Controls on Algal Bloom Propagation in the Kuwait Bay Utilizing: An Integrated Remote Sensing and Statistical Approach

Abstract:

Algal blooms have become a major concern over the last decade in Kuwait’s coastal waters where these blooms caused massive fish kill in a number of incidences. The purpose of this study is to accomplish the following: 1) identify the factors controlling algal bloom the development and propagation using Aqua-MODIS satellite data products (from 07/2002 to 07/2012), (2) identify the spatial and temporal variations in Chlorophyll-a (Chl-a) production in relation to the controlling factors; and 3) develop conceptual and predictive models (using in-situ and satellite-based datasets) that account for reported historical blooms and can successfully predict algal bloom proliferation in space and time. To achieve these goals, the following tasks were accomplished: 1) in-situ Chl-a data was correlated with satellite-based (MODIS and MERIS) Chl-a data products (OC3M, GIOP, GSM, and OC4E); 2) Chl-a concentration (from OC3M) were correlated spatially and temporally with potential controlling factors (SST, Turbidity, Euphotic Depth, Precipitation, Photosynthetically Available Radiation, Wind Vectors etc.); 3) the stepwise regression method was applied to identify the most significant controlling factors and to determine their order of importance; and 4) a back-propagation artificial neural network (ANN) was constructed to predict the bloom occurrences in time (first layer process) and space (second layer process). Findings include: 1) the Aqua-MODIS OC3M Chlorophyll-a algorithm correlated best with in-situ measurements (RMSE: 2.42, Mean Bias: 32.2%); (2) maximum OC3M Chl-a concentration was observed throughout the months of August through October (Temp. range: 18.4° to 22.3 °C); 3) the stepwise regression identified SST, secchi disk depth, wind direction, OC3M and wind speed as the most indicative temporal factors (SST: most significant R²: 80.1%) and the OC3M, distance to shore, GSM, SST and GIOP as the most indicative spatial variables; 4) the ANN model showed an excellent prediction performance (area under receiver operating characteristic [ROC] curve: 0.99), and (5) findings are being utilized for the development of an early warning system in the Kuwait Bay.