基于高光谱特征与人工神经网络模型对土壤含水量估算

发布时间：2018-12-05 20:21

【摘要】：土壤含水量(θ)是影响作物生长和作物产量的主要因素之一。旨在评估基于光谱特征参数的各种回归模型估算土壤含水量的精度,并比较人工神经网络(BP-ANN)和光谱特征参数模型的性能。2014年在室内获取砂土和壤土的土壤含水量和光谱反射率数据。结果表明:(1)当砂土容重为1.40 g·cm~(-3)时,900~970 nm最大反射率和900~970 nm反射率总和估算口达到极显著水平(R~2超过0.90);容重为1.50 g·cm~(-3)时,用蓝边最大反射率和900~970 nm反射率总和估算θ相关性最好(超过0.70);容重为1.60 g·cm~(-3)时,780~970 nm反射率总和与560~760 nm归一化吸收深度的R~2均超过0.90,达到极显著水平;容重为1.70 g·cm~(-3)时,900~970 nm最大反射率和900~970 nm反射率总和的R~2为0.88,呈极显著水平。(2)当土壤类型为壤土时,用900~970 nm最大反射率和900~970 nm反射率总和估算θ相关性最好。(3)蓝边反射率总和(R~2=0.26和RMSE=0.09 m~3·m~(-3))和780~970 nm吸收深度(R~2=0.32和RMSE=0.10m~3·m~(-3))估算砂土的含水量相关性最好。在估算壤土的含水量时,900~970 nm最大反射率(R~2=0.92和RMSE=0.05 m~3·m~(-3))与900~970 nm反射率总和估算模型的精度最高(R~2=0.92和RMSE=0.04 m~3·m~(-3))。(4)用人工神经网络模型能够更好地估算两种土壤的含水量(R~2=0.87和RMSE=0.05m~3·m~(-3))。因此,人工神经网络模型对θ估算具有巨大的潜力。
[Abstract]:Soil moisture content (胃) is one of the main factors affecting crop growth and crop yield. The purpose of this study was to evaluate the accuracy of various regression models based on spectral characteristic parameters for estimating soil water content. The performances of artificial neural network (BP-ANN) and spectral characteristic parameter model were compared. The soil moisture content and spectral reflectance data of sandy soil and loam soil were obtained indoors in 2014. The results are as follows: (1) when the bulk density of sand is 1.40g cm~ (-3), the summation estimation of the maximum reflectance of 900g 970 nm and the summation of reflectance of 900,970 nm reaches a very significant level (r _ (2) > 0.90); When the bulk density is 1.50g cm~ (-3), the correlation between the maximum reflectivity of blue edge and the sum of reflectance of 900g / 970 nm is the best (over 0.70). When the bulk density is 1.60g cm~ (-3), the total reflectance of 780g / 970 nm and the normalized absorption depth of 560g / 760 nm are over 0.90, which is extremely significant. When the bulk density is 1.70g cm~ (-3), the maximum reflectance of 900,970 nm and the sum of reflectance of 900,970 nm are 0.88, which is extremely significant. (2) when the soil type is loam, Theta correlation between the maximum reflectivity of 900g 970 nm and the summation of reflectance of 900g 970 nm is best. (3) the summation of blue edge reflectance (RP2O0.26 and RMSE=0.09 m3m-3) and the absorption depth of 780m970 nm (RG2O0.32 and RMSE=0.10m~3 m-) 3) the correlation of estimating the moisture content of sand is the best. In estimating the water content of loam, The summation estimation models of the maximum reflectance of 900N970 nm (RG2N0. 92 and RMSE=0.05 mt3m-3) and 900N970 nm have the highest accuracy (RM2M0. 92 and RMSE=0.04 mN3m-3). (4). Good enough to estimate the water content of the two soils (RC2N 0.87 and RMSE=0.05m~3 m-3).) Therefore, the artificial neural network model has great potential for 胃 estimation.
【作者单位】：中国农业大学资环学院土壤与水科学系;
【基金】：国家重点研发计划项目(2016YFD0800102) 国家自然科学基金项目(41371231)资助
【分类号】：S152.7

【相似文献】