基于多环境因子的新疆棉叶螨时空动态监测模型研究

Xinjiang region is important cotton-producing area in China, but in recent years, with promotion of cotton drip irrigation under plastic film technology and the continued warming of global climate, the ecological environment of cotton field becomes more conducive to cotton leaf mite, and increased risk of cotton spider mites disaster. Due to the lack of scientific and effective prediction method, cotton spider mites prevention and management is very difficult, so it has become a big obstacle of cotton production. The project study in precocious cotton-growing areas and special early maturing cotton-growing areas of Xinjiang and early-maturing cotton region as the area, with main factors associated with the cotton spider mites development and its rules as the research object, with the dynamic prediction of cotton leaf mite occurring time, regional and harm degree as the study target, refine cotton leaf mite problem domain features, then using multivariate nonlinear regression analysis method for model building and algorithm validation, for the more using model and spatial data for overlay analysis and buffer analysis, at last dynamic models of development of the cotton leaf mites occurred time and space were constructed，mite pests forecasting precision positioning and visualization were realized. The study results can provide scientific guidance for cotton leaf mite prevention from the traditional fields scattered control towards the regional comprehensive control and management, so good theoretical significance and practical value were put out.

新疆地区是中国重要的产棉区，但近年来随着棉花膜下滴灌的普及以及全球气候不断变暖，棉田生态环境变得更有利于棉叶螨的发生，棉叶螨成灾的风险加大。由于缺乏科学有效的预测预报手段，很难做到灾前预防，而灾后治理难度大，成为棉花生产的一大障碍。本项目以新疆地区有代表性的早熟棉区和特早熟棉区为研究区域，以棉叶螨成灾因子和时空变化规律为研究对象，以对螨害发生时间、区域和为害程度的动态预测为研究目标，提炼棉叶螨发生、变化的问题域特征，构建螨害发生程度的预测模型，揭示在新疆特殊种植模式和气候条件下棉叶螨消长机理与因子特征，并将模型预测结果与空间数据进行叠置分析和缓冲区分析，从而建立棉叶螨发生及蔓延的时空动态模型，为实现螨害区域化分级预报提供理论依据和建模方法，为螨害防治从传统的条田分散治理走向区域性综合防治提供科学指导。

棉花生产是新疆地区最重要的农业支柱产业，种植面积长期占一半左右。但由于气候变暖和长期连做，棉叶螨发生情况日益严重，尤其近年来呈不断上升趋势。大面积预防性用药成为防治的主要手段，在增加棉农经济负担的同时也带来环境污染，造成人为生态灾害。如何准确监测和提前预测，从而提前统防统治和精确防治，成为亟待解决的问题。因此该项目通过研究棉叶螨与生存环境之间的关系，针对新疆棉花种植特点，找出棉叶螨消长的主要影响因子和作用规律，结合土质、长势等空间数据建立预测叶螨发生程度的时空模型，提供叶螨为害程度和发生区域的预测预报。主要从以下几个方面展开了研究工作：首先，为精确获取棉田叶螨危害程度和发生面积，基于无人机多光谱数据构建了基于TVI、DVI和RDVI的logistic回归模型，分类准确率达到95%。通过对高光谱数据与棉叶螨严重程度之间相关性进行研究，发现759nm、634nm、661nm和693nm与螨害严重度均呈显著相关，相关系数分别为-0.797，0.447、0.473和0.579。结合其他光谱指数构建了多个监测模型。其次通过综合气象数据和棉花长势光谱指数构建了一定区域尺度下的棉叶螨预测logistic回归模型，实现了对农田级别的棉叶螨预测，经测试准确率达到82.9%。再次，基于灰色理论建立了年度棉叶螨预测GM(1,1)灾变模型，通过结合马尔科夫链和人工神经网络改进灾变模型实现了对一定区域的棉叶螨灾情的等级预测。总体上，该项目完成了从农田到一定区域范围的棉叶螨监测和预测，达到了预期目标。