| Popis: |
A feature-based verification method, commonly used for atmospheric model applications, has been applied to Chlorophyll-a (Chl-a) concentration forecasts from the Met Office Atlantic Margin Model at 7 km resolution (AMM7) North West European Shelf Seas model, and compared against gridded satellite observations of Chl-a concentration from the Copernicus Marine Environmental Monitoring Service (CMEMS) catalogue. A significant concentration bias was found between the model and observations. Two variants of quantile mapping were used to mitigate against the impact of this bias on feature identification (determined by threshold exceedance). Forecast and observed Chl-a objects for the 2019 bloom season (March 1 to 31 July), were analysed, firstly in space only, and secondly as space-time objects, incorporating concepts of onset, duration and demise. It was found that forecast objects tend to be too large spatially, with lower object numbers produced by the forecasts compared to those observed. Based on an analysis of the space-time objects the onset of Chl-a blooming episodes at the start of the season is almost a month too late in the forecasts, whilst several forecast blooms did not materialise in the observations. Whilst the model does produce blooms in the right places, they may not be at the right time. There was very little variation in forecasts and results as a function of lead time. A pre-operational AMM7 analysis, which assimilates Chl-a concentrations was also assessed, and found to behave more like the observations, suggesting that forecasts driven from these analyses could improve both timing errors and the bias. |
