遥感模型评估中顾及样本数量的统计指标:基于水色遥感的比较研究

摘要

为了推荐在遥感模型评估中引入不确定性的评价指标,即测量学界近年来提出的不确定度A类评定(UA).本文通过评估一系列水色遥感反演模型在各类数据集的表现,用计算机模拟了误差的各种近似分布,研究了样本个数n和统计指标RMSE、MAE与UA之间的关系.算法测试和计算机模拟的结果都表明,在样本个数少的情况下,遥感模型的不确定性(可信赖度)评价用UA比RMSE、MAE更适合,因为样本个数越多,建立的模型就越可靠(不确定度越小).此外,本文至少使用50个样本验证遥感模型,可使精度评价更加准确.%In order to introduce the uncertainty evaluation index in the remote sensing model evaluation,which is the uncertainty A (UA) proposed by the surveying community in recent years.This letter focuses on the impact of sampling size on those statistical variables,by testing a number of ocean color algorithms with different datasets and four kinds of approximate distributions of errors simulated by computer.The results indicate that RMSE,MAE and UA all vary with the number of observations n but present opposite trends.When n is less than 50,RMSE typically presents a upward trend from a small n to a large n.MAE in most of the results are similar to the RMSE,while UA always keeps a smooth downward trend with increasing n,Based on the results,we conclude that UA is better than RMSE and MAE to be employed for evaluating and presenting the uncertainty in remote sensing validation,because the more samples we take,the less uncertainty we get.In addition,using at least 50 samples to validate the remote sensing model,will make the evaluation more accurate.