Prediction of algal bloom occurrence based on the naive Bayesian model considering satellite image pixel differences
Bloom occurrence probability prediction is a critical issue for freshwater resource management and protection. As the mechanism of algal blooms is not understood, the construction of prediction model mainly depends on statistical data. Therefore, knowledge on prior bloom occurrence derived from stat...
Guardado en:
| Autores principales: | , , , , , , , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
Elsevier
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/6558812c97964613aa85dee33e1d1e7e |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Bloom occurrence probability prediction is a critical issue for freshwater resource management and protection. As the mechanism of algal blooms is not understood, the construction of prediction model mainly depends on statistical data. Therefore, knowledge on prior bloom occurrence derived from statistical data plays a significant role in establishing a prediction model. In this study, a naïve Bayesian model incorporating prior information was constructed to predict algal bloom occurrence probabilities 1–7 days in advance under different weather conditions in Dianchi Lake. The proposed model utilised the following data from the MODIS images, the floating algae index (FAI) for the previous 7 days and five meteorological variables (mean wind speed, air pressure, relative humidity on the prediction day, accumulated sunshine hours in the previous 3 days, and accumulated air temperature in the previous 7 days). The prior probabilities of each pixel were calculated on a monthly timescale to highlight the bloom’s temporal-spatial differences, and the 1–7 day posterior probabilities were calculated by combining the prior and conditional probabilities. The predictive effect was tested by the area under the receiver-operating characteristic curve (AUC), and the results showed the number of pixels with estimation predictions classified as ‘Good+’ and ‘Excellent’ were 91.3%, 91.4%, 82.7%, 83.3%, 89.0%, 86.6%, and 90.0% for the predictions on days 1–7, respectively. Additionally, the independent validation datasets showed that the percentage of correctly classified instances (CCI%) of approximately 90% of the pixels were greater than 50%. The proposed algal bloom prediction model based on the naïve Bayesian method has pixel-level prediction abilities, is applicable to other inland water systems, and can provide a reference for water environment risk forecasting and management. |
|---|