Random Forest Ensemble for River Turbidity Measurement From Space Remote Sensing Data

River turbidity, serving as an important evaluation index for monitoring water contamination and guiding pollution control, is mainly measured based on the data gathered from contacting turbidity sensors or contactless space satellites. Nevertheless, the prevalence of those abovementioned two measurements is strongly limited due to the disadvantages of low-density spatial distribution of sensor data and extremely high price of satellite data. To solve such difficulty, depending on the Google earth engine (GEE) that freely supplies hyperspectral remote sensing data, in this article, we propose a novel river turbidity measurement model based on random forest ensemble. First, by fully taking advantage of each spectral information and their tuned spectral information, a newly proposed full combination subspace is deployed to generate all the possible base random forests. Second, we introduce a novel error-minimization-based pruning algorithm to circularly delete poor base random forests in accordance with the dynamic threshold. Finally, a weighted average method solved by regularized linear regression is used to aggregate the entire remainder base random forests that are preserved after pruning, thereby yielding the final measurement result of river turbidity. Experiments corroborate the superiority of our proposed model over state-of-the-art competitors and its simplified counterparts.