Hokkaido University, Graduate School of Agriculture, Kita-9 Nishi-9 Kita-Ku, Sapporo 060-8589, Japan.
Hokkaido University, Research Faculty of Agriculture, Kita-9 Nishi-9 Kita-Ku, Sapporo 060-8589, Japan.
Sci Total Environ. 2019 Jun 10;668:696-713. doi: 10.1016/j.scitotenv.2019.03.025. Epub 2019 Mar 4.
Remote sensing (RS)-based models play an important role in estimating and monitoring terrestrial ecosystem gross primary productivity (GPP). Several RS-based GPP models have been developed using different criteria, yet the sensitivities to environmental factors vary among models; thus, a comparison of model sensitivity is necessary for analyzing and interpreting results and for choosing suitable models. In this study, we globally evaluated and compared the sensitivities of 14 RS-based models (2 process-, 4 vegetation-index-, 5 light-use-efficiency, and 3 machine-learning-based models) and benchmarked them against GPP responses to climatic factors measured at flux sites and to elevated CO concentrations measured at free-air CO enrichment experiment sites. The results demonstrated that the models with relatively high sensitivity to increasing atmospheric CO concentrations showed a higher increasing GPP trend. The fundamental difference in the CO effect in the models' algorithm either considers the effect of CO through changes in greenness indices (nine models) or introduces the influences on photosynthesis (three models). The overall effects of temperature and radiation, in terms of both magnitude and sign, vary among the models, while the models respond relatively consistently to variations in precipitation. Spatially, larger differences among model sensitivity to climatic factors occur in the tropics; at high latitudes, models have a consistent and obvious positive response to variations in temperature and radiation, and precipitation significantly enhances the GPP in mid-latitudes. Compared with the results calculated by flux-site measurements, the model performance differed substantially among different sites. However, the sensitivities of most models are basically within the confidence interval of the flux-site results. In general, the comparison revealed that models differed substantially in the effect of environmental regulations, particularly CO fertilization and water stress, on GPP, and none of the models performed consistently better across the different ecosystems and under the various external conditions.
遥感（RS）模型在估算和监测陆地生态系统总初级生产力（GPP）方面发挥着重要作用。已经使用不同的标准开发了几种基于 RS 的 GPP 模型，但模型对环境因素的敏感性存在差异；因此，有必要对模型的敏感性进行比较，以分析和解释结果，并选择合适的模型。在本研究中，我们在全球范围内评估和比较了 14 种基于 RS 的模型（2 种过程模型、4 种植被指数模型、5 种光能利用效率模型和 3 种基于机器学习的模型）的敏感性，并将其与通量站点测量的气候因素对 GPP 的响应以及自由空气 CO 富集实验站点测量的 CO 浓度升高进行了基准测试。结果表明，对大气 CO 浓度升高相对敏感的模型表现出较高的 GPP 增长趋势。模型算法中 CO 效应的基本差异要么通过绿色指数的变化来考虑 CO 的影响（9 个模型），要么引入对光合作用的影响（3 个模型）。温度和辐射的总体效应，无论在幅度还是符号上，在模型之间都存在差异，而模型对降水的变化反应相对一致。在空间上，模型对气候因素的敏感性在热带地区差异较大；在高纬度地区，模型对温度和辐射的变化有一致而明显的正响应，降水在中纬度地区显著增强 GPP。与通量站点测量计算的结果相比，不同站点之间模型性能存在很大差异。然而，大多数模型的敏感性基本上都在通量站点结果的置信区间内。总体而言，比较结果表明，模型在环境法规对 GPP 的影响方面存在很大差异，特别是 CO 施肥和水分胁迫，并且没有一个模型在不同的生态系统和各种外部条件下表现一致更好。
