Optimal sampling strategy of water quality monitoring at high dynamic lakes: A remote sensing and spatial simulated annealing integrated approach

Highlights: • Remote sensing and spatial sampling annealing integrated approach is developed for sampling optimization • The annual total suspended sediment (TSS) estimation errors dropped about 50% compared to traditional methods • Sampling accuracy was affected by the sampling strategy and water quality variations • Remote sensing provides valuable prior-knowledge for efficient sampling design. An efficient and precise spatial sampling design is critical to capture spatial and temporal water quality variations under cost and labor constraints. Therefore, it is practically essential to optimize the sampling locations using limited sampling numbers to obtain the most comprehensive water quality monitoring results considering both the spatial and temporal dynamics. Existing sampling methods were restricted due to lacking pre-information and specific sampling objective function. This paper proposed an optimal sampling strategy using remote sensing (RS) big data and spatial sampling annealing (SSA) integrated approach for sampling design. The proposed method involved spatial-temporal clustering of the total suspended sediment (TSS) using long-term remote sensing data (Terra/Aqua MODIS, 2000–2014), determining the required sampling numbers using geostatistical analysis, and SSA simulation following the objective function of minimization of the spatial-temporal mean estimation error using remote sensing data as references. Taking total suspended sediment (TSS) observations at Poyang Lake, China, as the case study and application region. Results showed that the RS + SSA sampling approach is superior to conventional sampling methods such as systematic, stratified, and expert sampling, concerning spatial and temporal sampling accuracy. TSS estimation errors of the whole lake were reduced by 18.11% and 29.34% on average when compared to systematic and stratified sampling under the same sample size. The annual TSS estimation errors were dropped by approximately 50%. The sampling accuracy was affected by the synthetic effects of sampling strategy (station numbers and spatial distributions) and water quality variations (coefficient of variation, CV). Sampling optimization is more efficient to improve the sampling accuracy than increasing sampling size, which requires more cost and human resources. Remote sensing showed great potential as ideal means to provide spatially contiguous and comprehensive data as prior-knowledge for efficient sampling design. This paper provides solutions and recommendations for evaluating existing monitoring stations in their representation of water quality or optimizing a new sampling network for future implications of more efficient and precise water quality sampling and routine monitoring.