Rivers of Memory: Mapping How Henan’s Heritage Lives Along Its Waterways

China’s inland waterways are being reconsidered for transport and waterfront development, and a new study asks a question that has long gone underquantified: how do river-network environments correspond to the contemporary cultural, social, and economic vitality of nearby heritage sites? Using 371 nationally protected heritage sites across Henan Province, researchers constructed a Heritage Site Vitality Index (HSVI) from seven min–max-normalized cultural, social, and economic indicators, with weights derived from the analytic hierarchy process. The analysis progressed from heritage spatial organization to river-network–vitality configurations and then to site-level associations, using pure-spatial DBSCAN, bivariate spatial analysis, ordinary least squares (OLS), geographically weighted regression (GWR), and locally compensated ridge GWR (LCR-GWR). The findings, published in Frontiers in Environmental Science, provide a diagnostic basis for differentiated heritage activation and spatial planning across a province spanning four river basins.

Henan Province, located in central China, spans the Yellow, Huai, Hai, and Yangtze river basins. Its west-to-east topographic gradient, dense plain river network, mountainous headwater areas, and long history of river-basin-based settlement development create substantial environmental and historical variation within a single administrative region. The expansion of railway and highway transport weakened the contemporary transport function of many inland rivers, but recent low-carbon transport policies and the continued role of waterway infrastructure in regional economic geography—including Henan Province’s ‘11246’ project—have renewed attention to inland navigation. As river corridors are reconsidered for transport and waterfront development, planning must also determine how water-network conditions correspond to the contemporary vitality of nearby heritage.

Existing inland-waterway research focuses primarily on engineering capacity, port and logistics performance, regional economic spillovers, and environmental effects. These frameworks explain how waterway infrastructure affects transport efficiency and regional development, but they usually treat cultural heritage as a background resource or an additional development sector rather than a principal evaluative dimension. Spatial cultural-heritage research has developed point-pattern analysis, accessibility evaluation, corridor and network identification, and multivariable explanations of heritage distribution, but these approaches rarely examine the relationship between those sites and contemporary inland water networks. Heritage-vitality and attractiveness studies use visitation, online attention, public services, and surrounding urban activity to evaluate contemporary heritage use, but generally do not place that activity within river-network and waterfront environments.

The study asks three questions: how are historically accumulated heritage resources organized spatially under the prespecified DBSCAN baseline, and how sensitive is that organization to parameter changes; where do heritage-vitality density and river-network density show synergy or directional mismatch across grid scales; and how do six explicitly defined site-level conditions relate to heritage vitality in global spatial models and geographically varying local models. These questions form one analytical chain: pure-spatial DBSCAN establishes the inherited spatial organization of heritage resources; KDE and bivariate Moran analysis identify province-wide river-network-vitality synergy and mismatch; OLS, spatial lag, and spatial error models provide the global site-level baseline and account for spatial dependence, while GWR and Monte Carlo tests identify geographic variation in the site-level relationships; LCR-GWR then evaluates whether the local coefficient surfaces remain stable under local collinearity.

DBSCAN revealed a dominant central-western agglomeration belt extending from Zhengzhou to Luoyang and a more dispersed peripheral pattern. The largest cluster contains 207 sites and connects the Zhengzhou-Luoyang concentrations into a broad central-western belt. Separate compact clusters occur around Kaifeng and Anyang. The Shangqiu cluster is dominated by the Han Liang royal mausoleums, while the Zhumadian cluster is dominated by the Zhugou revolutionary heritage complex; smaller northern clusters contain different combinations of city, tomb, temple, and settlement sites. These compositions show that the clusters represent several historical processes rather than one waterway-related heritage type. Noise points are more common in peripheral areas including Nanyang, Xinyang, and Sanmenxia, indicating dispersed spatial organization outside the principal density concentrations. Across the 15 prespecified parameter combinations, DBSCAN produced 2–13 clusters, showing that cluster boundaries depend on neighborhood scale and minimum cluster size.

At the 5-km grid scale, heritage-vitality density and river-network density showed a positive global spatial association (bivariate Moran’s I = 0.2109; permutation p = 0.001). High vitality–high river-network configurations occurred in parts of the central and northern plains, high river-network–low vitality configurations in parts of the eastern and southern plains, and several heritage agglomerations displayed high vitality despite low river-network density. Global bivariate Moran’s I values on the 2.5- and 10-km grids were 0.212908 and 0.206348, respectively, showing that the province-wide association and the coexistence of synergy and mismatch are stable across the three tested scales. Under the 14 one-at-a-time ±20% perturbations, Spearman’s ρ with the baseline ranged from 0.9983 to 1.0000, top-quintile overlap ranged from 94.7% to 100.0%, and global Moran’s I ranged from 0.3655 to 0.3850, compared with 0.3758 for the baseline. Moderate reweighting therefore preserves the principal site ranking and spatial autocorrelation pattern, while omission of a complete information component defines a clearer sensitivity boundary.

The OLS model is statistically significant overall, but its adjusted R² is 0.1201. A single province-wide combination of the six measured conditions therefore explains only a limited share of contemporary heritage vitality, leaving substantial variation associated with socioeconomic conditions, governance, visitor flows, conservation investment, and measurement factors. The Jarque-Bera test rejects residual normality. OLS residuals also remain significantly spatially autocorrelated under the primary symmetric 8-nearest-neighbor weights (Moran’s I = 0.3518; permutation p = 0.001). The significant LM-lag, robust LM-lag, LM-error, and robust LM-error tests show that the unexplained variation has a global spatial structure and justify comparison with spatial lag and spatial error models. Across k = 6, 8, 10, and 12, both models reduce residual spatial dependence relative to OLS. At k = 8, the spatial lag model has ρ = 0.6681 and the spatial error model has λ = 0.7144, and the filtered residuals are nonsignificant.

Adjusted R² increased from 0.1201 for OLS to 0.6000 for GWR, showing that geographically varying coefficients captured substantially more variation in heritage vitality. The 35-nearest-neighbor adaptive bisquare GWR produced six local coefficient surfaces. The Monte Carlo and LCR-GWR diagnostics support local spatial interpretation for heritage agglomeration, riverbank-line richness, and riparian greenness. River cost distance and land transportation accessibility remain regional-level control variables, while the Topographic Wetness Index remains a terrain-background control. The 999-iteration Monte Carlo tests support spatial variation in heritage agglomeration (p = 0.001), land transportation accessibility (p = 0.001), riverbank-line richness (p = 0.001), and riparian greenness (p = 0.006). River cost distance is marginal at the 0.05 threshold (p = 0.051), and TWI is nonsignificant (p = 0.992). LCR-GWR was then applied to the same 371 observations, with a location-specific penalty activated at 328 sites. R² changed from 0.7531 to 0.7293 and RMSE from 0.0378 to 0.0396. The residual Moran’s I of the original GWR remained significantly negative (−0.0926; permutation p = 0.001).

Coefficients for heritage agglomeration, riverbank-line richness, and riparian greenness changed sign across the province, while their spatial patterns remained relatively consistent under LCR-GWR. Local coefficients for heritage agglomeration range from −0.1096 to 0.1160. Coefficients are negative in parts of central and eastern Henan and positive at several peripheral sites, indicating spatial differentiation in the relationship between heritage agglomeration and vitality. Local coefficients for riverbank-line richness range from −1.3545 to 7.1865, with localized positive concentrations coexisting with extensive areas whose coefficients are close to zero. Local coefficients for the riparian greenness proxy range from −1.8595 to 1.0468, with negative coefficients predominating while some sites have positive coefficients, indicating spatial differentiation in the relationship between riparian greenness and heritage vitality. Together, the regional configurations and local coefficients show how heritage organization, riverbank form, and riparian ecology differentiate river-network–vitality relationships across the province.

The planning diagnosis combines the three river-network-vitality configurations with the robust GWR and LCR-GWR coefficient surfaces. The configurations identify the regional planning problem, while heritage agglomeration, riverbank-line richness, and riparian greenness identify the site-level conditions for field investigation. In high vitality-high river-network areas, planning should coordinate heritage use, river-corridor conditions, and service carrying capacity. Riverbank-line richness and riparian greenness can guide waterfront landscape, vegetation, water-quality, viewshed, and habitat surveys, while local evidence determines whether interpretation or recreational improvements are appropriate. In high river-network-low vitality areas, field investigation should examine heritage visibility, service provision, governance, and conservation investment. Physical interventions should also be evaluated against slope stability, flood exposure, water-erosion risk, and habitat continuity. High vitality-low river-network areas show that contemporary activation may depend more strongly on urban services, cultural communication, heritage agglomeration, or other local conditions than on dense river-network settings. Reversals in the heritage-agglomeration coefficients can guide comparison of coordination, service allocation, and competition among nearby sites. The study advances river-basin heritage analysis from resource mapping to the diagnosis of contemporary activation, providing a methodological reference for corresponding spatial comparisons in other river basins.

Subject of Research: Science

Article Title: Integrating cultural heritage into inland waterway revitalization: spatial heterogeneity and differentiated planning strategies in Henan, China

Article References: Li, H., Bai, Q., Song, Y., & Zhang, X. (2026). Integrating cultural heritage into inland waterway revitalization: spatial heterogeneity and differentiated planning strategies in Henan, China. Frontiers in Environmental Science, 14, Article 1928710. https://doi.org/10.3389/fenvs.2026.1928710

Image Credits: AI Generated

DOI: 10.3389/fenvs.2026.1928710

Keywords: cultural heritage, inland waterways, spatial analysis, geographically weighted regression, Henan, river basins, spatial planning, China, Integrating, cultural, heritage, inland