestrial ecosystem model (TECO) by using biometric and eddy covariance flux data measured at a temperate broad-leaved Korean pine forest in Changbai Mountain (CBS) of China from 2003 to 2005. We then estimated carbon stocks, carbon fluxes and uncertainties with posterior estimates of parameters. Biometric measurements consisted of foliage biomass, fine root biomass, woody biomass, litterfall, soil organic matter (SOM) and soil respiration. Important findings Residence times of carbon for most pools can be constrained by eddy covariance flux and biometric measurements, except for the passive soil organic matter pool. Estimated residence times of C ranged from 2 to 6 months for litter and microbial biomass pools, 1 to 2 years for foliage and fine root biomass, 8 to 16 years for slow SOM pool and 77-109 and 409-1 879 years for woody biomass and passive SOM pools, respectively. Model results showed that the prediction uncertainties of carbon stocks and accumulated carbon fluxes increased with time. When air temperature increased 10% and 20%, annual gross primary productivity (GPP) increased 6.5% and 9.9%, but annual net ecosystem productivity (NEP) changed with soil temperature. If soil temperature is constant, annual NEP increased 11.4%-21.9% and 17.6%-33.1%, while if soil temperature increased 10% and 20%, annual NEP decreased to a level that was lower than that under ambient temperature. Given the same climate condition and seasonal variation for leaf area index during 2003-2005, annual NEP and soil respiration in 2020 would be 163±12 and 721±14 g C·m~(-2)·a~(-1). Markov Chain Monte Carlo method is an effective way to estimate model parameters and to evaluate model prediction uncertainties. However, more studies are needed on a) estimation of residence time of C for passive soil organic matter, b) uncertainty analysis of input data and model structure and c) model-data fusion methods so as to improve the prediction accuracy of terrestrial ecosystem models.