【摘要】 遗址土体含水率、质地和密实度是土遗址长久保存的重要影响因素。可见—近红外光谱（VNIR）常用于土壤表面含水率的快速原位无损表征，但针对土遗址等文化遗产的数据集和反演方法研究少，夯土质地与密实度等关键特征对可见—近红外光谱—含水率模型的影响尚不明确。为提高含水率检测精度，研究遗址土质地与密实度对光谱—含水率模型的耦合影响，采集五处遗址不同质地的土样，在实验室内配制不同含水率的系列重塑土和散土样本，通过地物光谱仪获取光谱数据，经多元散射校正（MSC）处理，提取光谱特征波段，采用偏最小二乘回归法（PLSR）、多元线性回归法（MLR）、支持向量机法（SVM），构建土壤含水率的反演模型。结果表明，SVM模型的精度最高，基于通用波段（1379—1382 nm、1440—1450 nm、1982—1995 nm）建立的模型在验证集上的决定系数（R_P~2）为0.96，均方根误差（RMSE）为2.50，显著优于PLSR及MLR模型。遗址土质地、密实度会对土壤水分预测模型产生影响，改变特征波段和预测精度，然而其影响易被含水率、有机物含量等其他因素掩盖，造成光谱预测质地等参数的效果不佳。研究优化了反演模型，提升了基于VNIR的土遗址表面含水率分析精度和工程适用性，对土遗址预防性保护具有重要实践价值。更多还原

【Abstract】 The moisture content,texture,and compactness of soil are critical factors that affect the long-term conservation of earthen archaeological sites. Visible and near-infrared(VNIR) spectroscopy is commonly used for the rapid,in-situ,non-destructive analysis of surface moisture,although less research has been conducted on the datasets and inversion methods tailored to earthen cultural heritage sites. Moreover,the effects of key characteristics including soil texture and compactness on VNIR spectral-moisture content modeling remainsunclear. To improve detection accuracy and investigate the coupled effects of soil texture and compactness,soil samples of varying textures were collected from five heritage sites and used to make remolded and loose soil samples with different levels of moisture. Spectral data were then acquired using a field spectroradiometer,and spectral characteristic bands were extracted by processing the data with Multiplicative Scatter Correction(MSC). On the basis of these results,moisture content inversion models were constructed using Partial Least Squares Regression(PLSR),Multiple Linear Regression(MLR),and Support Vector Machine(SVM). The results demonstrate that SVM modeling achieved the highest accuracy,with models based on universal feature bands(1379—1382 nm,1440—1450 nm,1982—1995 nm) yielding a coefficient of determination(R_P~2) of0.96 and a Root Mean Square Error(RMSE) of 2.50 on the validation set,which significantly outperforms PLSR and MLR. The soil texture and compaction of earthen sites is known to affect soil moisture prediction and alter the characteristic wavelengths and forecast accuracy,though their effects can easily be hidden by other factors like moisture content and organic matter,leading to poor spectral prediction performance for parameters such as texture. This study optimizes inversion modeling and enhances the accuracy and engineering applicability of VNIR reflectance spectroscopy for analyzing surface moisture,and possesses significant practical value for the preventive conservation of earthen sites.更多还原