In silico simulation of dieback and resprouting (repositioning) in woody crown hydraulic networks

This deposit contains the source code, per-replicate data, aggregated statistics, dependency specification, and bilingual documentation for two in silico experiments on the Woody Crown Hydraulic Network (WCHN; RHLC in Portuguese). The WCHN/RHLC represents a woody crown as a rooted directed acyclic graph, following the framework developed by Prado (2024) and Prado and Trovão (2023). The experiments isolate two successive topological phases of drought-related crown regression and reconstruction. In the first experiment, cladoptosis or dieback is simulated by removing all final nodes at each round, progressively reducing the total number of nodes and woody axes. In the second experiment, repositioning or resprouting is simulated by transferring each eligible final node from its parent to its grandparent through lateral insertion. This operation preserves the total number of nodes while generating additional woody axes. Final nodes already connected directly to the initial node are not repositioned because they have no grandparent. Each experiment was conducted on 40 random networks over five successive rounds at two scales: 101 nodes and 900 nodes. The larger scale approximates the largest empirically reconstructed individual of Cenostigma pyramidale, a Caatinga species, with approximately 825 nodes. The 40 replicates are independent within each scale. The two scales are not statistically independent, because the 101-node network represents an earlier construction stage of the corresponding 900-node network generated with the same replicate-specific random sequence. Both experiments begin with exactly the same initial networks at each scale, allowing controlled comparison of their contrasting topological effects. Eight WCHN/RHLC properties are recalculated at every state: navigability, laterality, basitony, node sharing, potential topological plasticity, vulnerability, normalized symmetry, and normalized complexity. Vulnerability is calculated under the reformulated definition in which all nodes, including final nodes, are attackable except the initial node (root). Consecutive states are compared using two-sided paired Wilcoxon tests (rounds 0–1, 1–2, 2–3, 3–4, and 4–5; n = 40). Raw p-values are reported without multiple-comparison correction because the analysis is exploratory. Dispersions are calculated using population standard deviations. The simulations reveal contrasting topological signatures. Cladoptosis progressively reduces network size and most organizational properties while strongly increasing vulnerability. Repositioning/resprouting preserves network size, increases navigability, laterality, node sharing, and potential topological plasticity, and decreases vulnerability. These directional patterns are reproducible at both network scales. A fixed random seed (SEED = 2026) makes network generation and the per-replicate CSV deterministic. The computational environment was validated with Python 3.12.13 and SciPy 1.17.0 on Linux; the pinned dependency is provided in requirements.txt. In a compatible environment, running the script regenerates the deposited CSV and JSON files exactly. Contents: rhlc_dieback_rebrota.py — network generator and implementation of both experiments. experimentos_por_replica.csv — 960 per-replicate records: 2 experiments × 2 scales × 40 replicates × 6 states. summary.json — per-round means, population standard deviations, coefficients of variation, and paired Wilcoxon tests. requirements.txt — pinned external dependency required for statistical testing. README.md — bilingual methodological, computational, and reuse documentation. The source code is released under the MIT License. The data and documentation are released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).