区域尺度农田土壤有机碳时空演变模拟的不确定性评价研究

Soil organic carbon (SOC) is a key component of soil quality that sustains many important soil functions, and thus accurate characterization of SOC spatio-temporal evolvement is extremely important for maintaining and improving the soil fertility and the long-term stability of agricultural systems. However, any modeling process of SOC dynamics may inevitably involve some uncertainties due to limitations inherent in model input data (error, sample size, and availability) and the predicting model. Consequently, quantifying such uncertainties has become critical for achieving scientific decisions of effective management practice recommendations. In this study, the spatio-temporal evolvement of SOC over the last three decades in paddy soils in southern area of Jiangsu Province were simulated by using a biogeochemical process model DNDC. Regional soil sampling, rural household survey, long-term field observation, and remote sensing image classification were used to obtain soil, agricultural management, and land use data for optimizing internal parameters of DNDC model and integrating information on land use changes. The Monte Carlo method accounting for dependence between the sensitive parameters of the model, the generalized likelihood uncertainty estimation (GLUE) with integrated up-scaling (point-to-polygon) uncerntainty of SOC observations, and the Bayesian calibration method with integrated up-scaling uncerntainty of SOC observations were employed for quantifying the uncertainties of DNDC-modeled SOC dynamics, and the contributions of different model inputs (spatial parameters) to the total uncertainties as well. Implementation of the proposed methodologies in this research proposal will largely benefit the establishment of national-scale SOC accounting and uncertainty assessment framework.

土壤有机碳（SOC）是土壤质量及功能的核心，准确把握SOC时空演变规律对稳定和提高土壤肥力具有重要意义。然而由于观测数据及预测模型等因素限制，SOC时空演变预测中必然包含一定的不确定性，而定量这些不确定性是实现土壤资源管理科学决策的关键。本研究以苏南水稻土区为典型研究区、以DNDC模型模拟SOC时空演变为例，采用区域采样、农户调查、定位观测、遥感解译等手段获取土壤、农业管理及土地利用变化等数据。通过敏感参数相关性为约束条件改进蒙特卡罗方法；以地统计学随机模拟量化区域调查点SOC实测值尺度上推的不确定性，并将其分别融合到广义似然不确定性估计和正规贝叶斯校正方法中，以改进区域尺度SOC时空演变模拟的不确定性评价过程，明确区域尺度上模型主要空间化输入数据对不确定性的贡献，建立不确定性评价方法体系，为实现国家尺度农田SOC时空演变模拟的不确定性定量表达提供理论和方法依据。

认识和理解土壤有机碳（SOC）时空演变规律对于稳定和提高土壤肥力具有重要意义。然而，由于观测数据及预测模型等因素限制，SOC演变预测中必然包含一定的不确定性，而定量这些不确定性是实现土壤资源管理科学决策的关键。本研究基于区域土壤采样、管理措施调查、土壤肥力长期定位监测等数据和地统计学、DNDC模型预测、蒙特卡洛模拟（MC）等方法，明晰了样点数量特别是样点分布模式是SOC实测值尺度上推预测不确定性的重要影响因子，而空间化输入数据中的初始SOC含量及秸秆还田比例数据所包含的不确定性则是导致过程模型模拟SOC不确定性的两个最重要来源；在量化DNDC模拟SOC的不确定性方面，MC方法尚可接受，但推断的不确定性置信区间较宽。广义似然不确定性估计（GLUE）和贝叶斯校正（Bayes）两种方法量化DNDC模拟苏南稻麦轮作水稻土SOC演变的不确定性均可行，但GLUE方法计算效率极低，而Bayes方法相对适宜；基于不确定性分析，发现近35年来的前20年间苏南地区水稻土SOC显著增加，但后15年间有降低迹象，且这种变化与土地利用变化导致的土壤C输入波动密切相关。推荐管理措施下该区SOC未来仍有增加潜力，但面临巨大挑战。总计发表第1标注论文7篇，其中SCI论文5篇。此外，本研究基于Sobol敏感性分析、Bayes校正形成了独立于过程模型的SOC演变模拟不确定性量化方法也为实现国家尺度农田SOC时空演变模拟的不确定性定量表达提供了理论和方法依据。