Soil thermal conductivity (lambda), describing the ability of transferring heat in the soil, plays an important role in soil thermal behavior. The estimation of lambda at dryness (lambda(dry)) is essential for obtaining accurate lambda. This study aims to develop a new model for lambda(dry) across a wide range of the soil dry density (rho(d)) for soils with different textures. The lambda(dry) measurements of 75 soil samples from literature and 19 new soils from Qinghai-Tibet Plateau are used to establish the segmented relationships between lambda(dry) and rho(d) based on clustering algorithms. Our analyses reveal that when rho(d) < 1.4 g cm(-3), rho(d) significantly influences lambda(dry), while for rho(d & nbsp;)>= 1.4 g cm(-3), other soil properties must be taken into account. So, the performances of 12 widely used models are evaluated in these two different rho(d)& nbsp;ranges. Results show that when rho(d)& nbsp;< 1.4 g cm(-3), the previous developed models generally perform well. On the contrary, all these models produce large calculation errors when rho(d)& nbsp;>=& nbsp;1.4 g cm(-3). This further confirms the necessity of segmentation. Finally, with a demarcation point of 1.4 g cm(-3), a new model with different calculation methods is proposed herein for predicting dry. The new model exhibits the highest accuracy in predicting & nbsp;lambda(dry) with the highest correlation coefficient (R), lowest root mean square error (RMSE), and smaller mean bias error values; compared to the previous models, the new model RMSE values are reduced by 16.6% on average for soils with rho(d)& nbsp;< 1.4 g cm(-3)& nbsp;and 21.0% for rho(d)& nbsp;>=& nbsp;1.4 g cm(-3), respectively. Namely, the new model is highly suitable for studying lambda(dry)& nbsp;for different rho(d)& nbsp;due to its simplicity and applicability.

来源平台：INTERNATIONAL JOURNAL OF THERMAL SCIENCES