【作者】 王磊；

【导师】 冯克鹏；

【作者基本信息】 宁夏大学 ， 土木水利（专业学位）， 2022， 硕士

【摘要】 水资源短缺已成为制约宁夏农业发展的主要因素,提高农业用水效率是当地农业可持续发展必须解决的难题。土壤含水量是农田用水管理的基础信息。本文针对宁夏引黄灌区玉米种植区利用遥感监测土壤水分中存在的问题,开展基于无人机多光谱和热红外遥感的玉米土壤水分反演研究,为节水灌溉决策提供理论依据和技术支撑。在宁夏青铜峡灌区选取一块代表性玉米种植田,对玉米全生育期进行持续遥感监测,埋设土壤三参数传感器长期原位监测土壤水分,获取无人机遥感数据和地面实测数据,分析实验田块土壤水分变化规律、气候特征、土壤温度变化等,确定反演数据集中特征变量;研究不同冠层遮蔽条件下及全生育期土壤水分反演特点;比较相同数据输入模式下10cm、20cm、50cm不同深度土壤水反演效果;研究基于机器学习方法不同特征变量组合对土壤水分反演效果的影响,并选取代表性模式进行参数调优,综合确定最佳反演模型。主要研究成果有:1、实验田块各层土壤水分在两次灌溉之间呈线性耗散,耗散速度10cm层>20cm层>50cm层。根据土壤水分变化趋势可将土壤分为上下两部分,40cm层以上部分土壤水分受降水影响波动大,整体耗散较快,属于作物生长主要耗水区域,40cm层及以下部分土壤水分受降水影响波动小,整体耗散较慢,下层土壤含水量整体要高于上层,并且50cm层土壤水分最大。不同层土壤水分间存在极显著正相关关系。在灌溉影响下,土壤水分变异系数稳定在0.04～0.14的范围内。水平和垂直方向上,土壤水分变异系数与土壤水分平均值均有极显著负相关关系,并且可以用一元二次函数或一元三次函数来表示。灌溉后,变异系数会随土壤水分减少经过极小值和极大值点,对10cm和20cm层而言,变异系数极小值点和极大值点对应的土壤水分分别为0.31 m3/m3和 0.19 m3/m3。2、对土壤水分遥感反演,冠层未遮蔽前多光谱遥感较优,冠层完全遮蔽后热红外遥感较优,全生育期多光谱和热红外遥感均有较好表现,并且两者结合后反演效果有所提升;全生育期反演数据集中样本量最多,反演效果最优;随机采样方法在样本量较小时容易出现测试集R2有较大下降或测试集R2全部大于训练集的情况,可以尝试采取分层采样方法进行数据集划分。3、对10cm、20cm、50cm层土壤水分分别进行反演,比较不同层土壤水分反演特点和效果,设置不同的数据输入模式,比较不同特征变量组合的反演效果,评估了环境条件对反演效果的提升作用,确定了最佳模式,并讨论了模型参数调优对反演效果的影响,结果表明相对其他两层,10cm层土壤水分反演效果最优,反演的潜力最大。对全生育期反演10cm层土壤水分模型进行参数调优后,仅遥感信息的模式训练集R2达0.63,测试集R2达0.57;全部遥感信息组合土壤温度、近地面温度、热红外温度-气温差、近地面温度-气温差,训练集R2达0.82,测试集R2达0.80,具有最佳的反演效果。4、本文除波段反射率、植被指数、热红外温度等,引入了环境条件协同反演土壤水分,重点对比全部特征变量组合和仅使用遥感信息两种模式下的反演效果,结果表明环境条件能有效提升反演效果,仅使用遥感信息反演土壤水分的模型具有较大的提升潜力,并且可以尝试基于温度间的相关关系对环境条件进行替代,来简化和改进已有模型。更多还原

【Abstract】 Water scarcity has become the main factor restricting the agricultural development of Ningxia.Improving the efficiency of agricultural water use is a difficult problem that must be solved for the sustainable development of local agriculture.Soil moisture is the basic information for farmland water management.Aiming at the problems existing in the use of remote sensing to monitor soil moisture in the corn planting area of the Ningxia Yellow River Diversion Irrigation District,this thesis conducts a study on corn soil moisture inversion based on UAV multi-spectral and thermal infrared remote sensing,which provides theoretical basis and technical support for water-saving irrigation decision-making.This research selected a representative maize planting field in Qingtongxia Irrigation District of Ningxia,obtained UAV remote sensing data and ground measured data,analyzed the change law of soil moisture,climate characteristics,soil temperature changes,etc.in the experimental field.Besides,determined the characteristic variables in the inversion data set;studied soil moisture inversion characteristics under different canopy shading conditions and throughout the growth period;compared soil water inversion effects at different depths of 10cm,20cm,and 50cm using the same data input mode;studied the effects of different combinations of feature variables based on machine learning for soil moisture inversion,and selected representative models for parameters adjustment,comprehensively determined the best inversion model.All the work provided a reference for the implementation of remote sensing monitoring soil moisture in maize planting areas,precise irrigation and scientific decision-making in the Yellow River Diversion Irrigation District in Ningxia.The main research results are:1.The soil moisture of each layer of the experimental field is linearly dissipated between two irrigations,and the dissipation rate is 10cm layer>20cm layer>50cm layer.According to the changing trend of soil moisture,the soil can be divided into upper and lower parts.The soil moisture above the 40cm layer is affected by precipitation and fluctuates greatly,and the overall dissipation is relatively fast,which belongs to the main water consumption area for crop growth.The soil moisture in the 40cm layer and below is less affected by precipitation,and the overall dissipation is slow.The soil water content in the lower layer is generally higher than that in the upper layer,and the soil moisture in the 50cm layer is the largest.There is a very significant positive correlation between soil moisture in different layers.Due to the influence of irrigation,the coefficient of variation of soil moisture stabilized in the range of 0.040.14.In the horizontal and vertical directions,the coefficient of variation of soil moisture has a very significant negative correlation with the average value of soil moisture,furthermore,it can be represented by a quadratic function of one variable or a cubic function of one variable.After irrigation,the coefficient of variation will pass through the minimum and maximum points with the decrease of soil moisture.For the 10cm and 20cm layers,the soil moisture corresponding to the minimum and maximum points of the coefficient of variation are 0.31 m3/m3 and 0.19 m3/m3,respectively.2.For soil moisture remote sensing inversion,the multi-spectral remote sensing is better before the canopy is unshaded,while the thermal infrared remote sensing is much better after the canopy completely shaded.Both multispectral and thermal infrared remote sensing performed well in the whole growth period,and the retrieval effect was improved after the combination of the two methods.The inversion data set in the whole growth period has the largest number of samples,and its inversion effect is the best The random sampling method is prone to the situation that the test set R2 has a large drop or the test set R2 is all larger than the training set when the sample size is small.Therefore,the stratified sampling method can be used to divide the data set3.Inverting the soil moisture of 10cm,20cm and 50cm layers respectively,compared the characteristics and effects of soil moisture inversion in different layers,set up different data input modes,compared the inversion effects of different combinations of characteristic variables,and evaluated the impact of environmental conditions on the inversion.The optimal model was determined,and the influence of model parameter tuning on the inversion effect was discussed.The results showed that the soil moisture inversion of the 10cm layer has the best effect compared to the other two layers,and has the greatest inversion potential.After optimizing the parameters of the 10cm layer soil moisture model inversion in the whole growth period,only the remote sensing information model training set R2 reached 0.63,and the test set R2 reached 0.57;all remote sensing information combined with soil temperature,near-surface temperature,thermal infrared temperature-air temperature difference,the difference between ground temperature and air temperature mode’s training set R2 reaches 0.82,and its test set R2 reaches 0.80,which has the best inversion effect4.Apart from band reflectivity,vegetation index,thermal infrared temperature,this thesis introduces environmental conditions to synergistically retrieve soil moisture,and we focus on comparing the inversion effects of all the feature variable combinations and only using remote sensing information.The results showed that environmental conditions can effectively improve the inversion effect.The model that only uses remote sensing information to invert soil moisture has great potential for improvement,and you can try to replace environmental conditions based on the correlation between temperatures to simplify and improve existing models.更多还原