遥感同化动力学模型的南疆骏枣氮素动态监测方法与运移模拟

Southern Xinjiang has become the most important production base of high quality red dates due to the superior natural conditions. However, a series of problems have been caused because long-term unreasonable fertilization, such as inefficient use of nitrogen, decline of yield and quality, and environmental pollution. In those problems, detection of nitrogen changes is the premise of more fully fertilizing of nitrogen. This project focuses on jun jujube which is widely planted in southern Xinjiang. First of all, using EFAST(extended Fourier amplitude sensitive test) method to analysis global sensitivity and uncertainty of the EU_TOTATE_N model(Nitrogen dynamic model) parameters, correct and optimize parameters of the model. Using mathematical methods to study the influence mechanism and the function relation between salt, excessive nitrogen and nitrogen absorption, migration of the jujube tree in order to improve EU_TOTATE_N model. The second, the method of assimilating remote sensing into nitrogen dynamic model is implemented from researching different methods for retrieving Leaf Area Index, the selection of the assimilation time and space and the accuracy, efficiency and the ability to deal with uncertainty of some mainstream assimilation algorithms. And lastly, on the basis of above research results, using GF data to continuously monitor the dynamic change of nitrogen in time and space and using visualization technologies to simulate the migration process. Those problems including migration mechanism between the soil and jujube nitrogen, problems of discrete time, mechanism of remote sensing monitoring nitrogen are expected to be solved. The result of this project research have important scientific significance for promoting mechanism and high-precision of N remote sensing monitoring, moreover, it can provide theory basis for optimally fertilizing of jun jujube in southern Xinjiang.

南疆是我国重要的优质红枣生产基地，但长期不合理的施肥导致氮素利用效率低下、红枣产值下降和环境污染等一系列问题，快速有效地监测氮素变化是科学施肥的前提。项目以种植面积最广的骏枣为研究对象，首先，采用EFAST方法分析EU_TOTATE_N模型（氮动力学模型）参数的全局敏感性和不确定性，利用数学建模手段研究盐分和过量氮对枣树氮吸收和运移的影响机理及定量关系，以优化和改进EU_TOTATE_N模型；其次，从同化观测量叶面积指数反演、同化时间和空间尺度及主流同化算法的精度、效率和处理不确定性问题能力的系统分析等方面开展研究，实现遥感和氮动力学模型同化方法；最后，完成区域尺度的氮素时空动态变化监测与运移模拟，揭示土壤和枣树氮素之间的运移机制，解决氮素遥感监测时间不连续、机理性不强的问题。项目研究不仅对促进枣树氮素遥感监测机理化和精确化具有重要的科学意义，而且可为南疆枣园科学施肥提供理论依据。

首先，在广泛使用的肥料量的情况下，使用三个生长季节田间试验数据矫正和验证了WOFOST模型。模拟的萌芽期，开花期和成熟期的物候发育阶段的误差分别是–2，–3和–3天。叶片，茎，果实，总生物量和叶面积指数(LAI)的模拟生长动态与测量值吻合很好，显示的RMSE（均方根误差）值分别为0.14、0.33、0.37、0.62 t/ha和0.19, R^2(决定系数)值分别为0.95、0.98、0.99、0.99和0.95。.其次，假设LAI变化由氮素和其他营养差异导致，进而产生产量差异，尝试将接近最大发育阶段的单个遥感反演的LAI同化到经过校准的WOFOST模型，以提高当地枣园田间尺度的果实产量估算。与未同化模拟相比，强迫LAI后的同化提高了产量估算精度，2016年预测产量的R^2为0.62，RMSE为0.74(11.3％)t/ha，2017年的R^2为0.59，RMSE为0.87(11.3％)t/ha。.最后，研究的主要贡献是开发了SUBPLEX同化框架，将主要生长阶段的时间序列的遥感LAI同化到校准的WOFOST模型中，并将SUBPLEX算法的产量估算精度与广泛使用的集成卡尔曼滤波器(EnKF)同化进行了比较。结果表明，与未同化模拟相比，SUBPLEX和EnKF同化均显著提高了产量估算性能。SUBPLEX(R^2=0.78，RMSE=0.64(8.3％)t/ha和RPD(标准偏差(SD)/RMSE)=2.13)的同化精度略高于EnKF同化(R^2=0.73，RMSE=0.71(9.2％)t/ha和RPD=1.91)。该研究可以提供基于SUBPLEX算法的新同化方案来改善田间尺度的水果作物产量估算。