Noah-MP is a new-generation of LSM, which was developed to improve the performance of the Noah LSM (Chen et al., 1996; Chen and Dudhia, 2001). It is coupled to the WRF community weather and regional climate model (Barlage et al., 2015), and also available as a stand-alone 1-D model (Noah-MP v1.1). Noah-MP simulates several biophysical and hydrological processes that control fluxes between the surface and the atmosphere. These processes include surface energy exchange, radiation interactions with the vegetation canopy and the soil, hydrological processes within the canopy and the soil, a multilayer snowpack with freeze–thaw, groundwater dynamics, stomatal conductance, and photosynthesis and ecosystem respiration. The major components include a one-layer canopy, three-layer snow, and four-layer soil. Noah-MP provides a multi-parameterization framework that allows using the model with different combinations of alternative process schemes for individual processes (Niu et al., 2011). Alternative submodules for 12 physical processes can provide more than 5000 different combinations. Soil water fluxes are calculated by the Richards equation using a Campbell/Clapp–Hornberger parameterization of the hydraulic functions (Clapp and Hornberger, 1978).

We use an offline stand-alone 1-D mode (Noah-MP) with four soil layers: 0–10, 10–40, 40–100, and 100–200 cm. The selected Noah-MP physics options used in this study are similar to Barlage et al. (2015), Gao et al. (2015) and Chen et al. (2014) and are list in Table 1. In the default configuration of Noah-MP, the entire vertical soil profile was treated as one mineral ground texture only, and no organic soil matter is included.

The OAS research site has an organic LFH (forest floor) soil horizon, 8–10 cm deep. This study evaluates the impact of adding an organic soil horizon in the Noah-MP model using a similar approach to Lawrence and Slater (2008), which parameterizes soil thermal and hydrologic properties in terms of carbon density in each soil layer. Soil carbon or organic fraction for each layer is determined as fsc,i=ρsc,iρsc,max⁡, where fsc,ifsc,i is the carbon fraction of the each layer, ρsc,i is the soil carbon density, and ρsc,max⁡ is the maximum possible value (peat density of 130 kg m-3, Farouki, 1981). In this study, we assume that the topsoil layer is made up of 100 % organic matter, consistent with the 8–10 cm LFH horizon at OAS, with the carbon fraction equal to 1. The soil properties for this layer are calculated based on the parameters of organic soil. The second layer of the soil is considered to be a transition layer and made up of 30 % organic matter with the carbon fraction equal to 0.3. The soil properties of this layer are specified as a weighted combination of organic and mineral soil properties: P=1-fsc,iPm+fsc,iPo, where Pm is the value for mineral soil, Po is the value for organic soil, and P is the weighted average. The remaining soil layers were assumed to be 100 % mineral so we conducted sensitivitraction equal to 0. The soil properties for this layer are calculated based on the parameters of mineral soil. To investigate impacts of uncertainties of those parameters on simulations, we conducted sensitivity tests for key parameters such as saturated hydraulic conductivity, porosity, suction, and the Clapp and Hornberger parameter. Those parameters were perturbed within a 5–20 % range (except for hydraulic conductivity that is changed over 4 times below and above the default value) following the work of Letts et al. (2000). Results showed that the simulated top layer soil moisture is very sensitive to porosity, saturate hydraulic conductivity, saturated matric potential and the Clapp and Hornberger parameter, while other layers are not too sensitive to those parameters. For porosity, as the value increased, the topsoil moisture increased significantly. The saturated hydraulic conductivity mainly influences the unfrozen period. As the value increased, the topsoil moisture decreased. Saturated matric potential and the Clapp and Hornberger parameter only influence the frozen period. For saturated matric potential, the topsoil moisture decreased when the parameter value increased, while for the Clapp and Hornberger parameter, the topsoil moisture increased when the parameter value increased. Based on the site measurement, the soil bulk density of the top layer is about 160 kg m-3. As described in Letts et al. (2000), this organic soil can be defined as hemic peat, a medium humified organic soil. Table 2 gives the recommended parameters for hemic peat, with 0.88, 2.0, 0.0102, and 6.1 for porosity, saturated hydraulic conductivity, saturated matric potential, and the Clapp and Hornberger parameter, respectively (Letts et al., 2000). From the sensitivity test mentioned above, it seems that the recommended values from Letts et al. (2000) produced soil moisture and soil temperature close to observations.

The 30 min meteorological observations, including air temperature, specific humidity, wind speed, pressure, precipitation, downward solar, and long-wave radiation, at 36 m height from OAS were used as atmospheric forcing data to drive Noah-MP in an offline 1-D mode. Figure 2 shows the annual mean temperature (1.5 ∘C) and total precipitation (406 mm) at this site during the study period (1998–2009). The most significant climatic features during the study period are a prolonged drought that began in July 2001 and extended throughout 2003, and an extended wet period from 2004 to 2007.

Outputs from the Noah-MP simulations were evaluated against observations, using the root mean squared error (RMSE), square of the correlation coefficient (R2), and index of agreement (IOA) (Zhang et al., 2014). The IOA is calculated as IOA=1-∑i=1NMi-Oi2∑i=1N|Oi-O¯|+|Mi-O¯|2, where Mi and Oi are simulated and observed values of the same variable, respectively, and O¯ is the mean of the observed values. IOA ranges from 0 (no agreement) to 1 (perfect match).

The LSM spin-up is broadly defined as an adjustment process as the model approaches its equilibrium following the initial anomalies in soil moisture content or after some abnormal environmental forcing (Yang et al., 1995). Without spin-up, the model results may exhibit drift as model states try to approach their equilibrium values. To initialize LSMs properly, the spin-up time required for LSMs to reach the equilibrium stage needs to be examined first (Chen and Mitchell, 1999; Cosgrove et al., 2003). In this study, model runs for the year 1998 were performed repeatedly until all the soil-state variables reached the equilibrium state, defined as when the difference between two consecutive 1-year simulations becomes less than 0.1 % for the annual means (Cai et al., 2014; Yang et al., 1995). Yang et al. (1995) discussed the spin-up processes by comparing results from 22 LSMs for grass and forest sites, and showed a wide range of spin-up timescales (from 1 to 20 years), depending on the model, state variable, and vegetation type. Cosgrove et al. (2003) used four NLDAS-1 LSMs to discuss the spin-up time at six subregions covering North America, and showed that all models reached equilibrium between 1 and 3 years for all six subregions. In this study, we found that it requires 9 years for deep-soil moisture (100–200 cm layer) in Noah-MP to reach its equilibrium, 8 years for latent heat flux and evapotranspiration, but only 3 years for the surface soil moisture (Fig. 3). Cosgrove et al. (2003) and Chen et al. (1999) indicated that it takes a long time to reach equilibrium, especially in the deep soil layers and sparse vegetation, because the evaporation was limited by slow water diffusion timescales between the surface and deep soil layers. When using the groundwater component of Noah-MP, it might take at least 250 years to spin up the water table depth in arid regions (Niu et al., 2007). Cai et al. (2014) found that water table depth requires less than 10 years to spin up in a wet region, but more than 72 years for a dry region. For this boreal forest site where the water table depth is shallower (less than 2.5 m), it takes ∼ 7 years for water table depth to reach equilibrium. However, the freezing/thawing is a relatively slow process, so we set 10 years for the spin-up time for all the experiments discussed here.

We defined the simulation without incorporation of organic soil as the control experiment (CTL), and the simulation with the organic soil incorporated as the organic layer experiment (OGN). We first evaluated the CTL- and OGN-simulated soil temperature and moisture at the OAS site in relation to observations for the period of 1998–2009.

As shown in Fig. 4, the effects of including a 10 cm organic topsoil layer on simulated soil temperature are not uniform both throughout the soil depth and during the year. Figure 4a shows that the CTL and OGN simulations produced nearly identical top-layer temperatures which are in agreement with the observations except for a low bias in the winter period, especially during drought years 2002–2003. However, for deep layers (10–100 cm), soil temperature from the OGN is lower (higher) than the CTL simulation during summer (winter), especially for the drought years 2002–2003, leading to a good agreement between OGN and observations for second- and third-layer soil temperature (Fig. 4b, c). Lawrence and Slater (2008) indicated that strong cooling in summer is due to the modulation of early and midsummer soil heat flux, while higher soil temperature in fall and winter is due to less efficient cooling of organic soils. The soil thawing period in spring is significantly affected by the OGN parameterization since the thermal conductivity of the organic horizon is much lower than that of the mineral soil (∼ 0.4 W m-1 K-1 compared to ∼ 2.0 W m-1 K-1), which delays the warming of the deep soil layers after snowmelt. In winter, the organic soil layer insulates the soil and results in relatively higher wintertime soil temperatures for OGN compared with CTL. The difference is most pronounced in drought years (2002 and 2003) (Fig. 4). In summer, due to lower saturated thermal conductivity (0.25 W m-1 K-1 for organic compared to ∼ 6.04 W m-1 K-1 for mineral) in OGN, the downward transfer of heat from the topsoil layer is less and the deep soil temperature in OGN is lower than that in CTL.

In winter, with the presence of soil ice, the thermal heat conductivity in OGN (∼ 2.20 W m-1 K-1) is lower than that in CTL (6.04 W m-1 K-1); it reduces the upward transfer of heat from deep soils to topsoil and therefore results in higher deep-soil temperature in OGN. These results are consistent with studies that showed a simulated increase in winter soil temperature of up to 5 ∘C in boreal regions when including an organic layer (Koven et al., 2009; Rinke et al., 2008; Lawrence and Slater, 2008) in LSMs.

For the topsoil layer, the OGN parameterization increases the liquid soil water content in summer as water fills the larger pore space of organic soil, though the liquid soil water content in winter did not change much, due to the contrasting water retention characteristics of organic and mineral soil (Koven et al., 2009; Rinke et al., 2008; Lawrence and Slater, 2008). Higher porosity in OGN leads to an increase in total soil water content, while the lower topsoil temperature (Fig. 4a) in OGN enhances the ice content. Note that the observed soil water content during wet years may be higher than the site truth because the sensors were located in a low spot that is prone to flooding. This site got flooded in 2004 and the ground water has not dried since then; so the soil was oversaturated during the period of 2004–2008. In the second soil layer, the observed soil water content was incorrect after the site got flooded (2004–2008). With more precipitation during the wet period, the real soil water content should have a relatively high value. Since the OGN increases the soil water content, it should be closer to the true observation. From Fig. 5, it can be seen that the OGN improved the liquid water simulation in non-frozen periods. The soil moisture data are not reliable when the soil is frozen and are therefore not very useful during the winter. In late spring when snow starts melting, both CTL and OGN simulate the same topsoil temperature (Fig. 4). It is clear that the soil liquid water content is mainly controlled by precipitation, soil hydraulic conductivity, and runoff. The high porosity of organic soil in the topsoil layer helps to retain more snowmelt water and hence increases the topsoil layer liquid water content. For the deep soil layers, the soil liquid water content is highly influenced by the soil temperature. Liquid soil water content increases during soil-ice thawing period. The higher deep soil layer liquid water content in OGN is mainly because the soil hydraulic conductivity is higher for organic soil than mineral soil, so liquid water in the first layer can be transported downward quickly into the deeper layers. Although the organic soil layer is only added to the first two layers in this study, it still can affect the deep layer due to the infiltration characteristics of the topsoil.

The water retention characteristics of the organic soil horizon favor both higher water retention and reduced evaporation. The thermal conductivity is lower compared with that of the mineral soil, which then prevents the deeper soil from warming up rapidly after the snowmelt season. The lower thermal conductivity of the top organic soil affects the annual cycle of the ground heat flux. In summer, the top layer is warmer than the deep layers; the ground heat flux then transfers heat downward. Because air temperature is lower than land surface temperature, heat is transferred upward from soil to the land surface; the low thermal conductivity of the organic soil can prevent the soil from cooling. On the other hand, snowfall in winter may form a snow layer that will insulate the soil and make the simulations less sensitive to thermal conductivity. This may be the reason why the OGN-simulated winter soil temperature is higher compared to CTL simulations. With the organic soil layer on the top, the reduction of surface layer saturation levels in wintertime (Fig. 5) reduces the heat loss through evaporation. The winter soil temperature then becomes significantly higher compared with the CTL experiment. On the contrary, the higher soil water content in the topsoil layer during summertime (Fig. 5) increases the heat loss through evaporation; the summer soil temperature then becomes significantly lower compared with the CTL experiment.

Simulated differences in top-layer soil temperature and liquid soil water content lead to the differences in simulated surface energy fluxes. Figure 6 shows that the CTL run captures the observed monthly mean daytime sensible heat and latent heat flux reasonably well. However, SH is underestimated in spring and overestimated in summer. Accordingly, LH is overestimated in spring and underestimated in summer during most of the time period except for drought years 2002–2003 where LH is slightly overestimated. Generally, the OGN simulations show similar characteristics to the CTL, with improved correlation coefficients between observations and simulations: increasing from 0.88 (CTL) to 0.92 (OGN) for SH and from 0.94 (CTL) to 0.96 (OGN) for LH (Fig. 7). Overall, both CTL and OGN perform well in winter when snow is present and fluxes are small. During the spring snowmelt season, the OGN results are much better than the CTL (Figs. 6 and 7).

The OGN simulations also improved the underestimation of SH in spring in CTL, but they still overestimate summer SH. The reason for high bias in summer SH will be further discussed in Sect. 4.4. SH and especially LH show improvement in OGN compared to CTL, which is related to timing of soil thaw and warming in spring. CTL thaws the soil too early, causing a premature rise in LH in spring (April–May) and an associated underestimation of spring SH. The spring (April–May) fluxes are much improved in the OGN parameterization. However, both OGN and CTL retain a serious positive bias in SH from June to September, especially for wet years. The reduction of surface layer saturation levels in OGN led to lower soil evaporation and associated reductions in the total latent heat flux, and the reduction of LH is accompanied by a rise in SH (Fig. 6).

The quality of nighttime flux-tower data is questionable (Chen et al., 2015), especially for OAS located in a boreal forest. Therefore, we focused our analysis on daytime observation data. In general, the OGN parameterization improved the simulation of daily daytime LH in terms of both RMSE and IOA, and increased IOA for SH (Table 3). Nevertheless, compared with CTL, OGN increased the bias in SH slightly by ∼ 3 % (Table 3).

For the 12-year simulation period, the study site experienced a prolonged drought that began in July 2001 and extended throughout 2002 and 2003. We choose year 2002 and 2003 to represent typical drought years, and year 2005 and 2006 to represent typical wet years (Fig. 2), to examine the effect of the organic soil under different climate conditions. For drought years 2002–2003, OGN increased daytime SH especially in spring, and slightly decreased SH at nighttime (Fig. 8a, b, c, and d). LH is well simulated in both OGN and CTL (Fig. 8e, f, g, and h), with slightly increased daytime LH in OGN. OGN overestimates daytime SH compared with observations, while CTL underestimates daytime SH for spring (Fig. 8a). Both OGN and CTL overestimate SH for summer, autumn, and winter (Fig. 8b, c, d).

For wet years (Fig. 9), OGN produces higher daytime SH than CTL in general. For spring, OGN-simulated SH agrees with the observation better than CTL, but it is similar to or slightly worse than CTL for other seasons. Simulated LH for both OGN and CTL agree with observations well, with an improvement by OGN in spring, because the snowmelt process dominates during spring months. For other seasons, the OGN results are close to CTL.

It is clear from Figs. 4, 8, and 9 that in both CTL and OGN, summer sensible heat fluxes are overestimated for wet and dry years. We hypothesized that such high bias in summer sensible heat flux is partly attributed to energy imbalance in observations. We then calculated the energy balance residual term: Rnet - (SH + LH + G) for summer months (June, July, and August). In wet years, G in CTL and OGN is close to observed values; modeled latent heat flux is underestimated by ∼ 10 W m-2; modeled sensible heat flux is overestimated by ∼ 30 W m-2; and the residual term is ∼ 17 W m-2. Hence, it is reasonable to argue that the surface energy imbalance (∼ 17 W m-2) in observations contributes to a large portion of the ∼ 30 W m-2 high bias in sensible heat fluxes. In dry years, the summer energy imbalance (∼ 15 W m-2) is nearly equal to the high bias in sensible heat flux (∼ 15 W m-2).

In the previous section, it is clear that the incorporation of the top organic layer helps improve the simulation of the diurnal cycle of the surface energy and hydrologic components in spring season. In the following, we focus on a detailed analysis of the annual cycle of the surface energy and hydrology variables for dry (Fig. 10) vs. wet years (Fig. 11). Between June and September as shown in Fig. 10h, the upper two soil layers were unfrozen. The topsoil is wetter in OGN for both dry and wet years compared with CTL because organic soil can retain more water. As discussed in Sect. 4.2, for the deep soil layers, the liquid water content is influenced by the soil temperature and the movements of the soil liquid water content between soil layers. Since the soil hydraulic conductivity is higher for OGN than mineral soil, the water moves faster into deep soil layers than CTL; therefore the OGN simulates higher soil liquid water content in deep layers. OGN has a major impact on the daily cycle of soil temperature. Consistent with discussions in Sect. 4.2, the soil temperature below 10 cm simulated by OGN is lower in summer and higher in winter than that of the CTL simulation, and the OGN simulation shows less bias than the CTL simulation (Fig. 4). In the OGN simulation, the water moves faster into deep layers than in the CTL simulation, leading to more infiltrated water in the deep soil and hence a higher base flow. Consequently, the total runoff is increased. Due to the high soil porosity of the organic soil, OGN simulation shows higher soil-ice fraction at the topsoil layer during the freezing periods. The higher water capacity and higher soil-ice fraction of the organic soil then reduce liquid water content/soil moisture, leading to less evaporation (i.e., latent heat flux) during spring freezing periods, and a compensating increase of the sensible heat flux.

By adding an organic soil layer, the soil-ice content becomes higher due to higher porosity. For dry years, the impact of the organic soil on surface and subsurface runoff is not significant (Fig. 10e, f). The increase in the summer latent heat flux and sensible heat flux are compensated by a decrease in soil heat flux, leading to a significant decrease in summer soil temperature. In winter, the latent and sensible heat fluxes are not modified by the organic soil, but increased soil heat flux leads to an increased soil temperature in winter. The most prominent change by including the organic soil layer is the partition between vegetation transpiration and direct ground evaporation (Fig. 12a and b), where the OGN simulation slightly increased ground surface evaporation and vegetation transpiration.

For wet years (Fig. 11), the impact of the organic soil on surface and subsurface runoff becomes more significant, especially for subsurface runoff. The organic soil decreases the surface runoff during the summer season, and increases the subsurface runoff during the freezing periods, while it decreases the subsurface runoff during summer season. Because of the higher surface layer soil-ice content, the increase of subsurface flow should be due to the production of a wetter soil profile by OGN. The sensible heat flux also increases significantly in spring, with an associated reduction in latent heat flux and soil heat flux. The summer soil temperature also decreases but to a lesser degree than that in dry years, because the soil heat flux decreases less compared with dry years. Unlike dry years, there is a significant runoff change in wet years, and the ground evaporation is also decreased (Fig. 12c and d). OGN produces more soil-ice content and higher soil porosity, and leads to higher soil water content than CTL simulations as the higher ice content severely restricts movement of water out of the soil column. In the wet season, by adding an organic topsoil layer, the soil water increases due to the infiltration of the soil water into the deep soil. This then leads to an increase in the subsurface runoff. As a consequence, the volumetric liquid water becomes higher in summer for OGN compared with CTL simulation.