Assessing the simulated soil thermal regime from Noah-MPLSM v1.1 for near-surface permafrost modeling on the Qinghai-Tibet Plateau

Abstract. Land surface models (LSMs) are effective tools for near-surface permafrost modeling. Extensive and rigorous model inter-comparison is of great importance before application due to the uncertainties in current LSMs. This study designed an ensemble of 6912 experiments to evaluate the Noah land surface model with multi-parameterization (Noah-MP) for soil temperature (ST) simulation, and investigate the sensitivity of parameterization schemes at a typical permafrost site on the Qinghai-Tibet Plateau. The results showed that Noah-MP generally underestimates ST, especially that during the cold season. In addition, the simulation uncertainty is greater in the cold season (October-April) and for the deep soil layers. ST is most sensitive to surface layer drag coefficient (SFC) while largely influenced by runoff and groundwater (RUN). By contrast, the influence of canopy stomatal resistance (CRS) and soil moisture factor for stomatal resistance (BTR) on ST is negligible. With limited impacts on ST simulation, vegetation model (VEG), canopy gap for radiation transfer (RAD) and snow/soil temperature time scheme (STC) are more influential on shallow ST, while super-cooled liquid water (FRZ), frozen soil permeability (INF) and lower boundary of soil temperature (TBOT) have greater impacts on deep ST. Furthermore, an optimal configuration of Noah-MP for permafrost modeling were extracted based on the connectivity between schemes, and they are: table leaf area index with calculated vegetation fraction, Jarvis scheme for CRS, Noah scheme for BTR, BATS model for RUN, Chen97 for SFC, zero canopy gap for RAD, variant freezing-point depression for FRZ, hydraulic parameters defined by soil moisture for INF, ST at 8 m for TBOT, and semi-implicit method for STC. The analysis of the model structural uncertainties and characteristics of each scheme would be constructive to a better understanding of the land surface processes on the QTP and further model improvements towards near-surface permafrost modeling using the LSMs.


). Details about the processes and 148 optional parameterizations can be found in Yang et al. (2011a).

149
In this study, the dynamic vegetation option in VEG process was turned off for   (1) Gap=F(3D structure, solar zenith angle)   The sensitivities of physical processes were determined by quantifying the 191 statistical distinction level of performance between parameterization schemes. The

192
Independent-sample T-test (2-tailed) was adopted to identify whether the distinction 193 level between two schemes is significant, and that between three or more schemes was 194 tested using the Tukey's test. Tukey's test has been widely used for its simple 195 computation and statistical features (Benjamini, 2010

202
To extract the optimal combinations of parameterization schemes, the connection 203 frequency (CF) between parameterizations was calculated:

205
(2) Donating the colRMSEs concentrated below the 5th percentile as the ''best  Obviously, for two given parameterization schemes, a large CF has an advantage 211 in terms of optimal combination.   conducted to test whether the difference between parameterizations within a physical 276 process is significant (Fig. 5). In a given sub-process, any two schemes labelled with VEG(1), VEG(3) and VEG(4) were labeled with the letter "A", "C" and "B", 285 respectively. As described above, the VEG process was sensitive for ST simulation.   The CF was calculated to extract the optimal combination of parameterization 309 schemes for ST simulation (Fig. 6). The CF between any two schemes from the same 310 physical process was zero as expected. Consistent with Fig. 5 Table 1).

333
The simulated results of the optimal scheme combination well captured the 334 variation of ST (Fig. 2). Despite the overestimation of ST at the shallow soil layers from 335 April to July, the optimal combination well produced the ST during the cold season and 336 of the deep layers (Fig. 2).   water, λ of frozen soil is generally expected to be greater than that of unfrozen soil.

365
As list in Table 1, VEG process includes three options to calculate the variation of  . As a result, the BTR process was insensitive at all layers.

391
CRS(1) and CRS(2) had no significant difference at most layers except the last two 392 layers. However, the performance difference between CRS(1) and CRS(2) at the last 393 two layers is very small (Fig. 3 and 5).