Identifying Influential Variables for Chlorophyll-a Concentration using Permutation Variable Importance Measure

Chlorophyll-a (Chl-a), a critical indicator of aquatic ecosystem health, is influenced by hydrological, meteorological, and water quality (WQ) parameters. This study employs the Random Forest (RF) machine learning algorithm, integrating remote sensing and in-situ data, to develop a robust model for estimating Chl-a concentrations in three distinct regions of Gorgan Bay, Iran. The Permutation Variable Importance (PVI) index ranks the significance of WQ, hydrometric, and meteorological parameters. The RF model demonstrates strong performance, achieving an R² of ۰.۸۸ and an RMSE of ۱.۵۱ mg/m³, accurately capturing complex relationships between environmental variables and Chl-a. Particulate Organic Carbon (POC), pH, and river discharge emerge as the most influential variables. Limitations, including the exclusion of nutrient parameters (e.g., nitrogen, phosphorus) and incomplete remote sensing data due to cloud cover, may affect model generalizability. These findings support targeted WQ monitoring and management strategies for Gorgan Bay's ecosystem preservation.

Keywords:

Chlorophyll-a (Chl-a) , Random Forest (RF) , Permutation Variable Importance (PVI) , Sensitivity Analysis , Gorgan Bay (GB)