【作者】 郭熙；

【导师】 赵小敏；

【作者基本信息】 江西农业大学 ， 作物栽培与耕作学， 2010， 博士

【摘要】 农田养分的精确管理就是应用先进的信息技术手段,提高对农田养分资源的监测、评估和管理水平,同时提高农业生产技术的智能化水平。施肥模型是施肥技术的重要基石,是农田养分精准管理重要的一方面。本研究在不同条件下不同种类的水稻当季肥料利用率的研究基础上,构建了改进型养分平衡法施肥模型、神经网络施肥模型,并以GIS为工具建立县域水稻分区的肥料效应函数模型,在此基础上研发基于GIS平台和施肥推荐模型有机集成的农田养分信息管理系统。1.以江西省17个县217个试验点265次“3414”试验数据,分析得到不同土壤类型、不同地形地貌条件下江西省早、中、晚稻养分平衡施肥模型中的相关参数。研究结果表明：江西省水稻氮肥肥料利用率平均值为32.04%,P2O5肥料利用率为27.96%,K20的肥料利用率为50.69%。不同土壤类型对肥料利用率及其产量有着不同程度的影响。不同土壤类型中NPK当季肥料利用率的高低基本上表现为潴育型潮沙泥田>潴育型黄泥田＞潴育型麻沙泥田>潴育型鳝泥田＞潴育型黄沙泥田＞潴育型红沙泥田>淹育型红沙泥田>淹育型黄泥田>潜育型潮沙泥田>潜育型鳝泥田。不同地貌类型对肥料利用率和水稻产量也有一定的影响,氮、磷、钾当季利用率的相对高低顺序为平原>丘陵>山地。肥料当季养分利用率的高低与土壤肥力水平成反比,即高肥力田块由于供肥能力强,肥料利用率低,而低肥力田块由于供肥能力弱,肥料利用率反而高。不同水稻品种对肥料利用率也要一定影响,本研究表明中稻的肥料利用率大于晚稻、早稻,但差异不显著。2.针对传统的目标产量法施肥模型的缺陷,尤其是在县域推广进行测土配方施肥时,无法一一做试验确定其参数的不足,利用已进行的试验计算相关参数并添加相关调整系数对传统的养分平衡法水稻施肥模型进行改进。构建的改进型养分平衡法施肥模型为FQ=Np×(G-B/Al)/(Fn×FUR×Fl×F2×F3)。研究结果表明：利用全省不同土壤类型、不同地貌类型、不同水稻类型肥料利用率修正系数的改进型养分平衡法施肥模型,可以在全省进行推广使用。在奉新的试验表明改进型实际产量最高,传统型次之,习惯施肥再次,空白处理最低；改进型养分平衡法实际产量平均值比习惯施肥处理高135kg/hm2,增幅达1.9%,比传统养分平衡法处理高77.81 kg/hm2,增幅为1.1%。增产节支效益方面,改进型增产节支效益最高,传统养分平衡法次之,习惯施肥最小。改进型比传统型高出367.91元／公顷,增幅达20.63%。改进型比习惯施肥处理高出499.15元／公顷,增幅达30.21%。同时,可以看出,改进型推荐施肥的处理更接近目标产量。3.以BP算法为工具,利用试验结果,构建了江西省基于人工神经网络的水稻施肥模型。研究结果表明：构建的"4-4-10-3”的水稻施肥神经网络结构,有较好的训练精度和测试精度。施NPK肥模型结果与实际值的相关性为0.997,、0.982,,0.972。误差率均控制在1%以内,具有较好的泛化性功能。在奉新县选择了六个试验点进行了不同处理的产量试验表明神经网络模型实际产量最高为6846.75kg/hm2,改进型次之为6813.75kg/hm2,习惯施肥再次之为6600.5kg/hm2,空白处理最低为3967.95kg/hm2。同时对处理间差异显著性进行分析,结果表明改进型养分平衡法施肥以及神经网络模型施肥的产量差异不显著,习惯施肥与改进型养分平衡法、神经网络模型法处理的产量差异显著。4.结合传统试验数据,综合GIS和建模技术,从大尺度县域入手解决小尺度区域施肥问题；以弋阳县为例探讨分析丘陵县域土壤养分区域化分布的状况,实现丘陵耕地地力综合分区,得到各区水稻最佳经济效益施肥方案。研究结果表明：弋阳县土壤养分综合分区,共九个区,V区面积最大,达8005.22公顷,占全县耕地面积的34.02%,这个区域也是弋阳县最重要的粮食生产区域；面积大小依次是Ⅳ区、Ⅲ区、Ⅶ区、Ⅱ区、Ⅷ区、Ⅵ区、Ⅸ区,面积最小的是Ⅰ区,仅36.54公顷,仅占全县耕地面积的0.16%。所形成的具有代表性的各分区肥料效应函数方程分别代表各区肥力等别土壤施肥量与产量的关系,针对性强,解决了以往单个肥料效应方程难以代表不同地块及不同肥力水平的问题；相对以往的数据显著地增加了代表性。5.以C#为平台,利用ArcEngine的二次开发功能,在构建施肥模型的基础上,建立了施肥的实体-关系型数据库,形成施肥子模块,并形成农田养分管理信息系统。研究结果表明：系统采用可视化开发语言C#和地理信息系统二次开发平台ArcEngine,可以脱离GIS工具软件的运行环境、功能完全可以不逊于通用型GIS软件。农田养分管理信息系统的功能分为文件管理、信息查询、属性编辑、施肥推荐等相关模块,每个模块中又分若干个子功能,同时可以最大限度的满足农业部门土肥业务管理的需求,系统达到了既提高办事效率和减轻业务人员的劳动强度的要求。更多还原

【Abstract】 Accurate management of paddy nutrient is to improve the level of surveillance、evaluation and management, at the same time to improve the intelligence of agricultural techniques by advanced information technology, fertilization model is an important cornerstone for Fertilizer Technique, and is most one for Accurate management of paddy nutrient. Based on the research of fertilizer apparent recovery fraction at current season with different species of rice at different conditions, this research forms modified nutrient balance model and neural network model, establishes regionalized fertilization model of rice sativa by GIS. In conclusion, the organic integrated research and development of paddy nutrient information management is formed based on GIS and fertilizer recommendation model.1.Based on 265 dates from 217 test points of "3414" in 17 rational countries of Jiangxi Province, at last comes to a relative parameters conclusion in nutrient balance model of early-season rice、mid-season rice and late-season rice with different soil types and topography. Research results show:At Jiangxi Province the average for Nitrogen fertilizer utilization ratio is 32.04%, for P2O5 fertilizer utilization ratio is 27.96%, and for K20 fertilizer utilization ratio is 50.69%. Different soil types have different affection degree of fertilizer utilization and production. In different soil types NPK apparent fertilizer utilization at current season is different, hydromorphic tidal field sand clayey soil is the highest, then are hydromorphic yellow clayey soil、hydromorphic Ma sand clayey soil,hydromorphic eel clayey soil,hydromorphic yellow sand clay soil.hydromorphic red sand clay soil,submergenic red sand clay soil.submergenic yellow clayey soil,gleyic tidal field sand clayey soil and gleyic eel clayey soil. Different topographies affect fertilizer utilization and production, NPK apparent fertilizer utilization at current season is highest in flats, then is hilly land and mountainous area. Fertilizer apparent recovery fraction at current season is in inverse proportion of soil fertility, because the nutrient-supply capacity of high productivity fields are stronger, fertilizer utilization is lower; low productivity fields go by contraries. Different rice varieties affect fertilizer utilization, this research shows that fertilizer utilization of mid-season rice is higher than early-season rice and late-season rice, but diversity is not so remarkable.2.In allusion to fertilization model defects of the traditional target yield method; Especially, when conducted in the county to promote soil testing and fertilizer, we are unable to do testing in order to determine the lacks of parameters; Using tests which have been carried out, calculating the parameters and adding a related adjustment factor, this research improves the traditional model of nutrient balance method for Rice Fertilization, and the improved nutrient balance model is FQ=Np×(G-B/Al)/ (Fn×FUR×Fl×F2×F3).The research results show:The improved nutrient balance model with fixed coefficients of fertilizer utilization ratio for different soil types, different topographies, and different rice varieties, can be promoted and used in the province. The tests at Fengxin show:The highest actual production is the improved model, the second is the traditional model, the third is the farmers’fertilizer practice, the lowest is without any action; With disposing actual production averages the improved nutrient balance model is 135kg/hm2 (Growth rate reached 1.9%) higher than the farmers’fertilizer practice, and is77.81 kg/hm2 (Growth rate reached 1.1%) higher than the traditional nutrient balance model. With the yield efficiency saving the improved model is highest, the traditional nutrient balance model is second, the farmers’fertilizer practice is least; the improved model is 367.91 yuan/ha. (Growth rate reached 20.63%) higher than the traditional model, and is 499.15yuan/ha. (Growth rate reached 30.21%) higher than the farmers’fertilizer practice. At the same time, disposing fertilize of the improved nutrient balance model is Closer to the target output.3.With BP algorithm as a tool, using the test results, we construct the model of rice fertilization based on Jiangxi Province artificial neural network. And the results show that: Construction of the "4-4-10-3"neural network structure of rice fertilization has better training accuracy and testing accuracy. The correlation coefficient between the model results of NPK fertilizer applied and the actual value are 0.984,0.966,0.987; error rate are less than 1%with in the control; these have good generalization of sexual function. In Fengxin county, we select six pilot sites to fertilize with different ways, the research results show:The neural network model’s actual production is 6846.75 kg/hm2 with the highest, the improved model’s is 6813.75 kg/hm2 with the second, the farmers’fertilizer practice’s is 6600.5 kg/hm2 with the third, and without any action’s is 3967.95 kg/hm2 with the lowest. At the same time, we analyze variations between the different ways with significance analysis, the results show:the variations for production between the improved nutrient balance model and the neural network model are not obvious, but the variations for production between the farmers’fertilizer practice, the improved model and the neural network model are obvious. 4.Union tradition empirical datum, Synthesizes GIS and the modeling technology, starts from the county with large-scale problem of fertilizer to solve small-scale regional. This research analysed soil nutrient regionalizing distributed condition for hill county territory at Yiyang County, obtained the hill farming soil fertility comprehensive district, then got best economic value paddy fertilization program in all areas of the district. The findings indicated:The soil Nutrient at Yiyang County can be colligating zoned nine districts, the largest district is V which is 8005.22 ha., counting for 34.02%of the total farmland, and is the most district of grain production area at Yiyang County; The districts from large to small are IV III VII. II VIII VI IX; The smallest district is I which is 36.54 ha., only counting for 0.16%of the total farmland. The representative district fertilizer effect function equations represent the districts such as soil fertility, and relationship between yield fertilizer and output; These equations have strong pointed, solve the problem that the equation of a single fertilizer effect is difficult to represent different plots and different levels of fertility; and obviously increase data representation than before.5.Take C# as the platform, using the secondary development function of ArcEngine. basing on building a fertilization module, we establish entity-relational database for fertilization, form the fertilizer submodule and paddy nutrient information management system. The research results show:The system uses the non-visual development languages, C# and the secondary GIS development function of ArcEngine, then it’s operating environment can be divorced from GIS, it’s functions like the general-purpose GIS software. The functions of paddy nutrient information management system divide into File management, information query, attribute editing, fertilizer recommend, and others relevant modules, every modules can divide into several functions. The system can make the best of fertile business management needs for the agricultural sector, and achieve business personnel requests at both improving efficiency and reducing work intensity.更多还原