Enhancing the interpretability of ecological constraint mechanisms of urban sprawl with GIS-integrated ordered weighted averaging

Abstract Data aggregation of ecological indicators is the basis of constructing ecological constraint mechanisms for urban development. The aggregation method of geographic-information-system–integrated (GIS-integrated) ordered weighted averaging (OWA) has attracted significant attention because it reflects decision-makers’ risk-taking attitude and controls the trade-off between indicators. However, the OWA algorithm disrupts the ranking information of the indicators in the process of data aggregation, which breaks the direct link between the evaluation results and the input factors. We selected the Golden Triangle of Southern Fujian Province along the southeastern coast of China to carry out a multi-scenario decision-making analysis of ecological space constraints to interpret the GIS-OWA evaluation. We develop an ecological constraint index that considers the importance of ecosystem services and the sensitivity of the ecological environment. This ecological space soft-constraint index uses a regular increasing monotone fuzzy semantic operator to generate the order weights for OWA. On this basis, we propose two indexes of attribute contribution and location contribution to investigate the main factors that affect the GIS-OWA evaluation from pixel scale to local scale. The results show that the factor ratios on the ranked layer differ significantly and undergo spatial clustering. The fractions of the three factors (i) importance of water conservation (IWC), (ii) slope, and (iii) normalized difference vegetation index (NDVI) on the first ranked layer are 40.4%, 15.9%, and 15.7%, respectively. The dispersion in attribute value widens the range of the attribute contribution beyond that of the location contribution. When the exponent is 0.3, the three factors IWC, importance of biodiversity (IB), and NDVI of the cell (row: 1418; column: 951) have attribute contributions of 0.724, 0.133, and 0.070, respectively. A set of key factors exists in the distribution of the attribute contribution at the pixel scale, and the set remains relatively stable for different evaluation strategies. The sets of key factors for cells in different OWA-equivalence areas differ significantly, and an optimal window size exists for the spatial scope of the set under a given threshold. Thus, the contribution index is an appropriate tool for key-factor analysis with the GIS-OWA algorithm.