Efficient estimation of chlorophyll a concentration in artificial upwelling

Artificial upwelling is an effective method to boost the primary productivity in the sea area, which provides a strong support for fishery. The chlorophyll a concentration is a key indicator to evaluate the marine productivity. This paper aims to provide a high-precision model to approximate the chlorophyll a concentration of artificial upwelling. After eliminating the abnormal values from the experimental data, we first used the global linear fitting (GLF) and the global quadratic fitting (GQF) models respectively to obtain reference errors. In order to improve the precision, we introduced the idea of the piecewise fitting to the model. The training set was partitioned into two, four and eight segments to find the optimal number of segments that the model can achieve the best precision. These models calculate the least squares solutions of the error function on each segment. To satisfy the engineering requirement, which is the continuity at boundary points, constraints were added to the boundary points of adjacent segments. Further, the quadratic fitting was proposed and applied on segments where the errors of fitting were relatively large when applying the constrained piecewise linear fitting (CPLF) model, namely the local quadratic fitting (LQF) model. Finally, we partitioned one more variable on the piecewise linear fitting model to examine if the predictive precision can be enhanced. Compared with the work of predecessors, where the GLF, GQF models and intelligent algorithms were applied to this issue, the results of our tested models showed higher precision in predicting chlorophyll a concentration and less complexity in model operation.