A comparative analysis of Landsat-8 and MODIS data for forecasting cyanobacterial blooms in small-scale reservoirs

Abstract

Cyanobacterial blooms pose significant risks in freshwater ecosystems and human activities. Short-term prediction technologies of such blooms enhance decision-making processes to mitigate their detrimental impacts. Data from earth observation satellites proves invaluable for monitoring cyanobacterial blooms across diverse aquatic environments due to its consistent and systematic surveillance of water surfaces. Smaller water bodies, crucial for water supply in numerous countries, have been largely overlooked in developing predictive models for cyanobacterial blooms using satellite-derived data in previous research. With its low spatial and high temporal resolution, MODIS has been employed to forecast cyanobacterial blooms across vast water bodies, including large lakes and coastal areas. However, to our knowledge, high spatial resolution satellites such as Landsat-8 have not been previously utilized in developing models for small-scale water bodies. In this study, we constructed models to forecast Dolichospermum spp. concentrations with 7-day lead time, a genus of scum-forming cyanobacteria found in small Japanese reservoirs, using variables related to water quality, hydrology, meteorology, and Landsat-8 and MODIS data, integrated through three machine learning algorithms. We established three distinct temporal intervals for satellite-derived land surface temperature (LST) and normalized difference turbidity index (NDTI) to assess their temporal influence on bloom occurrences. The optimal model, employing MODIS data and the XGBoost algorithm, achieved an R² value of 0.84 and a root mean squared error of 0.44 in log₁₀ (cells/L). Satellite data from 23 to 38 days before a Dolichospermum spp. bloom enabled the most accurate models in two reservoirs, with LST generally showing higher relative importance than NDTI. This investigation emphasizes the potential of satellite-derived LST and NDTI as critical predictors in accurately predicting cyanobacterial blooms in small-scale reservoirs.