Sampling biases in MODIS and SeaWiFS ocean chlorophyll data

Although modern ocean color sensors, such as MODIS and SeaWiFS, are often considered global missions, in reality it takes many days, even months, to sample the ocean surface enough to provide complete global coverage. The irregular temporal sampling of ocean color sensors can produce biases in monthly and annual mean chlorophyll estimates. We quantified the biases due to sampling using data assimilation to create a “truth field”, which we then sub-sampled using the observational patterns of MODIS and SeaWiFS. Monthly and annual mean chlorophyll estimates from these sub-sampled, incomplete daily fields were constructed and compared to monthly and annual means from the complete daily fields of the assimilation model, at a spatial resolution of 1.25° longitude by 0.67° latitude. The results showed that global annual mean biases were positive, reaching nearly 8% (MODIS) and N5% (SeaWiFS). For perspective the maximum interannual variability in the SeaWiFS chlorophyll record was about 3%. Annual mean sampling biases were low (b3%) in the mid-latitudes (between −40° and 40°). Low interannual variability in the global annual mean sampling biases suggested that global scale trend analyses were valid. High latitude biases were much higher than the global annual means, up to 20% as a basin annual mean, and over 80% in some months. This was the result of the high solar zenith angle exclusion in the processing algorithms. Only data where the solar angle is b75° are permitted, in contrast to the assimilation which samples regularly over the entire area and month. High solar zenith angles do not facilitate phytoplankton photosynthesis and low chlorophyll concentrations occurring here are missed by the data sets. Ocean color sensors selectively sample in locations and times of favorable phytoplankton growth, producing overestimates of chlorophyll. The biases derived from lack of sampling in the high latitudes varied monthly, leading to artifacts in the apparent seasonal cycle from ocean color sensors. A false secondary peak in chlorophyll occurred in May–August, which resulted from lack of sampling in the Antarctic. Persistent clouds, characteristic in the North Pacific, also produced overestimates, again by selectively sampling only the high growth periods. In contrast, areas characterized by thick aerosols showed chlorophyll underestimates to nearly −30% in basin monthly means. This was the result of selective sampling in lower aerosol thickness periods, which corresponded with lower phytoplankton growth periods. A combination of MODIS and SeaWiFS sampling was most effective at reducing mid-latitude biases due to inter-orbit gaps, sun glint, and sensor tilt changes. But these biases were low using a single sensor, suggesting multiple sensors had little effect in reducing global and regional monthly and annual mean biases. Ocean color data are an invaluable source of information about global biological processes. However, these results suggest that sampling errors need to be considered in applications involving global and regional mean chlorophyll biomasses as well as seasonal variability and regional trend analysis. © 2007 Elsevier Inc. All rights reserved.