Mapping distributions of marine organisms using environmental niche modelling - AquaMaps K. Kaschner, J. Ready, S. Kullander, R. Froese and many more….INCOFISH, FishBase…
AquaMaps Basic Concept Environmental envelope type modeling approach Predictor Preferred min Preferred max MinMax P Max Species-specific environmental envelopes Relative probability of occurrence (HSPEN) (HCAF) (HSPEC) INTRODUCTION
HCAF table Environmental data per 0.5 degree latitude / longitude square Contents –Bathymetry –Mean annual SST (Sea surface temperature) –Mean annual Salinity –Mean annual Chlorophyll A (now primary production) –Mean annual Sea ice concentration (replacing distance to ice edge) –Mean annual distance to land –Etc.
AquaMaps Basic Concept INTRODUCTION P c = P Bathymetry c * P SST c * P Salinity c * P ChloroA c * P IceDist c * P LandDist c
AquaMaps Basic Concept INTRODUCTION European flounder (Platichthys flesus)
AquaMaps Basic Concept INTRODUCTION European flounder (Platichthys flesus)
Environmental Envelopes: Sources of Information Envelopes can be defined based on expert knowledge / published information –E.g. depth ranges for fishes -> FishBase automatically generated based on species records (point data) ENVELOPES
Automated Envelope Generation: 1. Step: Selection of Species Records ENVELOPES
Automated Envelope Generation: 1. Step: Selection of Species Records Minimum: n = 10 records with reliable species ID & location information ENVELOPES European flounder (Platichthys flesus), n = 65
2. Step: Selection of “Good” Records Cross-check with known FAO areas of occurrence (e.g. FishBase) ENVELOPES
2. Step: Selection of “Good” Records Cross-check with known FAO areas of occurrence (e.g. FishBase) (N.B. Chilean e.g. dealt with by non-native status exclusion) ENVELOPES
2. Step: Selection of “Good” Records Cross-check with known FAO areas of occurrence (e.g. FishBase) ENVELOPES European flounder (Platichthys flesus), n = 33
3. Step: Grouping over “Good” Cells Mean annual SST [C] ENVELOPES Mean annual SST [C] Frequency Non-grouped records (n = 33) Records grouped over cells (n = 20) Minimum: n = 10 cells
4. Step: Calculate Percentile Ranges ENVELOPES Mean annual SST [C] Max =16.75Min =1.6575% = % = 9.06
4. Step: Calculate Percentile Ranges ENVELOPES Mean annual SST [C] - 2SD = 4.09Mean = SD = SD = SD = 19.51
4. Step: Calculate Percentile Ranges ENVELOPES Min25%75%Max Depth SST [C] Salinity [ppu] ChloroA [?] IceDist [km] LandDist [km]
4. Step: Calculate Percentile Ranges ENVELOPES 25% -75 % Percentile = “Preferred range”
4. Step: Calculate Percentile Ranges ENVELOPES 25% -75 % Percentile = “Preferred range”
4. Step: Calculate Percentile Ranges ENVELOPES 25% -75 % Percentile = “Preferred range”
4. Step: Calculate Percentile Ranges ENVELOPES Mean annual SST [C] Max =16.75Min = % = % = 7.27
4. Step: Calculate Percentile Ranges ENVELOPES Min10%90%Max Depth SST [C] Salinity [ppu] ChloroA [?] IceDist [km] LandDist [km]
4. Step: Calculate Percentile Ranges ENVELOPES 10% -90 % Percentile = “Preferred range”
4. Step: Calculate Percentile Ranges ENVELOPES 10% -90 % Percentile = “Preferred range”
5. Step: Broadening of Min-Max Ranges ENVELOPES Mean annual SST [C] Max =1.5 * Interquartile = % = % = 7.27 Min =1.5 * Interquartile = Note that if true value is more extreme then this is kept
ENVELOPES Min10%90%Max Depth SST [C] Salinity [ppu] ChloroA [?] IceDist [km] LandDist [km] Step: Broadening of Min-Max Ranges
6. Step: Ensure Minimum Range Width ENVELOPES Mean annual SST [C] ΔMin = 1 °C ΔMin = 2 °C
ENVELOPES 6. Step: Ensure Minimum Range Width 1 °C 2 °C 1 ppu 2 ppu km 4 km 2 km 4 km Min10%90%Max Depth SST [C] Salinity [ppu] ChloroA [?] IceDist [km] LandDist [km]
ENVELOPES 7. Step: Store Envelope in HSPEN Min10%90%Max Depth SST [C] Salinity [ppu] ChloroA [?] IceDist [km] LandDist [km]
Model Algorithm Predictor Preferred min Preferred max MinMax P Max Relative probability of occurrence MODEL ALGORITHM
Model Algorithm MODEL ALGORITHM P c = P Bathymetry c * P SST c * P Salinity c * P ChloroA c * P IceDist c * P LandDist c – Multiplicative approach: Each parameter can act as “knock-out” criterion Redundant parameters have no effect on distribution
Model Output ALGORITHM
Model Output ALGORITHM
Effects of Individual Predictors MODEL ALGORITHM Bathymetry
Effects of Individual Predictors MODEL ALGORITHM SST
Effects of Individual Predictors MODEL ALGORITHM Salinity
Effects of Individual Predictors MODEL ALGORITHM Chlorophyll A
Effects of Individual Predictors MODEL ALGORITHM Distance to ice edge
Effects of Individual Predictors MODEL ALGORITHM Distance to land
Additional Rules If Min IceEdgeDist > 1000 km then ignore parameter (Rethinking – data changing to ice concentration) If Max LandDist > 1000 km then Max LandDist = maximum distance (4000 km) MODEL ALGORITHM
Preliminary Results EXAMPLES Atlantic herring (Clupea harengus), n = 7500
Preliminary Results EXAMPLES Atlantic herring (Clupea harengus), n = 7500
Preliminary Results EXAMPLES Atlantic cod (Gadus morhua), n = 215
Preliminary Results EXAMPLES Atlantic cod (Gadus morhua), n = 215
Preliminary Results EXAMPLES Tropical two-wing flyingfish (Exocoetus volitans), n = 330
Preliminary Results EXAMPLES Tropical two-wing flyingfish (Exocoetus volitans), n = 330 Data cleaning needed
Preliminary Results EXAMPLES Tope shark (Galeorhinus galeus), n = 110
Preliminary Results EXAMPLES Tope shark (Galeorhinus galeus), n = 110
Preliminary Results EXAMPLES Orange roughy (Hoplostethus atlanticus), n = 116
Preliminary Results EXAMPLES Orange roughy (Hoplostethus atlanticus), n = 116
Preliminary Results EXAMPLES Coelacanth (Latimeria chalumnae), n = 10
Preliminary Results EXAMPLES Coelacanth (Latimeria chalumnae), n = 10
Preliminary Results EXAMPLES Coelacanth (Latimeria chalumnae), n = 10
Preliminary Results EXAMPLES Red lionfish (Pterois volitans), n = 65
Preliminary Results EXAMPLES Red lionfish (Pterois volitans), n = 65
Points for Investigation DISCUSSION Advantages/disadvantages of envelope modeling in comparison to other habitat suitability modeling / mapping approaches (GARP, Maxent, Bioclim etc.) Minimum number of records required? Environmental data –Seasonal data –Historical and predicted future data –Categorical data? E.g. habitat types Multiplicative model (Geometric mean)? Weighting factors (e.g. known forcing factors)? Effects of effort biases? Others?
Existing modelling DISCUSSION Other presence only modelling –GARP (Genetic Algorithm for Rule-Set Parsimony) The ‘industry standard’ but a bit of a ‘black box’ –Maxent (Maximum entropy) – latest popular method A machine learning method, iterating algorithm Computationally quite fast (but not as fast as AquaMaps) –Bioclim – early simplistic method Uses similar approach to envelopes Moderately fast computation
AquaMaps compared DISCUSSION Advantages –Speed Simple calculations take very little time Can be done on-the fly over the internet ( Tools/AquaMaps) –FAO area use to block out areas of known absence Can be switched off to allow prediction of areas that could be invaded –Batch processing runs the whole database in one go – many species Potential Disadvantages –Accuracy? As yet unknown – testing underway but looks good at this scale –Resolution? 0.5 degree scale difficult to reapply at local scales without remaking HCAF But - Other methods also require the environmental data sets to be provided at the correct scale
Acknowledgements FishBase – Provision of data and interface –Occurrence records, depth data, FAO area assignment BADC (British Atmospheric Data Centre) – Provision of data from global climate models –Future and past environmental data (just beginning) –Plan to predict the effects of climate change of fish distributions using: Historical data - 100yrs ago and 50yrs ago Future modelled data - 20yrs time, 50yrs time, 100yrs time INCOFISH partners