土地退化是当前全球最严重的环境问题之一。对土地退化准确评价不仅为防治工作提供依据，也是建立监测预警系统的基础。土地退化发生范围广，采用人工调查方法局限性较大。遥感技术逐渐成为土地退化评价的重要技术手段和数据来源。随着土地退化监测手段的多元化，如何把这些多源数据尽可能地有效利用，将地面监测数据的精确性和遥感数据的易获取性结合起来提高土地退化评价精度，已经成为研究的一个热点。数据融合将遥感和地面监测数据提供的信息加以综合，消除多源数据间的冗余并综合互补信息，从而降低不确定性，最终获得对土地退化评价的一致性描述。本文以甘肃省境内的石羊河流域土地退化评价为研究对象，引入BP神经网络、模糊集理论和DS证据理论，建立数据融合模型，对土地退化进行评价，针对多源数据融合在土地退化评价中的应用，进行了如下几个方面的研究和探讨：(1)根据一定的原则，从地形、植被、土壤、气象水文、社会经济因子几个方面，选择适合于研究区域的土地退化指标，并根据应用要求进行合适的数据预处理过程。(2)提出利用BP神经网络进行土地退化评价。将地面监测数据作为训练样本训练神经网络，预处理后的指标数据作为输入数据进行土地退化评价。(3)提出利用DS证据理论进行土地退化评价。分别利用BP神经网络和模糊集理论构造基本概率分配函数，然后利用DS组合规则进行融合，并最终依据决策准则对土地退化进行评价。(4)将数据融合方法与常规的线性模型进行了对比分析，实验结果表明，数据融合方法提高了评价精度，且评价精度与地面监测点的代表性有关。(5)将实验方法应用到石羊河流域，实现了数据方法在土地退化评价上的应用。实验结果表明，石羊河流域土地退化较严重，呈两极分化和区域性分布，严重退化和未退化土地总共占流域总面积的62.69%，且分布情况与地貌单元和气候区基本吻合。

Land degradation is currently one of the most serious environmental problems. Accurate assessment of land degradation can not only provide basis for land degradation control, and also the foundation to built a monitoring and warning system. Due to a wide range of land degradation, artificial ground monitoring census method has a great limitations. Remote sensing technology gradually become the important evaluation technology and data sources because of its access speed, cover range, information and other characteristics. As the diversity of land degradation monitoring method, how to use these data effectively to improve elevation precision become a hot issue. Data funsion synthesize the information provided by remote sensing data and ground monitoring data, which will eliminate redundancy and comprehensive complementary information, reducing uncertainty and finally obtained the consistency description of the land degradation evaluation. This paper treated land degradation evaluation as the research topic, choose multi-indicators as funsion data, use BP neural network ,fuzzy set theory and DS evidence theory to establish a data fusion model for the land degradation evaluation. In this paper, the application of data fusion in the evaluation of land degradation has been studied. The main results and include of this work are as follows: (1) According to a certain principles, choose suitable indicators from terrain, vegetation, soil, meteorology, hydrology and social economic factors, and choose appropriate data preprocessing process according to the requirements. (2) This paper puts forward to use BP neural network in land degradation evaluation, take ground monitoring data as the training samples, the index data as input data for land degradation evaluation. (3)Thi paper puts forward to use the DS evidence theory in land degradation evaluation. Use BP neural network and fuzzy set thory to construct the probability distribution function. Take the ground monitoring data as the BP neural network sample, and make full use of the accuracy of ground monitoring data and the learning function of BP neural network, to built land degradation indicator and evaluation relationship model; use the statistical characteristics of the ground samples to structure fuzzy membership functions.And then use DS combination rules to combinate the basic probability distribution function, then according to the decision criteria to evaluate the land degradation. (4) Compare data fusion method with general linear model, the experiment results shows that the data fusion method improve the evaluation precision, and the accuracy is concerned with the representative of ground monitoring station. (5)Apply the experimental methods to the Shiyang River Basin, realization the application of data fusion in land degradation evaluation. The experimental results show that the Shiyang River Basin is in a serious land degradation condition, and has a polarization regional distribution, the serious and none degradation area is 62.69% of the whole basin. The distribution is consistent with geographic unit and climate zone.In addition to the natural factors ,the human activities is one of the factors that affected the land degradation.