2.5 Data calculations and statistical analyses
The soil P activation coefficient
and available P were characterized by P availability, and the P
activation coefficient was the proportion of available P content in
total P content (Li and Zhang, 1994; Sharma and Chowdhury, 2021). All
statistical tests were performed or produced using SPSS 18.0 software
(IBM Corp., Armonk, NY, USA). All data were tested using a z-test for
normality before analysis. Data meet normality if absolute z-scores for
skewness or kurtosis (skew or kurtosis values dividing their standard
errors) are less than 1.96 and 3.29 in small (n < 50) and
medium-sized (50< n < 300) samples, respectively.
Thus, some data were square root transformed (total P, P activation
coefficient) or log10 transformed (P forms excluding Ca10P), exponent
transformed (pH) to meet normality. Two-factor analysis of variance was
conducted to determine the effects of marsh degradation and soil
sampling depth on soil P and its forms; one-way analysis of variance
(ANOVA) followed by Duncan’s test was used to compare the differences in
the concentrations of available P, P activation coefficient and other
basic physiochemical properties among differently degraded marsh soils.
Difference was considered to be significant when p <
0.05. The figures except for the
structural equation model were drawn in Origin 9.0 (OriginLab Corp.,
Northampton, MA, USA).
Structural equation model,
combining factor analysis and path analysis, was broadly employed to
analyse the relationship between variables (including unobserved
variables) based on the covariance matrix of variables in natural
ecosystems (Bai et al., 2020; Hou et al., 2016; Lefcheck, 2016; Melese
et al., 2015). Therefore, we used a structural equation model to further
identify the effects of different P forms in regulating available P
variations and the complex interactions among P forms in soils.
Theoretical and experimental studies demonstrated that soluble P
(available P) was regulated by the direct and indirect effect of various
P forms
(Gama-Rodrigues
et al., 2014; Hou et al., 2016; Liu et al., 2019; Melese et al., 2015),
and the changes in soil environments (e.g. plant community and
hydrography) also influenced the transformation of P forms to available
P (Hu et al., 2021; Wang et al., 2021b). Thus, a
conceptual structural equation
models of the direct and indirect effects of soil P forms on available P
under marsh degradation are constructed in Figure 2 based the following
hypothesis: (1) marsh degradation has an indirect effect on available P
by the mobilisation of stable and moderate labile inorganic P, and
organic P; (2) the mobilisation of soil P in alpine wetlands mainly
follow a process from organic P to inorganic P, and from stable P to
moderate labile P and labile P in turn (Hou et al., 2016). The model was
fitted using a combined data from the three sites, four types of marsh
degradation, and seven depths, totalling 84 observations, because P
forms and available P from the same profile have the significant
differences (Table 4; available P of one-way analysis of variance: F
statistic value 5.185, p < 0.05). Thus, the minimum
sample size can meet a general rule
that the ratio of the total
number of samples to the number of variables is 5 times (Lefcheck,
2016). The categorical variables of marsh degradation were indicated by
the wetland degradation index that was calculated by Qinghai Provincial
Standards “Degradation classification of alpine marsh wetlands”
(DB63-T-1794-2020). The wetland degradation index was equal to the
average of water area ratio and importance value of hygrophytes.
Structural equation model was performed using the SPSS Amos 24.0
software package (IBM SPSS Inc., Chicago, IL, USA) after all variables
excluding marsh degradation were log10 transformed.
The best-fit model was derived
using the maximum likelihood based on
the relative/normed chi-square
test (χ2/df), normed fit index
(NFI), comparative fit index (CFI), and root square mean error of
approximation (RMSEA) of model fit (Hooper et al., 2008; Grace and
Bollen, 2005). Model fits with χ2/df ranges from 2.0
to 5.0, NFI > 0.95, CFI > 0.95, and RMSEA
< 0.08 can be accepted as a good structural equation model
(Hooper et al., 2008; Grace and Bollen, 2005).