Ecosystem services assessment and sensitivity analysis based on ANN model and spatial data: A case study in Miaodao Archipelago

Abstract

Ecosystem services (ESs) assessment is an important basis for the island protection and utilization plan development. The application of artificial neural network (ANN) to ESs assessment is a new attempt. Herein, an ANN model for the assessment of 5 ESs including carbon sequestration (CS), habitat quality (HQ), nutrient retention (NR), sediment retention (SR), and water yield (WY) of Miaodao Archipelago is established by taking the advantages of InVEST model and the powerful spatial analysis function of the geographic information system (GIS), as well as the self-learning and self-adaptive characteristics, and prediction function of ANN. The independent sample test shows that the correlation coefficient between the model prediction value and InVEST model simulation value is 0.88 (P < 0.001), and the average absolute error of the simulation is 10.33%. The 5 ESs of the Miaodao Archipelago in 2019 was then simulated using the trained model. The comparison with the corresponding data in 2010 suggests that CS, SR and HQ are in decline, while WY is in upward. The impacts of land use on ESs are then analyzed with the ecosystem service change index (ESCI) in different land use scenarios. An ANN model based quantitative method for evaluating the sensitivities of ecosystem service (ES) to typical environmental factors is proposed. The modeling results suggest that the ANN based model can accurately quantify the importance of different environmental factors to the island ESs, and give the comprehensive impact of multiple factors on the ESs. Therefore, the ANN model combined spatial data analysis provides an important means for ESs assessment and sensitivity analysis.

Keywords

• Ecosystem service
• Artificial neural network
• InVEST model
• Parameter sensitivity