PSR-BP Neural Network-Based Health Assessment of the Huangshui Plateau Urban Wetlands in China

Abstract

Wetland health assessment provides important basis for wetland restoration and management. However, it is quite tricky to select proper indicators from multiple assessment indicators that can truly reflect the health state of urban wetlands. In an attempt to address these problems, a pressure-state-response (PSR) and back propagation artificial neural network (BP) conjoined model was established for health assessment of several plateau urban wetlands in Xining City, China. The model was driven and verified through field monitoring and social questionnaire data for 4 consecutive years from 2016 to 2019. Results indicate that: (1) Eight health evaluation indexes, including population density, eutrophication level, increasing humidity, carbon dioxide absorption, air purifying, recreation, wetland management level and investment in ecological construction and protection were selected from 45 input indexes. (2) The health index of Huangshui National Wetland Park has been increasing year by year, with an average of comprehensive health score of 0.746, 0.790, 0.884, and 0.877, respectively. The indicators that contributed the most to the restoration effect were leisure and entertainment service value (2016), carbon dioxide absorption service value (2017), eutrophication (2018), and wetland management level (2019), respecially. (3) Compared with the single PSR method, the advantages of this method include; There are fewer evaluation indicators, more accurate results (excluding the interference of some highly variable indicators) and more sensitive to environmental changes. The current study proposed a novel method that may provide additional accurate and refined indicators for urban wetland health assessment.