Seasonal synchrony inflates spatial dependence in water-quality Markov chains: evidence from a fine-scale agricultural wetland.

Spatial Markov chains are increasingly used to infer spatial dependence among water-quality monitoring sites, usually from raw records retaining the common seasonal signal. We tested whether that dependence reflects interaction between sites or synchrony to a shared seasonal driver. Classic and spatial Markov chains were applied to full-year monthly water-quality index (WQI) records from eight connected sites in a Taihu-basin agricultural wetland (2023-2025, 288 site-months). On raw data, a site's transitions depended significantly on its neighbours' state (p = 0.038), but the own-versus-neighbour state table was near-degenerate, 22% of its cells near-empty, because synchrony kept each site and its neighbours in the same state. After deseasonalization, the table was fully populated and the dependence was no longer detectable (p = 0.60). The reversal held across six spatial-weight specifications (deseasonalized p = 0.25-0.96, against 0.0002-0.077 on raw data) and was corroborated by a Moran ladder (pooled Moran's I 0.80 to -0.11; spatial-autoregressive ρ 0.70 to -0.16). A two-way variance decomposition attributed 81.6% of WQI variance to time and 7.9% to site, while the production season opened a direct Good-to-Poor collapse channel absent in winter. A power analysis places the design's detection limit at a conditional-matrix deviation of about 0.19, above the deseasonalized value of 0.09, so weak spatial interaction cannot be excluded. The apparent dependence is therefore attributable largely to seasonal synchrony, the Moran effect, rather than to spillover. Such models should be deseasonalized, and their table sparsity and detection limit reported, before their dependence is read as spatial process.