基于集合经验模态分解算法的土壤肥力胁迫甄别与监测

土壤肥力是农作物生长所需的关键要素之一,其水平直接影响农作物的长势甚至产量,然而作物生长过程中受多种胁迫因素的综合影响。避免土壤肥力监测结果受到其他胁迫因素的干扰是土壤肥力精准监测的关键问题之一。该研究旨在分析利用集合经验模态分解（Ensemble Empirical Mode Decomposition,EEMD）算法进行土壤肥力胁迫甄别和监测的可行性。以河北省廊坊市大厂县冬小麦耕地为研究区域,以EEMD算法为基础,将分解后的本征模函数（Intrinsic Mode Function,IMF）分量按照年际、年间和年内的尺度进行合成,结合不同的时间尺度及胁迫特征剔除土壤水分胁迫、病虫害胁迫及重金属胁迫等因素,实现对土壤肥力胁迫的有效甄别提取。结合主成分分析相关方法,将有机质、全氮、有效磷以及速效钾4项养分指标转换为3个主成分,初步得到土壤肥力综合评价简易模型,后与分解结果进行拟合,构建并实现了土壤肥力综合水平的定量评价模型。结果表明：1）在年际、年间以及年内3组波动组分中,年际波动组分可以较好地反映研究区内土壤肥力胁迫作用对农作物长势的影响;2）利用最小二乘法对土壤养分指标测定数据进行线性拟合,拟合结果与原始数据的决定系数达到了0.857,能够较好地反映出原始数据的变化水平;3）最终土壤肥力水平评价模型评价结果与实测结果间的平均误差为11.82%,表明模型预测结果与实际情况契合程度高,模型反演结果能够较好地反映研究区的土壤肥力水平。该研究结合EEMD算法与统计学分析实现了土壤肥力胁迫的有效甄别及土壤肥力定量评价模型的构建,为遥感技术在土壤肥力研究领域的应用提供了参考。

Screening and monitoring of soil fertility stress using ensemble empirical mode decomposition

Soil fertility is one of the most important elements required for crop growth. The specific level of soil fertility directly affects the growth and even the yield of crops. However, the development of crop growth depends mainly on a variety of stress factors. Therefore, it is very necessary to avoid the interference of stress factors in the monitoring of soil fertility. The purpose of this study was to accurately identify and monitor the soil fertility stress using Ensemble Empirical Mode Decomposition (EEMD). The research area was taken as the winter wheat cultivated land in the central latitude and longitude of 116.98°E and 39.88°N of the Dachang Hui Autonomous County, Langfang City, Hebei Province, China. The total area was 176.29 km2, among which the cultivated land accounted for 11 581.54 hm2. The landform was mainly plain with mostly tidal brown or tidal soil. The decomposed Intrinsic Mode Function (IMF) components were synthesized using EEMD, according to the annual, inter-annual and intra-annual scales. Different time scales and stress characteristics were combined to eliminate the stress of soil water, pest, and heavy metal, in order to obtain the effective screening and extraction of soil fertility stress. The organic matter, total nitrogen, available phosphorus, and available potassium were also collected from the field soil samples in each planting area of winter wheat. A principal component analysis was performed on the four nutrient indexes to convert them into the three principal components. A simple model was preliminarily obtained for the comprehensive evaluation of soil fertility. Finally, a quantitative evaluation model was established for the comprehensive level of soil fertility after fitting with the decomposition. The results showed that: 1) The components were mainly divided into three groups of fluctuation components after EEMD. Among them, the inter-annual fluctuation component better reflects the effect of soil fertility stress on crop growth in the study area. 2) The principal component analysis of the four indicators showed that the accuracy of the overall expression with the first three principal components fully met the needs, where the cumulative contribution rate reached 99.16%. The Least Square Method (LSM) was used to fit the soil nutrient index data. The correlation coefficient between the fitting and the original data reached 0.86, indicating the better representative for the change level of the original data. 3) The average error between the evaluation and the measurement was 11.82%, indicating that the predicted performance of the model was in better agreement with the actual situation. Therefore, the inversion of the model can be expected to better reflect the soil fertility level in the study area. The effective screening of soil fertility stress was achieved to quantitatively evaluate the soil fertility level using remote sensing image data.