利用ICESat-2/ATLAS修正SRTM高程误差的PSO-RF方法

PSO-RF Approach for Correcting Elevation Error of SRTM Using ICESat-2/ATLAS

摘要: 针对当前SRTM高程误差修正方法未能充分挖掘高程误差及其影响因素间的复杂非线性关系的不足，本文提出一种修正SRTM的PSO-RF方法。首先，从ICESat-2/ATLAS强光束数据中提取参考高程光子，计算各光子对应的SRTM高程误差及地形参数和地表覆盖类型参数；然后，构造随机森林(Random Forest, RF)算法和粒子群优化(Particle Swarm Optimization, PSO)算法相结合的SRTM高程误差修正模型，并用光子数据训练修正模型；最后，将训练所得PSO-RF模型应用于ICESat-2未覆盖的SRTM区域，得到修正后的SRTM，并与多项式回归(Polynomial Regression, PR)方法进行比较分析。选取美国内华达山脉和圣华金河谷区域的SRTM数据进行实验验证，实验结果表明：PSO-RF方法所构造的修正模型可有效减小SRTM高程修正的误差，修正精度优于PR方法。

Abstract: In response to the inadequacy of current SRTM elevation error correction methods for fully ex-ploring the complex non-linear relationship between elevation errors and their influencing fac-tors, this paper proposes a PSO-RF method for correcting SRTM. Firstly, reference elevation photons are extracted from ICESat-2/ATLAS strong beam data, and the SRTM elevation errors, terrain parameters, and land cover type parameters corresponding to each photon are calculated. Then, a SRTM elevation error correction model is constructed by combining the random forest (RF) algorithm and particle swarm optimization (PSO) algorithm, and the correction model is trained using photon data. Finally, the trained PSO-RF model is applied to SRTM regions not covered by ICESat-2 to obtain the corrected SRTM and compared and analyzed with the polynomial regression (PR) method. Experimental verification is conducted using SRTM data from the Nevada Mountains and San Joaquin Valley regions in the United States, and the results demonstrate that the correction model constructed by the PSO-RF method can effectively reduce the error of SRTM elevation correction, and the correction accuracy is superior to the PR method.