Optimal Spatiotemporal Scales to Aggregate Satellite Ocean Color Data for Nearshore Reefs and Tropical Coastal Waters: Two Case Studies

Abstract

Remotely sensed ocean color data are useful for monitoring water quality in coastal environments. However, moderate resolution (hundreds of meters to a few kilometers) satellite data are underutilized in these environments because of frequent data gaps from cloud cover and algorithm complexities in shallow waters. Aggregating satellite data over larger space and time scales is a common method to reduce data gaps and generate a more complete time series, but potentially smooths out the small-scale, episodic changes in water quality that can have ecological influences. By comparing aggregated satellite estimates of Kd(490) with related in-water measurements, we can understand the extent to which aggregation methods are viable for filling gaps while being able to characterize ecologically relevant water quality conditions. In this study, we tested a combination of six spatial and seven temporal scales for aggregating data from the VIIRS instrument at several coral reef locations in Maui, Hawai‘i and Puerto Rico and compared these with in situ measurements of Kd(490) and turbidity. In Maui, we found that the median value of a 5-pixels, 7-days spatiotemporal cube of satellite data yielded a robust result capable of differentiating observations across small space and time domains and had the best correlation among spatiotemporal cubes when compared with in situ Kd(490) across 11 nearshore sites (R2 = 0.84). We also found long-term averages (i.e., chronic condition) of VIIRS data using this aggregation method follow a similar spatial pattern to onshore turbidity measurements along the Maui coast over a three-year period. In Puerto Rico, we found that the median of a 13-pixels, 13-days spatiotemporal cube of satellite data yielded the best overall result with an R2 = 0.54 when compared with in situ Kd(490) measurements for one nearshore site with measurement dates spanning 2016–2019. As spatiotemporal cubes of different dimensions yielded optimum results in the two locations, we recommend local analysis of spatial and temporal optima when applying this technique elsewhere. The use of satellite data and in situ water quality measurements provide complementary information, each enhancing understanding of the issues affecting coastal ecosystems, including coral reefs, and the success of management efforts.

Multi‐Factor Coral Disease Risk: A new product for early warning and management

“…We foresee a similar issue for water quality metrics: Although we expect that the seasonal climatology and associated variability metric used in these models are fairly robust in the long term, the current models do not capture acute events caused by intense rainfall and associated runoff, which are known to influence disease (Haapkylä et al, 2011). Although we attempted to measure acute events with ocean color data (procedure described in Geiger et al, 2021), we found that the available data were too sparse to use in the models, with no satellite coverage for ~80% of the corresponding survey data. More importantly, the ocean color data unavailable during events were not random but aggregated during cloudy days; in other words, days that are most likely associated with rain events that can increase disease risk.…”

“…We used the mean (i.e., climatology) and associated variability in Kd(490) to represent seasonal changes because, to date, these values are too highly variable and too infrequently available in the coastal zone to use actual, or even 3-week composite, values. Additional details on the derivation of these metrics and their accuracy can be found in Geiger et al (2021). We include month in the model as a proxy for all other seasonally changing conditions.…”

Multi-Factor Coral Disease Risk Forecasting for Early Warning and Management

“…We foresee a similar issue for water quality metrics: while we expect that the seasonal climatology and associated variability metric used in these models are fairly robust in the long-term, the current models do not capture acute events caused by intense rainfall and associated runoff, which are known to influence disease (Haapkylä et al, 2011). Although we attempted to measure acute events with ocean color data (procedure described in Geiger et al, 2021), we found that the available data were too sparse to use in the models, with no satellite coverage for ∼80% of the corresponding survey data. More importantly, the ocean color data unavailable during events were not random, but aggregated during cloudy days; in other words, days that are most likely associated with rain events that can increase disease risk.…”

“…To characterize chronic water quality conditions in both model development and forecasting, we aggregated the diffuse attenuation coefficient at 490 nm, Kd(490), as a proxy for turbidity from VIIRS data (Kirk, 1994). We calculated long-term Kd(490) median and variability for each reef pixel by overlaying aggregated data from 2012-2020 (i.e., all data available at the time of study) within a 5-pixel buffer (750 m becomes ∼8.25 km resolution) following methods from Geiger et al, 2021 to increase data availability, as nearshore ocean color data are notoriously patchy. These metrics are indicative of spatial differences in water quality across reefs, providing information on locations that have chronically good or poor water quality and those that are exposed to a large range of water quality conditions throughout the year versus those with more consistent conditions.…”

“…We used the mean (i.e., climatology) and associated variability in Kd(490) to represent seasonal changes because, to date, these values are too highly variable and too infrequently available in the coastal zone to use actual, or even three-week composite, values. Additional details on the derivation of these metrics and their accuracy can be found in Geiger et al, 2021. We include month in the model as a proxy for all other seasonally changing conditions.…”