|
چکیده
|
Soil thermal conductivity is one of the main factors influencing the soil temperature regime. Over the recent years, by developing remote sensing technology, topographic attributes such as elevation, slope and aspect have become accessible from digital elevation models (DEMs). Based on the interactions between soil properties and topographically heterogeneous landscapes, this research investigates the application of topographical attributes (geographical and satellite information), along with soil physical properties, in order to create pedotransfer functions (PTFs) and estimate soil thermal conductivity, not reported so far. The hydrology experiment database of the Southern Great Plains 1997 (SGP97) available online at (http://www.cei.psu.edu/nasa_lsh/), was used to obtain these functions through applying linear regression, artificial neural networks (ANNs) and support vector machine methods. Twelve PTFs (PTF1-12) were developed to predict the thermal conductivity using different combinations of soil physical properties and topographical attributes. Improvement was attained via a combination of soil physical properties and topographical attributes (PTF5-11) rather than soil physical properties alone (PTF2-4). Thermal conductivity was better estimated by use of east-west slope degree, north-south slope degree and saturated water content along with texture fractions and bulk density as predictors in PTF7. The root mean square error decreased from 0.266 and 0.217 in PTF1 to 0.217 and 0.157 in PTF7 in the training and testing steps, respectively. ANNs were the best performing method with the relative improvement ranging from 3.52 to 19.35% and from 14.00 to 28.00% in the training and testing steps, respectively. The results displayed the successful estimation of the thermal conductivity in the training and testing steps, applying a combination of soil physical and topographical attributes.
|