Two potential equilibrium states in long-term soil respiration activity of dry grasslands are maintained by local topographic features

Spatial patterns of stability proxies and background variables

Figure 2 a, b show the spatial distribution of our two proxy variables, the average rank of R_s per position (rankR_s) and of the range of the ranks per position (rangeR_s) in kriged maps. The middle to southern areas were found to have the largest, whilst the north-eastern areas the smallest rankR_s values, whereas a slightly different pattern was characteristic for rangeR_s with some additional north-western large values. Similarly, larger average soil organic carbon content (meanSOC) and average soil water content (meanSWC) (Fig. 2 c, d) were detected at the western-middle-southern regions and smaller at the north-eastern part of the study site.

Correlations between stability proxies and background variables along DEMs: entire dataset

We investigated the potential direct effects of the different terrain attributes (local mean elevation (mALT), standard deviation of elevation (SD), topographic position index (TPI), slope (Sl), Easterness and Northness (East, North)) on the spatial distributions of our proxy variables by using the terrain attributes originating from differently smoothed DEM rasters. DEM1 was the original, 0.2 m resolution model, while DEMs 2–6 were progressively smoothed by a factor of two resulting in different resolution DEM rasters (DEM2: 0.4 m, DEM3: 0.8 m, DEM4: 1.6 m, DEM5: 3.2 m, DEM6: 6.4 m, respectively), and finally DEM7 met the resolution of the field measuring campaigns (10 m). The terrain attributes were filtered out from the rasters for the 78 measuring positions of the sampling grid.

On the basis of the correlation analysis we found an important difference in terrain attribute features between DEM 5 and 6, especially in SD, Sl, North and East. All subsequent results are then based on DEMs 1–5, which were found to be more similar to each other and to the original DEM1. The maps of terrain attributes with the box blur kernel from DEM1-5 can be found in the Supplementary Information (SI) together with the descriptions and calculations. As we couldn’t find any of the blur kernels superior to the other when considering correlations, the results hereafter are only presented for the box blur kernel calculations for simplicity.

When we considered the entire dataset (named hereafter: “A” dataset), we could only find significant correlation between rangeR_s and TPI at less smoothed DEMs but the correlation was very weak (black symbols and line in Fig. 3).

Any other correlation between the proxies and the terrain attributes could only be deduced indirectly from the positive correlations between rankR_s and meanSOC, meanSWC (cf. Table 1b). These correlations were scale-independent, i.e., we detected them at every DEMs. In general, the larger the soil carbon content and soil moisture at a position (cf. Figure 2c,d, showing quite similar patterns to the proxy patterns in the figure upper row), the larger the R_s activity detected and the opposite was true for lower carbon content and soil moisture positions.

As we investigated the background of these correlations more thoroughly in dataset A in terrain attributes (cf. the maps in SI, Table 1a), we found that meanSOC showed negative correlation with ALT, mALT, SD and Sl, while positive with North and East at DEMs 1–5. Similarly, meanSWC correlated negatively to SD, Sl, except for DEM5. Several terrain attributes were then responsible for the meanSOC patterns, i.e., higher absolute elevation and neighbouring surface heterogeneity, as well as steeper slope positions facing more South-West could be characterized with lower meanSOC. The opposite features were characteristic for higher meanSOC level positions on lower elevations with lesser neighbouring heterogeneity and gentle slopes facing mostly North and East. Similarly, meanSWC was higher at smaller surface heterogeneity with more gentle slopes, while lower at more heterogeneous surfaces with steeper slopes. rankR_s followed these patterns with higher R_s activity in the middle-southern part of the study area, while, in contrast, lower R_s activities were characteristic at lower meanSOC and meanSWC at the north, north-east facing locations in the study site, mainly on local ridges as found on the basis of the direct TPI correlations.

Correlations between stability proxies and background variables along DEMs: subgroups

We also checked the correlations within different data subgroups corresponding to specific rankR_s or meanSOC categories because we hypothesized that these kinds of groupings could enable us to grasp some important characteristics of the stability.

Subgroups:

• Subgroups were created on the basis of rankR_s ± SD: S (Smaller than mean-sd), M (Middle between mean ± SD), L (Larger than mean + SD).

• C1, C2, C3, C4, C5 (from the smallest to the largest meanSOC quintiles).

Direct correlation between terrain attributes and proxies showed considerable variation depending on the subgroups and variables (Table 1b collects scale-independent correlations, valid at almost each of the DEMs between 1 and 5, or scale-dependent ones, valid only in several of the less smoothed DEMs 1, 2, 3).

It seems that the meanSOC pattern related negatively to ALT, mALT detected in dataset A could have acted as a driver for the negative rankR_s and ALT, mALT correlations in the groups M and C1. It was in subgroup C1 that TPI, SD and Sl acted negatively on rankR_s as well, most probably more directly through the patterns generated by terrain attributes in meanSWC. Further negative correlations were found between rangeR_s and TPI in S data (see also blue symbols and line in Fig. 2), like in dataset A (cf. Fig. 3 black symbols and line), as well as between rangeR_s and North in C4. Accordingly in the long run, local valleys but mostly constant slope positions (with TPI close to zero, cf. blue symbols in Fig. 3) with lower neighbouring surface heterogeneity and gentle slopes with more elevated meanSOC and meanSWC could be characterized with larger R_s activity with higher variability (through the negative TPI–rangeR_s correlation) in these subgroups per se, similarly to dataset A. The opposite was likely to be the case for local ridges.

The subgroups mentioned here were restricted groups of measuring positions, where carbon content in the soil was the lowest of all (C1) or, as in subgroup S, coincided with low meanSOC levels (Fig. 4), and low meanSWC levels as well.

Furthermore, measuring positions, grouped by either on the basis of rankR_s or on the basis of meanSOC, occupied more or less well delimited spatial areas within the sampling grid (Fig. 5, positions coloured according to C1–C5, where e.g., C5 category, indicated with asterisk occupied the lowest altitudinal positions, C4 was found mostly around C5, while C1 category could be found along the edges of the study area on the crests), which would also be characteristically different in respect of the terrain attributes, especially in SD, Sl, as found by the correlations (cf. Table 1).

Finally, rangeR_s was fitted to rankR_s using the following equation:

where µ is the mean rankR_s (42.61), σ is the standard deviation of rankR_s (25.38), and a (4,291.73) is a model parameter. The correlation, approaching a bell-shaped curve (Fig. 6, both the curve and the model parameters are statistically highly significant, p < 0.0001), visualized together with the subgroups showed that both low and high rankR_s could be associated with small rangeR_s with larger stability and a typically resistant response, while middle values of rankR_s corresponded to larger rangeR_s with a more flexible, resilient response of R_s activity. Furthermore, it was also showed that rankR_s categories more or less fitted to C1–C5 meanSOC categories (cf. square symbols of C1 in S-M-group regions, C2 in M, while asterisks mostly in the upper half of the rankR_s range), giving strong evidence of SOC as a controlling factor in R_s stability. The smallest and the largest rankR_s values could correspond to the largest potential stability (in terms of resistance) in the activities, rankR_s being either low in general (cf. Fig. 2a north and north-east regions) due to low meanSOC and meanSWC (Fig. 2c,d) or high (cf. Figure 2a more the middle and southern regions within the study plot) in the opposite cases. On the other hand, medium rankR_s with larger rangeR_s overlapped spatially with C2-C3 and M groups with medium meanSOC levels, and these positions showed a more resilient response.

Source: Ecology - nature.com