Enhancing pedotransfer functions (PTFs) using soil spectral reflectance data for estimating saturated hydraulic conductivity in southwestern China

- An Ann-based model was developed to estimate Ks based on soil properties and soil spectra.
- Soil spectra can enhance the accuracy of estimating Ks.
- Spectra may have potential in Ks estimating at regional scale.

Soil pedotransfer functions (PTFs) are alternatives to expensive field and laboratory methods for acquiring soil hydraulic properties. Traditional PTFs generally require soil physiochemical properties (such as particle size distribution, bulk density, soil organic matter, etc.) as inputs. Soil spectral information is, more recently, being considered as an additional input property for improving the performance of PTFs. To date, however, there are only a few studies focused on this issue. Using 171 in-situ saturated soil hydraulic conductivity (Ks) observations in southwestern China and an Artificial Neural Network model (ANN), we developed three different PTFs, with different inputs, to estimate the saturated hydraulic conductivity. The inputs include the basic soil properties (denoted as BSP, bulk density, % clay, % silt and % sand) and soil spectral data (SPRD, dimension reduced by using the principal component analysis (PCA) method). We evaluated the performance of the new PTFs using the determination coefficient (R2), root mean square error (RMSE) and Akaike's information criterion (AIC). The results showed that including the SPRD as an additional input improved the performance of the PTF for estimating Ks. Compared to the PTF that solely used BSP as inputs, the R2 of the PTF, which used both the BSP and SPRD as inputs, increased by 0.33 (from 0.09 to 0.42) and the RMSE and AIC decreased by 0.26 (from 1.38 to 1.12) and 18.16 (from 273.96 to 255.8), respectively. These results indicate the potential of SPRD to improve the performance of PTF in estimating hydraulic conductivity.