Particulate organic matter is a part of the labile, easily decomposable, pool of soil organic matter; therefore its study in soil is very important. Estimating particulate organic matter through some related variables at an acceptable level of accuracy is very importance. In this study three model were utilized and compared for predicting particulate organic matter. (i) wavelet function of artificial neural network, (ii) perceptron function of artificial neural network and III) multiple linear regression. The highest R² value and lowest NRMSE in the regression models were 0.94 and 0.071, respectively, in the wavelet function of artificial neural network were 0.97 and 2.10 E-05, respectively. These values in perceptron function of artificial neural network were intermediate. The results indicated that although wavelet function of artificial neural network was the most reliable model for predicting POM, but multiple linear regression was also an appropriate predictor due to its acceptable R², especially when the logarithmic forms of the data were used in the models. The results revealed that the best inputs for modelings are soil physicochemical properties only. Wavelet function of the ANN was utilized in the POM prediction for the first time.

颗粒有机物是不稳定的，易分解的土壤有机物库的一部分。因此在土壤中的研究非常重要。通过一些相关变量以可接受的准确度估算颗粒有机物非常重要。在这项研究中，使用了三种模型并进行了比较，以预测颗粒有机物。（i）人工神经网络的小波函数，（ii）人工神经网络的感知器功能，以及III）多元线性回归。最高R ²回归模型中的NRMSE值分别为0.94和0.071，在人工神经网络的小波函数中分别为0.97和2.10 E-05。人工神经网络的感知器功能中的这些值是中间值。结果表明，虽然人工神经网络的小波函数是预测POM的最可靠模型，但是多元线性回归由于其可接受的R ²也是合适的预测因子，尤其是在模型中使用对数形式的数据时。结果表明，建模的最佳输入仅是土壤物理化学性质。人工神经网络的小波函数首次用于POM预测中。