Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term...
Transcript of Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term...
![Page 1: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/1.jpg)
Long-Term Trends in Ocean Chlorophyll: Update from a
Bayesian Hierarchical Space-Time Model
Matthew Hammond, Claudie Beaulieu, Stephanie Henson, & Sujit Sahu
![Page 2: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/2.jpg)
National Oceanography Centre
Ocean Colour• Phytoplankton are responsible for ~50% total biospheric production
• Chlorophyll is a phytoplankton pigment that can be remotely sensed
• Thus is listed as an essential climate variable (Bojinski et al., 2014)
• Continuous satellite ocean colour record ~22 years
![Page 3: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/3.jpg)
National Oceanography Centre
Phytoplankton Trend Studies
Boyce & Worm (2015)
• There are conflicting trends
in observational studies
• Arising due to• Differences in technique
• Short record length
• Large interannual variability
![Page 4: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/4.jpg)
National Oceanography Centre
Ocean Colour Trend Studies
Henson et al. (2016)
• Studies suggest average of
~32 years of record required
• Albeit region dependent
• Although this uses a linear
regression technique
![Page 5: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/5.jpg)
National Oceanography Centre
Sea-Ice DetectionLimitations of Previous Studies• Linear regression approach considers
individual grid cells independently
• Incorporating spatial correlation should improve ability to detect trends
• We also use a Bayesian framework so we can quantify uncertainty
![Page 6: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/6.jpg)
National Oceanography Centre
Sea-Ice DetectionSpatio-Temporal Model• 3 level hierarchical model
‐ Data level‐ Process level: Regression coefficients, Temporal and Spatial
Correlation‐ Prior level: Uninformative - as preliminary
• 2 models to show effects of spatial correlation
• ESA OC-CCI v3.1 dataset
• Trend determined in individual Longhurst regions
![Page 7: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/7.jpg)
National Oceanography Centre
Initial Model Comparison
Hammond et al. (2017)
Chl
Dev
iatio
n (m
gm-3
)
![Page 8: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/8.jpg)
National Oceanography Centre
Model Comparison• Spatial correlation
consistently leads to improved accuracy
• Up to 2 %yr-1 difference between the models
• Including spatial correlation leads to realistic assessment of uncertainty
Hammond et al. (2017)
![Page 9: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/9.jpg)
National Oceanography Centre
Model Information as Prior• Biogechemical models can compensate for the shortness of the
observed record
• We use information from the analysis of CMIP5 models‐ 14 Models‐ Up to 6 ensembles per model‐ Period covering 1997-2039 combining RCP 8.5 and historical scenarios
• We analyse trends in each model and ensemble to get a multi-model mean and variance
• This is then incorporated into our model as an informative Bayesian prior
![Page 10: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/10.jpg)
National Oceanography Centre
Bayesian Inference
𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷 ∝ 𝑳𝑳𝑷𝑷𝑳𝑳𝑷𝑷𝑳𝑳𝑷𝑷𝑳𝑳𝑷𝑷𝑷𝑷𝑳𝑳 × 𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷
Vague Prior
![Page 11: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/11.jpg)
National Oceanography Centre
Bayesian Inference
𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷 ∝ 𝑳𝑳𝑷𝑷𝑳𝑳𝑷𝑷𝑳𝑳𝑷𝑷𝑳𝑳𝑷𝑷𝑷𝑷𝑳𝑳 × 𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷𝑷
Vague Prior Specific Prior
![Page 12: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/12.jpg)
National Oceanography Centre
Impact of Priors
Hammond et al. (In Review)
North Pacific Tropical Gyre Province
![Page 13: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/13.jpg)
National Oceanography Centre
Impact of Priors
• Slight reduction in uncertainty in the majority of regions
• 15/23 regions
• General biasing effect towards zero
• Effect in most cases fairly small• Large range of model projections?• Priors for other parameters?
Hammond et al. (In Review)
![Page 14: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/14.jpg)
National Oceanography Centre
Global Outlook
• Trends generally positive in the Southern Ocean and Atlantic
• Up to 1.5%yr-1
• Global average small net positive trend
• 0.08 %yr-1
• Most uncertainty in trend estimates in the North Atlantic and Kuroshio Current regions
Hammond et al. (In Review)
![Page 15: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/15.jpg)
National Oceanography Centre
Conclusions• We find greatly improved accuracy in estimating chl trends when using spatial
correlation terms
• Using prior information from CMIP5 models helps to constrain trends and their uncertainties, particularly important while the observational record is still short
• The statistical model used here provides a robust framework for incorporating all available information and can be applied to improve understanding of global change in a range of key climate variables
![Page 16: Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian … · 2020. 5. 6. · Long-Term Trends in Ocean Chlorophyll: Update from a Bayesian Hierarchical Space-Time Model. Matthew](https://reader033.fdocuments.in/reader033/viewer/2022051815/603ed4efd7bcfb246f117cb4/html5/thumbnails/16.jpg)
Making Sense of Changing Seas
Thank you
Hammond, Beaulieu, Sahu, and Henson. (2017) - “Assessing Trends and Uncertainties in Satellite-Era Ocean Chlorophyll Using Space-Time Modeling”. Global Biogeochemical Cycles.
Hammond, Beaulieu, Henson, and Sahu. (2018) – “Assessing the presence of discontinuities in the ocean color satellite record and their effects on chlorophyll trends and their uncertainties”. Geophysical Research Letters.
Hammond, Beaulieu, Henson, and Sahu. (In review) - “Updated Global Ocean Color Trends and Uncertainties Constrained Using Biogeochemical Models”.