The most accurately at all microsites determined was to be the

R_{ref} parameter (CV at all studied microsites did not exceed 2%, which is significantly less than the error, artificially added to the input data with

CV = 15%). This was probably due to a sufficiently large number of measurements performed at each microsite. As noted above,

R_{ref} is the soil respiration at the average measured soil temperature (

T_{ref}), which in this case shows similar (with measured

R_{soil}) differences between microsites:

R_{ref} increased in the DEP–FL–EL–TUS–STUS sequence, i.e., 40, 73, 120, 171 and 277 mgC m

^{−2} h

^{−1}, respectively. In this case,

T_{ref} was ≈13 °C. For some microsites, the coefficient of variation of

Q_{10} turned out to be slightly higher: we were able to determine it most accurately (

CV = 5%) for the microsite on a flat surface (FL). In addition to FL, the accuracy of determining the

Q_{10} parameter increased with an increase in the height of the microsite position in the relief and, accordingly, the degree of its drainage: in the DEP–EL–TUS–STUS sequence,

CV of

Q_{10} was 10, 9, 7, and 6%. It can be assumed that the accuracy of determining the

Q_{10} parameter would have to depend, not on the error artificially introduced to check the stability of the model, but on the natural error caused by the variation of fluxes that is maximal at FL and minimal at STUS (

Figure 4). However, in this case, we are dealing with different

CVs. When considering the measurement results, this is the variability of the measured soil respiration in principle, without reference to any specific temperature or hydrological conditions, whereas in the case of the model,

CV illustrates the stability of the determination of the parameters when artificial error is introduced into the input data. If the error of the determination of the parameters is significantly (by an order of magnitude) greater than the error of the input data (15% in this case), then this model is ill-conditioned [

49,

50]. In the latter case, one of the solutions may be a reduction in the number of model parameters, that is, its simplification, or an increase in the amount and variety of input data.

When parametrizing the model, the dependence on GWL was reflected in a somewhat worse way than that on soil temperature: CV of parameter a at the microsite TUS reached 38%, while at EL and STUS—14–15%. CV of parameter b did not exceed 5% for all microsites.

In general, the use of linear dependence between

R_{soil} and GWL was dictated by somewhat limited information on the level of soil respiration at different moisture levels. GWL most often fluctuated in a narrow range near the soil surface over the four-year period of regular observations. We were able to register periods of sharp declines in GWL only in some months of 2014 and 2015. On the one hand, this problem can be resolved in the future by conducting more prolonged and more frequent observations, and on the other hand, by monitoring GWL at each microsite independently. Nevertheless, parameter

a, which characterizes the slope of the straight line in the dependence between

R_{soil} and GWL, increases from drier to moister microsites, which illustrates the inverse relationship between soil respiration and moisture, and a large contribution of GWL to the variability of

R_{soil} in depressions. Taking into account that

CVs of all parameters (except for

a at the microsite TUS) did not exceed

CVs of the input data with artificially introduced “noise,” the parameterization can be considered successful, and the model stable and well conditioned [

49,

50].