Remote　sensing　soil　moisture　estimation　is　a　complex　process　due　to　the　influence　of　the　uneven　subsurface,　resulting　in　many　uncertainties.　This　paper　focuses　on　the　Qinghai-Tibet　Plateau　as　the　study　area　and　discusses　the　uncertainties　of　soil　moisture　estimation　from　three　aspects:　feature　variables,　target　variables,　and　estimation　models.　The　study　concludes　that　the　Mean　Decrease　in　Impurity　feature　selection　method　with　K=12　significantly　outperforms　other　feature　selection　methods　in　selecting　feature　variables.　Using　measured　data　as　the　model　target　variable　results　in　the　highest　precision　and　smallest　standard　deviation　(R=0.9257,　RMSE=0.0388cm3/cm3).　The　enhanced　generalized　regression　neural　network　(EGRNN)　model　has　higher　accuracy,　smaller　standard　deviation　values　(R=0.9422,　RMSE=0.0341　cm3/cm3),　and　stronger　adaptability　in　soil　moisture　estimation.　　©　2024　The　Authors.