摘要：
Land cover information about spatial distributions and dynamic characteristics of cover types is important for geographic condition monitoring, ecological and environmental modeling, resources management, and global change research. With proliferation of land cover information and its widening applications, research on its accuracy assessment is becoming increasingly urgent. Methods for accuracy assessment in land cover information may be divided into two kinds, information users' and producers' methods, depending on the sources and variety of data and metadata that may be used for accuracy evaluation. For validation of land cover change information from users' perspectives, especially for local accuracy quantification and analysis, it is necessary to investigate logistic regression. It is possible to use localized landscape pattern indices, such as class, homogeneity and patch size, which can be extracted from the maps being assessed, as covariates or explanatory variables for logistic regression modeling. Based on certain validation samples, regression coefficients can be estimated, with the resultant regression models applied to perform predictions of local accuracies. This paper seeks to investigate comparative performances of two alternative strategies for selective use of covariates: direct strategy (directly using pattern indices extracted from the change map being assessed as covariates), and composite strategy (using composites of patterns indices extracted from the individual single-date classification maps concerned in change analysis). Empirical studies in part of Wuhan show that there are no significant differences between the predictions obtained by the two strategies. Therefore, it is concluded that the direct strategy is a more straightforward and convenient choice for local accuracy predictions in land cover change information.%土地覆盖信息是地理国情监测、生态环境建模等领域的重要组成.逻辑回归方法是土地覆盖变化信息局域精度量化颇有潜力的方法之一.在逻辑回归建模中,可从待检验地图提取若干像素尺度的景观指数(如像素类别、同质性、异质性、斑块大小等),用作协变量;基于验证样本,可估计逻辑回归系数,从而进行未采样点位的的精度预测.关于协变量的选择有两种策略:一是直接使用土地覆盖变化图中的若干景观指数作为协变量(直接策略),二是使用两时相的土地覆盖图中相应的景观指数的组合作为协变量(组合策略).基于武汉部分城区的验证结果显示,两种策略下的逻辑回归预测结果的差异不明显.因此,从提高计算效率的角度,推荐采用直接策略进行土地覆盖变化信息的局域精度预测.
