A novel multi-source data fusion method based on Bayesian inference for accurate estimation of chlorophyll-a concentration over eutrophic lakes

A novel multi-source data fusion method based on Bayesian inference (BIF) was proposed in this study to blend the advantages of in-situ observations and remote sensing estimations for obtaining accurate chlorophyll-a (Chla) concentration in Lake Taihu (China). Two error models (additive and multiplicative) were adopted to construct the likelihood function in BIF; the BIF method was also compared with three commonly used data fusion algorithms, including linear and nonlinear regression data fusion (LRF and NLRF) and cumulative distribution function matching data fusion (CDFF). The results showed the multiplicative error model had small normalized residual errors and was a more suitable choice. The BIF method largely outperformed the data fusion algorithms of CDFF, NLRF and LRF, with the largest correlation coefficients and smallest root mean square error. Moreover, the BIF results can capture the high Chla concentrations in the northwest and the low Chla concentrations in the east of Lake Taihu.

Automated summary

A novel multi-source data fusion method based on Bayesian inference was proposed for accurate chlorophyll-a concentration estimation in Lake Taihu. The multiplicative error model was found to be more suitable as compared to the additive model. The Bayesian inference method outperformed other data fusion algorithms, showing the largest correlation coefficients and smallest root mean square error, and successfully captured the spatial distribution of chlorophyll-a concentrations in Lake Taihu.