Studying spatio-temporal changes in phytoplankton by means of remote sensing

The ocean is teeming with microscopic organisms called phytoplankton. Phytoplankton comprises two main groups: photosynthetic cyanobacteria and the single-celled algae that drift in the sunlit top layers of oceans. They provide food, directly or indirectly for virtually every other marine creature. They emit much of the oxygen that permeates our atmosphere and their fossilized remains, buried and compressed by geological forces, are transformed into oil. In addition, they play a huge role in the cycling of carbon dioxide from the atmosphere to the biosphere and back, cycling that helps to control Earth’s climate [1].air jordan 1

Oakley sunglasses Promo Code

How do we study phytoplankton from space you may ask. Well, we take advantage of the fact that oxygen-producing photosynthesis only occurs in organisms that have a pigment called chlorophyll-a. This pigment enables the phytoplankton to absorb blue light, which would otherwise be scattered by the sea water. The more phytoplankton there are in an area of the ocean, the more chlorophyll-a there is and the darker the area appears from space. Therefore, through calibration between remotely sensed reflectance and in situ measures of chlorophyll concentration, we are able to estimate chlorophyll-a concentration from remote sensors and use it as a proxy for phytoplankton biomass. A rapid increase in phytoplankton biomass, in terms of number of cells per volume unit or chlorophyll concentration, is commonly known as “phytoplankton bloom”. This blooms may occupy large extensions and are normally seen from space.

nike air max 1 premium

Phytoplankton bloom in Argentinian sea as seen from MODIS/Aqua.
Phytoplankton bloom in Argentinian sea as seen from MODIS/Aqua.

The Argentinian continental shelf is one of the richest areas of the world oceans both in terms of phytoplankton biomass but also for the great abundance of economically important species. However, it is also the most poorly studied ecosystem. There was then, the interest and the need to study the spatio-temporal variations of satellite chlorophyll and phytoplankton blooms in the continental shelf and shelf break of the Argentinian patagonic region. There was also an especial interest in this area because the Argentinian Spatial Agency (CONAE) along with Brazilian Spatial Agency (AEB), are developing a new mission called SABIA-Mar that will provide ocean color data with an especial focus in this region. Therefore, we used 11-years of MODIS/Aqua L3 products of 4 km spatial resolution and 8 days temporal resolution.

All the raster processing was done with Free and Open Source Software, especially GRASS GIS and R, and taking advantage of the new temporal modules in GRASS GIS. Here, I will show some examples of commands used for the study and the results obtained with them.

After downloading all the raster data (from NASA ocean color site) and importing them into GRASS (see wiki for more details), we first analyzed the availability of valid data over the study region for the whole study period and characterized its monthly and annual variability.

oakley sunglasses usa

## total amount of valid data ##
# valid data count
t.rast.series input=cla output=count_2003_2013 method=count
# percentage of valid data
r.mapcalc expression="perc_valid_data_2003_2013=(count_2003_2013*100.0)/506.0"

ray ban reading sunglasses

total_perc_vd
Total percentage of valid data for the period 2003-2013.

nike air max release dates

## monthly amount of valid data ##
# monthly counts of valid data
t.rast.aggregate input=cla output=cla_monthly_count base=cla_monthly_count granularity="1 months" method=count sampling=start
# monthly sums of valid data
for month in "0 jan" "1 feb" "2 mar" "3 apr" "4 may" "5 jun" "6 jul" "7 aug" "8 sep" "9 oct" "10 nov" "11 dec" ; do
 set -- $month ; echo $1 $2
 t.rast.series input=cla_monthly_count method=sum where="start_time=datetime(start_time, 'start of year', '"${1}" month')" output="${2}"_count_sum
done
# monthly percentages of valid data (number of maps per month for some months changes in leap years)
# we need to count how many maps for each month we have
for i in 01 02 03 04 05 06 07 08 09 10 11 12 ; do
 t.rast.list -s input=cla columns=name where="strftime('%m', start_time)='"${i}"'" | wc -l
done
# then, just simple r.mapcalc operations to get percentages

girls oakley sunglasses

Monthly percentage of valid data.
Monthly percentage of valid data.

Then, we studied the spatio-temporal variability of satellite chlorophyll-a concentration and mapped descriptive statistics such as mean and standard deviation for the whole time series (2003-2013) and inter-annualy. We also obtained seasonal and monthly climatologies for those statistics and analyzed anomalies from the mean.

authentic nike air jordan

# annual variations
for method in average stddev ; do
 t.rast.aggregate input=cla output=cla_yearly_${method} base=cla_yearly_${method} granularity="1 years" method=${method} sampling=start
done

nike air max running

fig11years_average
Mean annual concentration of chlorophyll-a (MODIS-Aqua L3).

# seasonal climatology
for i in "01 02 03" "04 05 06" "07 08 09" "10 11 12" ; do
set -- $i ; echo $1 $2 $3
 for m in average stddev ; do
 t.rast.series input=cla method=${m} where="strftime('%m', start_time)='"${1}"' or strftime('%m', start_time)='"${2}"' or strftime('%m', start_time)='"${3}"'" output=${m}_${1}
 done
done

# monthly climatology
for i in 01 02 03 04 05 06 07 08 09 10 11 12 ; do
 for m in average stddev ; do
 t.rast.series input=cla method=${m} where="strftime('%m', start_time)='${i}'" output=${m}_${i}
 done
done
Fig12months_average_eng
Mean monthly concentration of chlorophyll-a (MODIS- Aqua L3).

In general, the results showed that spatially, chlorophyll-a concentration changes during the cycle of a year and among years, both in attained values and distribution and extension of high concentration areas. However, a certain constancy was observed in location and timing of bloom (and high concentration) occurrence. For example, the largest extension of high chlorophyll-a values is usually observed in spring, covering the shelf break area, coastal tidal fronts and midshelf until approximately 52° S, with a latitudinal north-south progression direction. Temporally, spatial aggregates of the whole study area showed that chlorophyll-a presents a more or less marked annual cycle with high values in spring that persist during summer and decrease towards autumn and winter, as seen in monthly and seasonal climatologies. It was also evident, both in temporal and spatial analysis, a second (normally less important, but not always) bloom in autumn.ray ban sunglasses australia

Annual and monthly anomalies also reflected a high level of variation in the continental shelf, alternating with years of great area of positive anomalies and years with predominance of values below the mean. The largest areas of positive anomalies were observed in 2003, 2010 and 2011. These anomalies may be explained by wind anomalies changing front’s positions and anomalies in rivers discharges as a consequence of higher precipitations because of global phenomena such as El Niño [2-3]. The mechanism by which these phenomena operate is still under study. Some authors suggest that global effects of ENSO (El Niño Southern Oscillation) upon winds (as a consequence of changes in sea surface temperature) may affect phytoplankton distribution [3]. Some others sustain that what changes is phytoplankton community composition and/or bloom starting date [4-5]. In Southern Southamerica, ENSO has strong effects on rainfall, especially on spring rainfall. This increases the discharges by great rivers such as Rio de la Plata (RDP). In fact, high concentrations of chlorophyll-a were detected close to RDP discharges during El Niño events, probably because of the larger extension of the plume and the higher nutrient supply.

# yearly anomalies
t.rast.mapcalc in=cla_yearly_average out=cla_yearly_average_anomaly basename=cla_yearly_average_anomaly expression="cla_yearly_average-average_cla"
mean_fig11years_anomaly
Annual anomalies from the mean.
# monthly anomalies
t.rast.aggregate input=cla output=cla_monthly_average base=cla_monthly_average granularity="1 months" method=average sampling=contains
for i in 01 02 03 04 05 06 07 08 09 10 11 12 ; do
 for map in `t.rast.list -s input=cla_monthly_average columns=name where="strftime('%m', start_time)='"${i}"'"` ; do
 r.mapcalc expression="anomaly_${map}=${map}-average_${i}"
 done
done
monthly_anomaly_2011_eng
Monthly anomalies from the mean for 2011.

To study certain phenologic indexes, we needed complete (without gaps) series. Therefore, we compared the two most used methods of gap-filling and reconstruction, i.e.: DINEOF (Data Interpolating Empirical Orthogonal Functions) and HANTS (Harmonic Analysis of Time Series). DINEOF is available at GHER site, but we used the version available in the package sinkr for R. HANTS on the other hand is available as an add-on for GRASS GIS (r.hants). Below, some examples of HANTS use:

r.hants -lh input=`g.list rast pat=*_CHL_*arg sep=comma` nf=5 fet=0.1 dod=11 range=0.01,65.0 base_period=46 suffix=_hants1 amplitude=amp_hants1 phase=pha_hants1
r.hants -lh input=`g.list rast pat=*_CHL_*arg sep=comma` nf=6 fet=0.1 dod=11 range=0.01,65.0 base_period=46 suffix=_hants2 amplitude=amp_hants2 phase=pha_hants2

One of HANTS benefits is that it allows to know the dominant frequency making use of the amplitude maps generated as outputs, and hence know how many cycles we have in one year, or whatever the base_period is. This is achieved as follows:

# extract dominant frequencies
for i in 1 2 ; do
 r.series input=`g.list rast pat=amp_hants* sep=,` output=dom_freq_hants${i} method=max_raster
done

For DINEOF, we had to export our strds, import it into R, create a spatio-temporal matrix which is the input for DINEOF, run the command, recreate raster maps from output spatio-temporal matrix, and import them back into GRASS (more details on an upcoming post). Next, we described and analyzed the spatial variability of phenological indexes, such as date of maximum concentration and bloom starting date estimated by 2 different methods: the maximum rate of change and a threshold-based method.

# Date of maximum concentration of chlorophyll-a
t.rast.mapcalc -n inputs=cla output=date_max_cla expression="if(cla == max_cla,start_doy(),null())" basename=date_max_cla
t.rast.series input=date_max_cla method=maximum output=series_date_max_cla
t.rast.aggregate input=cla granularity="1 year" method=max_raster output=yearly_max_index basename=yearly_max_index

These maps were then reclassified to DOY, averaged and reclassified again to month. The next map shows the mean date of occurrence of maximum concentration.

bloom_md_dineof_month_eng
Month of maximum chlorophyll concentration.
# Bloom starting date - Method: Maximum rate of change
# 1. get the slope
t.rast.mapcalc2 expression="slope_cla = (cla[1]-cla[0])/8.0" basename=slope_cla
# 2. get the maximum slope per year
t.rast.aggregate input=slope_cla granularity="1 year" method=max_raster output=max_slope_index basename=max_slope_index

The same as before, these maps were reclassified to DOY, averaged and reclassified again to month, to get the mean bloom staring date based on maximum rate of change.

bloom_sd_dineof_month_eng
Month of bloom start based on maximum rate of change.

The method based on a threshold was programmed in R following [6] and using 5% above the median as threshold for bloom determination. We exported our strds (spatio-temporal raster data set) to R, run the script there, and imported maps back into GRASS GIS.

Finally, we estimated and described statistical indexes such as bloom area and bloom occurrence frequency. We first reclassified maps according to different thresholds. We considered 3 different thresholds: 5, 10 and 15% above the median of the series in each pixel.

# bloom occurrence
for threshold in 005 010 015 ; do
 t.rast.mapcalc -n input=cla output=cla_${threshold}_higher_median_class expression="if(cla > ${threshold}_higher_median, 1, null())" basename=cla_${threshold}_higher_median_class
done

To get bloom area according to each threshold, we just obtained basic statistics for each series and then processed those reports in R.

for threshold in 005 010 015 ; do
  t.rast.univar cla_${threshold}_higher_median_class > stats_${threshold}
done

With the time series of classified maps (bloom/no bloom) according to the different thresholds, we estimated total and yearly frequency of bloom occurrence and percentage of occurrence (as a proxy for bloom duration).

# yearly frequency and percentage of bloom occurrence
# counts
for threshold in 005 010 015 ; do
 t.rast.aggregate input=cla_${threshold}_higher_median_class method=count granularity="1 year" output=${threshold}_year_bloom_freq basename=${threshold}_yearly_bloom_freq
done
# percentages
for threshold in 005 010 015 ; do
 for i in `seq 1 11` ; do
  r.mapcalc expression="${threshold}_year_bloom_perc_${i}=(${threshold}_year_bloom_freq_${i}*100)/vd_yearly_count_${i}"
 done
done

The following figure shows the inter-annual variation in percentage of bloom occurrence. Years that showed positive anomalies in previous analysis, also showed a higher percentage of bloom occurrence, i.e.: they showed bloom level concentrations for a “longer” period.

fig11years_005_yearly_bloom_perc_dineof
Annual variations in percentage of bloom occurrence.

Phenomena such as El Niño imply a change in sea surface temperature, and multiple lines of evidence suggest that changes in phytoplankton biomass and ocean productivity are related to ocean warming product of global warming [7]. At least two mechanisms are driving this trend: a physical effect upon vertical stratification as a consequence of warming of ocean higher layers that affects phytoplankton by limiting nutrient supply, and a direct effect of warming on phytoplankton metabolic rates. It is well-known that phytoplankton changes and changes in their phenology (bloom starting date, bloom duration, etc.) may have deep effects upon biogeochemical cycles, climatic patterns, fisheries and the general structure and functioning of marine ecosystems. These global effects are a strong motivation to study phytoplankton changes under global warming [1,7].

As a very general conclusion, we may say that the spatial variability in each moment, or for the aggregation of a certain period, is likely to be dependent on environmental differences among diverse areas and the particular dynamics associated to geographic position. The temporal variability, on the one hand, can be related to seasonal regular cycles in lightning conditions, nutrient flux, vertical stratification, among others. On the other hand, the inter-annual variability observed in chlorophyll-a concentration and phenological indexes considered might be related to external or extrinsic forces associated with climate changes. This study intended to be a baseline on the spatio-temporal patterns of variation of chlorophyll-a concentration in the continental shelf and shelf break of the Argentinian patagonic region, though it is acknowledged that ocean dynamics is too complex to be addressed with a unique index. Nevertheless, it is a first approach to the problem and set the basis to continue researching more effective methods to study algal blooms (and their variability) in the Argentinian sea, with the final goal of including these kind of products in models that allow to predict the occurrence of harmful algal blooms, the dynamics of marine system (under extractive pressure) and the effects of global changes over climatic and biogeochemical cycles. Finally, this work also intended to contribute with science data to the development and planning of CONAE-AEB SABIA-Mar mission, which will provide high-resolution ocean color data over Argentinian and Brazilian coastal zones and continental shelf.

This work is part of my MS thesis (In Spanish) defended on February 25th, 2015. You can download it here.

[1] Falkowski, P. (2012). Ocean science: the power of plankton. Nature, 483:S17–S20.

[2] He, X., Bai, Y., Pan, D., Chen, C.-T. A., Cheng, Q., Wang, D., & Gong, F. (2013). Satellite views of the seasonal and interannual variability of phytoplankton blooms in the eastern china seas over the past 14 yr (1998-2011). Biogeosciences, 10:4721–4739.

[3] Machado, I., Barreiro, M., & Calliari, D. (2013). Variability of chlorophyll-a in the south-western atlantic from satellite images: Seasonal cycle and enso influences. Continental Shelf Research, 53:102–109.

[4] D’Ortenzio, F., Antoine, D., Martinez, E., & Ribera d’Alcala, M. (2012). Phenological changes of oceanic phytoplankton in the 1980s and 2000s as revealed by remotely sensed ocean color observations: OCEANIC PHYTOPLANKTON PHENOLOGY CHANGES. Global Biogeochemical Cycles, 26:GB4003.

[5] Solari, L. C., Gabellone, N. A., Claps, M. C., Casco, M. A., Quaini, K. P., & Neschuk, N. C. (2014). Phytoplankton chlorophyte structure as related to ENSO events in a saline lowland river (Salado River, Buenos Aires, Argentina). Ecology and Evolution, 4:918–932.

[6] Brody, S. R., Lozier, M. S., & Dunne, J. P. (2013). A comparison of methods to determine phytoplankton bloom initiation: METHODS TO DETERMINE BLOOM INITIATION. Journal of Geophysical Research: Oceans, 118:2345–2357.

[7] Lewandowska, A. M., Boyce, D. G., Hofmann, M., Matthiessen, B., Sommer, U., & Worm, B. (2014). Effects of sea surface warming on marine plankton. Ecology Letters, 17:614–623.

Leave a Reply

Your email address will not be published. Required fields are marked *