Ecological Nowcasting in Chesapeake Bay

1. Ecological Nowcasting in Chesapeake Bay Christopher Brown NOAA Satellite Climate Studies Branch CICS - ESSIC University of Maryland, College Park

2. Motivation for Study • Detect and predict distribution pattern of organisms that affect society, both beneficial and harmful • Few existing methods work well and in near-real time • satellite remote sensing using spectral signatures • mechanistic modeling

3. Habitat Model Hybrid Statistical – Mechanistic Approach • Develop multi-variate empirical habitat models • Drive habitat models using real-time data acquired from a variety of sources

4. Hybrid Statistical – Mechanistic Approach • Based on concept of niche • Identify the geographic locations where ambient conditions coincide with the preferred habitat of target organism • Old technique employed in new way • GAP analysis: retrospective analysis • Nowcasting: near-real time

5. Ecological Nowcasting of Two Species in Chesapeake Bay • Dinoflagellate Karlodinium micrum • Sea Nettle, Chrysaora quinquecirrha

6. Karlodinium micrum • A common estuarine dinoflagellate found along the U.S. East Coast • Seasonally abundant in Chesapeake Bay • Contributed to several fish kills in Chesapeake Bay • Significant blooms confined to a relatively narrow range of salinity and temperature Photomicrograph of the dinoflagellate Karlodinium micrum.

7. SST Habitat Model Salinity K. micrum Nowcasting Procedure • Estimate current surface salinity and temperature fields • Georeference salinity and SST fields • Apply habitat model • Generate image illustrating the relative abundance of K. micrum Relative Abundance of K. micrum

8. Habitat Model • Neural Network (NN) employs sea surface temperature, salinity and month to predict the relative abundance of K. micrum at low, medium and high or “bloom” concentrations • NN trained with samples (n = 151) of in-situK. micrum abundance and various environmental variables • A test data set (n = 81) was extracted from the available data to assess the model’s performance

9. X1 X2 X3 Xn PE1 f (W h X + b) wij aPE I N P U T S X Classify a= 1 0 -1 PE out f (W h X + b) aPE PE2 f (W h X + b) h h h aPE = foutput(Woutputhfhidden (Whiddenh X + bhidden) + boutput) h h h PEm f (W h X + b) Xi * wij Schematic Representation ofNeural Network Input Layer Hidden Layer Output Layer

10. Issues and Advantages of Neural Networks • Issues • “Black Box” • Advantages & Uses • Useful for representing and processing inexact and sparse data and for performing approximate reasoning over uncertain knowledge and ill-defined problems • Useful in discerning patterns and relationships • No a-priori distribution assumed

11. K. micrum Neural Network Performance

12. 35 30 25 20 15 10 5 0 Surface Salinity • Generated using hydrodynamic model developed for the Chesapeake Bay • Model forced using near-real time input • Model attributes: • Horizontal Resolution: 1-5 kilometers • Vertical Resolution: 1.52 meters • Error: 2 - 3 ppt Model generated surface salinity in Chesapeake Bay for April 20, 2005

13. 35 30 25 20 15 10 5 0 Sea-Surface Temperature Two Sources: • Generated by hydrodynamic model • Error: 2 - 3 °C • Derived from NOAA AVHRR satellite imagery • Resolution: 1 km • Weekly composite • Bias: 0.5 °C; STD: 1.0°C Sea-surface Temperature (ºC) Model generated sea-surface temperature in Chesapeake Bay for April 20, 2005

14. Nowcast vs. In-Situ Comparison May 27, 2004 - Nowcast May 23-26, 2004 - In-situ 0-10 cells/ml 10-2000 cells/ml >2000 cells/ml

15. ephyra juvenile medusa (adult) egg strobila scyphistoma larva polyp Life Cycle of Chrysaora From: T.L. Bryant and J.R. Pennock (eds). 1988. The Delaware Estuary: Rediscovering a Forgotten Resource. University of Delaware Sea Grant College Program. Newark, DE. Introduction: Sea Nettles • Chrysaora ephyra and medusa seasonally populate Chesapeake Bay • Chrysaora is biologically important and impacts recreational activities • Knowing the distribution of Chrysaora would provide valuable information

16. SST Likelihood of Chrysaora Habitat Model Salinity Sea Nettle Nowcasting Procedure • Estimate current surface salinity and temperature fields • Georeference salinity and SST fields • Apply habitat model • Generate image illustrating the likelihood of encounter of Chrysaora

17. Sea Nettle Habitat Model • Models developed to predict: • Probability of encountering Chrysaora • Density of Chrysaora • Analyzed relationship between Chrysaora, salinity and sea-surface temperature • Samples collected in surface waters (0 –10 m) of Chesapeake Bay (n = 1064) • 2/3 model training • 1/3 model testing

18. Sea Nettle Habitat Nettle medusa occupy narrow temperature (26-31 °C) and salinity (10-16 PSU) range. Salinity optimum = 13.5 PSU.

19. Probability of Encountering Sea Nettles • Combination of salinity and SST is a good predictor of Chrysaora presence • If SST < 34°C: • p = elogit / (elogit + 1), • where, • logit = -8.120 + (0.351*SST) - (0.572* |SAL - 13.5|) • Hosmer-Lemeshow Goodness of Fit P = 0.493

21. Preliminary In-Situ Verification • Model prediction performed well on bay-wide scale • Fail to predict lateral variation in mainstem • Additional comparison with in-situ measurements underway

22. Nowcast WWW Sites Sea Nettle and K. micrum nowcasts are generated daily and are available on the World Wide Web. http://coastwatch.noaa.gov/seanettles http://coastwatch.noaa.gov/cbay_hab/index.html

23. Future Directions and Work • Continue nowcast validation and refine habitat models of Chrysaora and Karlodinium • Develop habitat models for additional HAB species in Chesapeake Bay • Incorporate additional environmental variables into habitat models and nowcast system to enhance HAB prediction capability • Generate historical distribution patterns of occurrence and relative abundance from retrospective salinity and temperature to document interannual variability

24. Regional Ecosystem Modeling • Objective: Develop a fully integrated, bio-physical model of Chesapeake Bay and its watershed that assimilates in-situ and satellite-derived data. • Purpose: • Near-Real Time Applications: Nowcasting and forecasting of marine organisms, ocean health, and coastal conditions • Climate Research: Estimating effect of climate change on the health of coastal marine ecosystems • Partners: NOAA, CICS-ESSIC, other UMD departments, Meteorology, and programs, e.g. UMCES. SeaWiFS True-Color Image of Mid-Atlantic Region from April 12, 1998. Image provided by the SeaWiFS Project, NASA/Goddard Space Flight Center and ORBIMAGE

25. Thank You!

26. Probability of Encountering C. quinquecirrha Nowcasting Sea Nettles Likelihood of Encountering C. quinquecirrha On July 29, 1999

27. Interannual Variability Probability of Encountering C. quinquecirrha July 29, 1999 July 25, 1996 Likelihood of Encountering C. quinquecirrha in July 1996 and 1999

28. Future Directions and Work • Continue validation and refine habitat model of K. micrum with new observations • Develop habitat models for additional HAB species in Chesapeake Bay • Incorporate additional environmental variables into habitat models and nowcast system to enhance HAB prediction capability

29. Problem With Empirical Approach Phytoplankton succession is a series of false starts in everchanging directions toward a momentarily defined unispecific specificity which is never achieved because the environment immediately changes and alters the direction of competitive pressure. G. E. Hutchinson, 1967 A Treatise on Limnology Environmental conditions change and may not be applicable for forecasting the distribution of species under future climatic conditions, i.e. species acclimate to environment.