Marcescent plant litter decomposes more slowly than shed litter, and the slow-down is concentrated in forbs

Summary

Across 34 herbaceous species grown in a common garden, litter that stood as persistent standing-dead (marcescent) tissue lost substantially less mass than tissue that was directly shed: species-level mean mass loss was 38.5% for marcescent versus 56.6% for shed litter, a large and highly significant paired difference (Wilcoxon signed-rank p < 0.001, z = −4.49, matched-pairs rank-biserial r = 0.77; marcescent decomposed more slowly in 30 of 34 species). The effect depends strongly on functional group: forbs showed a marcescent-minus-shed deficit of −23.1 percentage points versus only −2.2 for grasses (Mann-Whitney U = 18, p = 0.0007), and the marcescent slow-down was significant within forbs (p = 0.0001) but not within grasses (p = 0.24). Microbial biomass on the litter (Cmic) was positively associated with decomposition (Spearman ρ = 0.43, p < 0.001, n = 58). All findings except the within-grass test survive Benjamini-Hochberg and Holm multiplicity correction. This confirms, quantitatively, that marcescence retards decomposition — but as a forb-specific phenomenon rather than a uniform one.

Provenance and methods

Input was the single deposited workbook DataMarcescencedecomposition fin.xlsx, downloaded by Figshare id and md5-verified (0c66ac4ef3d82ef09ed2c0332a4796b6) before use. The Decomposition sheet gives, per litter sample, the species, family, functional group (Fgroup: Forb/Grass/Legume), replicate, litter type (I initial, M marcescent, S shed), and percent mass loss (LossMass [%]); the PLFA sheet gives microbial biomass (Cmic, µg C/g) per species/litter/replicate.

The analysis was pre-registered before execution (see claude-science-brief.md). To guard against pseudoreplication, mass loss was first averaged to one value per species per litter type; the 34 species with both a marcescent and a shed value form the paired sample. Four declared tests were run: T1, a paired Wilcoxon signed-rank test of marcescent vs shed mass loss; T2, a Mann-Whitney U test of the per-species (marcescent − shed) difference between the two well-sampled functional groups (Forb n = 27 vs Grass n = 7; Legume, n = 2, excluded as too few); T3a, the within-group paired test for each of forbs and grasses; and T3b, a Spearman correlation of microbial biomass against mass loss on the joined sample. The correlation/contrast family {T2, T3a-forb, T3a-grass, T3b} was corrected with both Benjamini-Hochberg (FDR) and Holm. All computation is deterministic (analysis.py, seeded).

Data records

Three figures and three tables were produced. fig-1 is the paired species-level marcescent-vs-shed mass loss. fig-2 is the per-species difference by functional group. fig-3 is microbial biomass against mass loss. tbl-1 is the species-level marcescent/shed/difference table; tbl-2 the per-functional-group summary with within-group tests; tbl-3 the raw, BH- and Holm-adjusted p-values across the test family. A Frictionless Data Package describes the tables.

Technical validation

The source file was md5-verified before use. The primary result is strong and consistent: marcescent litter decomposed more slowly in 30 of 34 species (T1, p < 0.001, large effect). The functional-group dependence is significant (T2, p = 0.0007; BH-adjusted 0.001) and coheres with the within-group tests (forbs p = 0.0001, BH 0.0002; grasses non-significant, p = 0.24). The microbial-biomass association (T3b, ρ = 0.43, p < 0.001; BH 0.001) is a correlation, not a manipulated effect. No test outside the pre-registered set was run. The one non-significant result (within-grass) is reported as such.

Usage notes

This is a mass-loss and functional-group analysis, not a mechanistic decomposition model. The functional-group contrast rests on 27 forb vs 7 grass species and excludes legumes (n = 2); the grass result is under-powered rather than a demonstrated null. The Cmic association is observational and cannot separate microbial cause from litter-quality confounds (marcescent tissue differs in C:N and chemistry). Reusers should treat the marcescence slow-down as a robust, forb-concentrated effect and the microbial link as a hypothesis for follow-up.

Code availability

analysis.py is self-contained: it downloads the pinned file by Figshare id, verifies md5, runs the four pre-registered tests with multiplicity correction, and writes the figures, tables, and results.json. Re-running reproduces every number.

Claims

See claims.json for the atomic, individually addressable assertions, each tied to the figures and tables that support it and each carrying its statistics and confidence.