基于HSI模型的土壤氮素空间预测不确定性的可视化表达

VISUALIZATION OF UNCERTAINTY ASSOCIATED WITH SPATIAL PREDICTION OF TOTAL N IN TOPSOIL USING HSI MODEL

任何模型都无法完全准确地再现真实世界,因此各种空间预测始终面临着不确定性的困扰。从20世纪90年代开始,土壤空间预测研究中输出结果的不确定性分析与评价日益受到重视。由于不确定性信息的可视化表达有助于更加直观地揭示与展示真实世界,因此是一种增强不确定分析与决策支持之间联系的有效途径。本文首先对不确定性可视化表达的主要技术与方法进行了简要论述,进而以北京市城市边缘带样区基于地统计学的土壤表层全氮含量的空间预测作为案例研究,以相对误差表示预测结果的不确定性,应用色调-饱和度-亮度(HSI)色彩模型实现土壤属性预测结果和不确定信息的可视化同步表达。在HSI模型中,色调值被用来表示预测值,白度表示不确定性。在案例研究的全氮预测输出结果中,颜色发白的区域表示预测结果的不确定性较高,需要做进一步补充采样。实际应用过程中发现,二维图例对不确定性的可视化表达效果有一定的限制作用,仍需在二维图例完善或创建新型图例方面开展深入的研究工作。

Spatial prediction is always facing challenges of uncertainty because it is almost impossible for any model to represent exactly the real world. Since the 1990s, more and more attention has been given to analysis and evaluation of uncertainties in output of spatial soil prediction. Visualization can be a way to increase communication between uncertainty analysis and decision-making, since it can help expose and express reality more intuitively. First of all, different techniques or ways to visualize uncertainty were briefly reviewed. And a case study was conducted of the statistics-based spatial prediction of total N content in the topsoil of peri-urban Beijing. Relative errors were deemed as expression of uncertainty in the prediction. A Hue-Saturation- Intensity （HS1） color model, which is a psychologically appealing color model, was recommended to visualize simultaneously soil attribute prediction resuhs and uncertainties therein. The HIS model uses hue to visualize prediction values and whiteness to visualize uncertainty. A two-dimensional legend was designed to supplement＇the visualization. This case study indicated that visualization of both prediction and prediction uncertainty offers a possibility to enhance visual exploration of data uncertainty and compare different prediction methods or predictions of totally different variables. The whitish area of the visualized output can be sim- ply interpreted as unsatisfactory predictions, where additional samples may be needed for a better prediction. The limitation of using the two-dimensional legend is that it is not easy to match the colors between the HSI coded image and the legend because the pale colors are always difficult to distinguish.