A Hidden Markov Model based unscented Kalman Filtering framework for ecosystem health prediction: A case study in Shanghai-Hangzhou Bay Urban Agglomeration

Abstract

Urban agglomeration is one maturely spatial organized form in the urbanization development process, and often presents in the area with prominent ecological and environmental problems. As a widely used scientific indicator, ecosystem health aims at measuring the ecosystem variation along with the development of urban agglomeration. Assessing the ecosystem health of urban agglomeration in the past stages is the foundation of sustainable development of urban agglomeration. Based on a hidden Markov Model and unscented Kalman filtering method, this research monitor and predict the ecosystem health of Shanghai-Hangzhou Bay Urban Agglomeration (SHBUA). More specifically, we propose hidden Markov model (HMM) decoding scheme and HMM-based Kalman filter framework to simulate and predict the ecosystem health status in SHBUA. The feasibility and simulation accuracy of two prediction cases is verified in ecosystem health assessment of urban agglomerations. Unscented Kalman filter (UKF) is used to refine results of the prediction procedure. We construct the HMM-UKF scheme and HMM decoding scheme using Future Land Use Simulation (FLUS)-based land use data. Analysis of results of two prediction cases indicates the spatial heterogeneity of the UKF correction effect. We construct sequences extension strategies to meet the requirements of the UKF algorithm. The result shows that sequences extension strategies optimize the prediction mechanism of the HMM-UKF scheme. Theoretically, UKF can derive interpretable ecological health prediction results based on both measurement and decoding values of EHI. The application of UKF improves the prediction accuracy of HMM from 74.60% to 81.27%. This research may provide an available framework for quantitative prediction of ecosystem health based on short-term time series, using the sequence expansion strategy as an option to improve the performance.