Woltz, Victoria L. Zhu, Zhiliang Peneva-Reed, Elitsa I. 2022 Raster digital data Reston,VA U.S. Geological Survey https://doi.org/10.5066/P9JAE5JC Mangrove species dominance on Pohnpei island, Federated States of Micronesia was modeled with two geospatial model types: k-nearest neighbor (KNN) and random forest (RF) and a common set of predictors. Dominant mangroves were defined as species comprising the largest basal area per field plot. The RF model predicted species dominance for each species separately, resulting in 8 maps (one for each species). The maps of Rhizophora stylosa and R. mucronata dominance were combined because these species were difficult to tell apart in field identification (resulting in the 7 maps presented here). The KNN model produced one map, which shows all species' dominance locations in one raster layer. The KNN model results were the best based on field data and in field knowledge of the area but the RF results are still shared here. To map mangrove species dominance on Pohnpei island, Federated States of Micronesia. 2016 2017 ground condition Complete None planned Pohnpei, Federated States of Micronesia (FSM) 158.082478 158.408487 7.031884 6.737392 USGS Thesaurus geospatial analysis modeling forest resources USGS Metadata Identifier USGS:62bf11ebd34e82c548ced84c Getty Thesaurus of Geographic Names Pohnpei (island) USGS Biocomplexity Thesaurus Trees Integrated Taxonomic Information System (ITIS) 2022 ONLINE_REFERENCE Washington, D.C. Integrated Taxonomic Information System (ITIS) http://itis.gov Zhiliang Zhu ECOSYSTEMS: OFFC OF ASSOC DIR FOR ECOSYSTEMS Senior Physical Scientist mailing and physical

Mail Stop 300, 12201 Sunrise Valley Dr

Reston VA 20192 US 703-648-4243 zzhu@usgs.gov expert advice Most mangroves within the plots are identified to the species level, except for some individuals of genus Rhizophora, which were removed from the dominance mapping analysis. Mangrove species (and/or putative hybrids) inventoried in our surveys included Bruguiera gymnorhiza, Lumnitzera littorea, Rhizophora apiculata, R. x lamarckii, R. mucronata, R. stylosa, Sonneratia alba, and Xylocarpus granatum. The mangrove species, Pemphis acidula and Heritiera littoralis, were not encountered during our surveys, though both have been recognized for its presence on the island. Kingdom Plantae Subkingdom Viridiplantae Infrakingdom Streptophyta Superdivision Embryophyta Division Tracheophyta Subdivision Spermatophytina Class Magnoliopsida Superorder Rosanae Order Malpighiales Family Rhizophoraceae Genus Bruguiera Species Bruguiera gymnorhiza Burmese mangrove Kingdom Plantae Subkingdom Viridiplantae Infrakingdom Streptophyta Superdivision Embryophyta Division Tracheophyta Subdivision Spermatophytina Class Magnoliopsida Superorder Rosanae Order Myrtales Family Combretaceae Genus Lumnitzera Species Lumnitzera littorea Kingdom Plantae Subkingdom Viridiplantae Infrakingdom Streptophyta Superdivision Embryophyta Division Tracheophyta Subdivision Spermatophytina Class Magnoliopsida Superorder Rosanae Order Malpighiales Family Rhizophoraceae Genus Rhizophora Species Rhizophora apiculata mangrove Kingdom Plantae Subkingdom Viridiplantae Infrakingdom Streptophyta Superdivision Embryophyta Division Tracheophyta Subdivision Spermatophytina Class Magnoliopsida Superorder Rosanae Order Malpighiales Family Rhizophoraceae Genus Rhizophora Species Rhizophora mucronata mangrove Kingdom Plantae Subkingdom Viridiplantae Infrakingdom Streptophyta Superdivision Embryophyta Division Tracheophyta Subdivision Spermatophytina Class Magnoliopsida Superorder Rosanae Order Myrtales Family Lythraceae Genus Sonneratia Species Sonneratia alba Kingdom Plantae Subkingdom Viridiplantae Infrakingdom Streptophyta Superdivision Embryophyta Division Tracheophyta Subdivision Spermatophytina Class Magnoliopsida Superorder Rosanae Order Sapindales Family Meliaceae Genus Xylocarpus Species Xylocarpus granatum Kingdom,Kingdom Plantae,Plantae Subkingdom,Subkingdom Viridiplantae,Viridiplantae Infrakingdom,Infrakingdom Streptophyta,Streptophyta Superdivision,Superdivision Embryophyta,Embryophyta Division,Division Tracheophyta,Tracheophyta Subdivision,Subdivision Spermatophytina,Spermatophytina Class,Class Magnoliopsida,Magnoliopsida Superorder,Superorder Rosanae,Rosanae Order,Order Malpighiales,Malpighiales Family,Family Rhizophoraceae,Rhizophoraceae Genus,Genus Rhizophora,Rhizophora Species,Species Rhizophora stylosa,Rhizophora x lamarckii none none Victoria L Woltz ECOSYSTEMS: OFFC OF ASSOC DIR FOR ECOSYSTEMS Physical Scientist mailing and physical

Mail Stop 300, 12201 Sunrise Valley Dr

Reston VA 20192 US 703-648-4523 vwoltz@usgs.gov The RF model that was used to create this map was created in RStudio. This Code is provided online with this map publication. The RF model that was used to create this map was created in RStudio. This Code is provided online with this map publication. Victoria L Woltz ECOSYSTEMS: OFFC OF ASSOC DIR FOR ECOSYSTEMS Physical Scientist mailing and physical

Mail Stop 300, 12201 Sunrise Valley Dr

Reston VA 20192 US 703-648-4523 vwoltz@usgs.gov All RF models were significant at a 95% confidence interval (p<0.05). Once the models were created for each species, the model algorithms were used to “back predict” the data (or to find the predicted species dominance at field plots). This initial model evaluation was good for B. gymnorhiza, R. apiculata, R. mucronata, S. alba and X. granatum. However, L. littorea, R. x lamarckii, and R. stylosa had poor area under the receiver operating characteristics curve (AUC) values meaning the model had difficulty distinguishing between dominant and nondominant locations for these species. L. littorea and R. x lamarckii also had low Cohen's Kappa values from back predicting indicating only moderate agreement with field data. Upon further inspection of the models, a 10-fold, 1,000-time cross-validation yielded low cross-validation kappas ranging between 0 and 0.34 for all models, which equates to no agreement to fair agreement. Cross-validation kappas may be low due to imbalances between the numbers of dominant and nondominant plots despite the precautions that were used to minimize this effect. The relatively high RF model performance after back prediction compared with the fair to no agreement according to cross-validation kappas tell us it is likely that our RF models were overfitted. The KNN model outperformed the random forest model based on field data and in field knowledge of the area. Though the KNN model underpredicts minority species, it predicted B. gymnorhiza, R. apiculata and S. alba as the three most common dominant mangrove species on Pohnpei as is true in the field data. Conversely, the RF model lowers the dominance of R. apiculata and S. alba to similar levels of more rare species in the field data such as R. mucronata, R. x lamarckii, and L. littorea. No formal logical accuracy tests were conducted Data set is considered complete for the information presented, as described in the abstract. Users are advised to read the rest of the metadata record carefully for additional details. No formal positional accuracy tests were conducted No formal positional accuracy tests were conducted Field Dominant species, defined as species comprising the largest basal area per field plot, were analyzed separately with two geospatial model types: k-nearest neighbor (KNN) and random forest (RF) and a common set of predictor. Basal area for training and testing purposes was found using plot data from a total of 273 plots which were inventoried in 2016 and 2017 for forest structure and mangrove species (Peneva-Reed et al., 2019). Plots were 10 m in radius with 3 m radius subplots. Predictor variables with 5 m resolutions were created in ArcMap 10.7.1, including principal components of WorldView-3, WorldView-2 and QuickBird satellite imagery composites; distance from water; elevation; and island side (leeward or windward). Due to heavy cloud cover, satellite image composite were used. The composites were comprised of WorldView-3 eight band images from from July, September and October 2018; WorldView-2 three band image from December 2013; and Quickbird four band images from January 2007 and June 2005. The WorldView-3, WorldView-2, and Quickbird images were resampled to the 5 meter resolution from 1.24, 1.64, and 2.62 m resolution, respectively. The DEM was derived in ArcMap using the elevation data collected on the island. To decrease spectral similarities between mangrove species, the images were analyzed using unsupervised principal component analysis (PCA) (Abdollahnejad et al., 2017; Liu et al., 2017). The first four principal components of the image analysis were used as predictors as they captured 99.55% of the variance. In our model, distance to water and elevation are proxies for unavailable data such as inundation frequency; the salinity gradient and soil physiochemical characteristics such as nitrogen, phosphorous, and sulfide (Crase et al., 2012, McKee et al., 2002). The digital elevation model (DEM) was created by extrapolating elevation data collected on the island [D. B. Gesch, personal communication September 26, 2019] (Thorne et al., 2019). None of the predictors were multicollinear (p<0.05) when tested with the rfUtilities package (Evans et al., 2011; Evans and Murphy, 2019) in RStudio (RStudio Team, 2018). RF model creation and evaluation were performed using the rfUtilities and randomForest packages in RStudio (Evans et al., 2011; Evans and Murphy, 2019; RStudio Team, 2018; Breiman et al., 2018). One random forest model was created for each species (Bruguiera gymnorhiza, Lumnitzera littorea, Rhizophora apiculata, R. x lamarckii, R. mucronata, R. stylosa, Sonneratia alba, and Xylocarpus granatum) for a total of eight RF models. RF models were performed in two ways; species data that had balanced dominant and nondominant plots were input into the randomForest function, while less balanced species data were input into the rf.classBalance function (Breiman et al., 2018; Evans and Murphy, 2019). All RF models were run as classification with 1,000 trees without replacement and with ~36% out-of-bag samples. The sum of sensitivity and specificity was maximized to determine the threshold of each species being dominant (Beaumont et al., 2016). Models were evaluated by determining their significance, running cross-validations and finding the performance metrics of back predictions (predicting values of the known plots using the created model). The 10-fold cross-validation was run 1,000 times, with 90% of the data used for training and 10% used for testing (Beaumont et al., 2016). Results from the RF model showed B. gymnorhiza was predicted to be dominant in 35% of the forest, with all other species dominant in 7% or less of the forest. In some areas of the mangrove forest no species were predicted to be dominant. In these areas, species may be more equal in terms of basal area. In other areas, more than one species was predicted to be dominant. This is not surprising since these species were modeled separately and because they have much overlap in the range of predictors in which they can survive and thrive. All RF models were significant at a 95% confidence interval (p<0.05). Once the models were created for each species, the model algorithms were used to “back predict” the data (or to find the predicted species dominance at field plots). This initial model evaluation was good for B. gymnorhiza, R. apiculata, R. mucronata, S. alba and X. granatum. However, L. littorea, R. x lamarckii, and R. stylosa had poor area under the receiver operating characteristics curve (AUC) values meaning the model had difficulty distinguishing between dominant and nondominant locations for these species. L. littorea and R. x lamarckii also had low Cohen's Kappa values from back predicting indicating only moderate agreement with field data. Upon further inspection of the models, a 10-fold, 1,000-time cross-validation yielded low cross-validation kappas ranging between 0 and 0.34 for all models, which equates to no agreement to fair agreement. Cross-validation kappas may be low due to imbalances between the numbers of dominant and nondominant plots despite the precautions that were used to minimize this effect. The relatively high RF model performance after back prediction compared with the fair to no agreement according to cross-validation kappas tell us it is likely that our RF models were overfitted. The KNN model outperformed the random forest model based on field data and in field knowledge of the area. Though the KNN model underpredicts minority species, it predicted B. gymnorhiza, R. apiculata and S. alba as the three most common dominant mangrove species on Pohnpei as is true in the field data. Conversely, the RF model lowers the dominance of R. apiculata and S. alba to similar levels of more rare species in the field data such as R. mucronata, R. x lamarckii, and L. littorea. References: Abdollahnejad, A.; Panagiotidis, D.; Joybari, S.S.; Surový, P. Prediction of dominant forest tree species using quickbird and environmental data. Forests 2017, 8: 42. Crase, B.; Liedloff, A.C.; Wintle, B.A. A new method for dealing with residual spatial autocorrelation in species distribution models. Ecography 2012, 35: 879-888. Beaumont LJ, Graham E, Duursma DE, Wilson PD, Cabrelli A, Baumgartner JB, et al. Which species distribution models are more (or less) likely to project broad-scale, climate-induced shifts in species ranges? Ecol Modell. 2016;342: 135-146. Breiman L, Cutler A, Liaw A, Wiener M. Breiman and Cutler's random forests for classification and regression. [R package version 4.6–14]. 2018. Available from: https://cran.r-project.org/ Evans JS, Murphy MA. Random forests model selection and performance evaluation. [R package version 2.1–5]. 2019. Available from: https://cran.r-project.org/ Evans, J.S.; Murphy, M.A.; Holden Z.A.; Cushman, S.A. Modeling species distribution and change using random forest. In Predictive species and habitat modeling in landscape ecology: Concepts and applications. Drew C, Wiersma Y, Huettmann F, eds. New York: Springer Science and Business Media. 2011, P139-159. Liu, L.; Coops, N.C.; Aven, N.W.; Pang, Y. Mapping urban tree species using integrated airborne hyperspectral and LiDAR remote sensing data. Remote Sens. Envir. 2017, 200: 170-182. McKee, K.L.; Feller, I.C.; Popp, M.; Wanek, W. Mangrove isotopic (δ15N and δ13C) fractionation across a nitrogen vs. phosphorus limitation gradient. Ecology 2002, 83: 1065-1075. Peneva-Reed, E.I., Woltz, V.L., and Zhu, Z., 2019, Aboveground mangrove biomass data collected from and species dominance maps of Pohnpei, Federated States of Micronesia: U.S. Geological Survey data release, https://doi.org/10.5066/P9JAE5JC. RStudio Team. RStudio: Integrated Development for R. RStudio, Inc., Boston, MA URL http://www.rstudio.com/. 2018. Thorne, K.; Buffington, K.M.; MacKenzie, R.A.; Krauss, K.; Ellison, J.C.; Peneva-Reed, E.; et al. Modeling mangrove ecosystem sea-level rise vulnerability for Pohnpei, Micronesia [abs.]. In Fall Meeting, San Francisco, California. Proceedings: Washington, D.C., American Geophysical Union 2019, pap. GC51K-0995 Peneva-Reed, Elitsa I., Woltz, Victoria L., and Zhu, Zhiliang 20190820 csv, shapefile, raster, and R code Reston, VA US Geological Survey https://doi.org/10.5066/P9JAE5JC Field Data, Maps, and Code 20160322 20170901 Peneva-Reed, E.I., Woltz, V.L., and Zhu, Z., 2019, Aboveground mangrove biomass data collected from and species dominance maps of Pohnpei, Federated States of Micronesia: U.S. Geological Survey data release, https://doi.org/10.5066/P9JAE5JC. Biomass Data Collected from Pohnpei The field data was used in the model to find species dominance The steps are outlined in the methods and in the R Code used to create this data which is also provided online with this map publication. 2021 N/A Raster Universal Transverse Mercator 57N 0.9996 159.0 0.0 500000.0 0.0 coordinate pair 5.0 5.0 METERS WGS_1984_UTM_Zone_57N, WKID: 32657, Authority: EPSG WGS_1984 6378137.0 298.257223563 Bruguiera.gymnorhiza.dominance_RFmodel.tif Map of Bruguiera gymnorhiza dominance on Pohnpei, FSM Producer defined OID Object identifier ESRI 0 1 Object ID Value This attribute defines if the species is predicted to be dominant in this location of the mangrove forest. Producer defined 1 The species is predicted to be dominant at locations where the value is “1” Producer defined 0 The species is not predicted to be dominant at locations where the value is “0” Producer defined Count The number of cells in this category ESRI This is the number of cells in this category Lumnitzera.littorea.dominance_RFmodel.tif Map of Lumnitzera littorea dominance on Pohnpei, FSM Producer defined OID Object identifier ESRI 0 1 Value This attribute defines if the species is predicted to be dominant in this location of the mangrove forest. Producer defined 1 The species is predicted to be dominant at locations where the value is “1” Producer defined 0 The species not is predicted to be dominant at this location Producer defined Count The number of cells in this category ESRI This is the number of cells in this category Rhizophora.apiculata.dominance_RFmodel.tif Map of Rhizophora apiculata dominance on Pohnpei, FSM Producer defined OID Object identifier ESRI 0 1 Value This attribute defines if the species is predicted to be dominant in this location of the mangrove forest. Producer defined 1 The species is predicted to be dominant at locations where the value is “1” Producer defined 0 The species is not predicted to be dominant at locations where the value is “0” Producer defined Count The number of cells in this category ESRI This is the number of cells in this category Rhizophora.stylosa.mucronata.dominance_RFmodel.tif Map of Rhizophora stylosa and R. mucronata dominance on Pohnpei, FSM. The modeled results for Rhizophora stylosa and Rhizophora mucronata were combined because R. mucronata and R. stylosa are difficult to distinguish without flowers and commonly occur in the same areas and therefore could have been confused in the field survey which this map is based off of. Producer defined OID Object identifier ESRI 0 1 Value This attribute defines if the species is predicted to be dominant in this location of the mangrove forest. Producer defined 1 The species is predicted to be dominant at locations where the value is “1” Producer defined 0 The species is not predicted to be dominant at locations where the value is “0” Producer defined Count The number of cells in this category ESRI This is the number of cells in this category Rhizophora.x.lamarckii.dominance_RFmodel.tif Map of Rhizophora x lamarckii dominance on Pohnpei, FSM Producer defined OID Object identifier ESRI 0 1 Object identifier Value This attribute defines if the species is predicted to be dominant in this location of the mangrove forest. Producer defined 1 The species is predicted to be dominant at locations where the value is “1” Producer defined 0 The species is not predicted to be dominant at locations where the value is “0” Producer defined Count The number of cells in this category ESRI This is the number of cells in this category Sonneratia.alba.dominance_RFmodel.tif Map of Sonneratia alba dominance on Pohnpei, FSM Producer defined OID Object identifier Producer defined 0 1 Value This attribute defines if the species is predicted to be dominant in this location of the mangrove forest. Producer defined 1 The species is predicted to be dominant at locations where the value is “1” Producer defined 0 The species is not predicted to be dominant at locations where the value is “0” Producer defined Count The number of cells in this category ESRI This is the number of cells in this category Xylocarpus.granatum.dominance_RFmodel.tif Map of Xylocarpus granatum dominance on Pohnpei, FSM Producer defined OID Object identifier ESRI 0 1 Value This attribute defines if the species is predicted to be dominant in this location of the mangrove forest. Producer defined 1 The species is predicted to be dominant at locations where the value is “1” Producer defined 0 The species is not predicted to be dominant at locations where the value is “0” Producer defined Count The number of cells in this category ESRI This is the number of cells in this category Victoria L Woltz ECOSYSTEMS: OFFC OF ASSOC DIR FOR ECOSYSTEMS Physical Scientist mailing and physical

Mail Stop 300, 12201 Sunrise Valley Dr

Reston VA 20192 US 703-648-4523 vwoltz@usgs.gov Unless otherwise stated, all data, metadata and related materials are considered to satisfy the quality standards relative to the purpose for which the data were collected. Although these data and associated metadata have been reviewed for accuracy and completeness and approved for release by the U.S. Geological Survey (USGS), no warranty expressed or implied is made regarding the display or utility of the data for other purposes, nor on all computer systems, nor shall the act of distribution constitute any such warranty User must be able to open a raster file. File can be opened in ESRI's ArcMap, ArcGIS Pro, or QGIS. 20220926 Victoria L Woltz ECOSYSTEMS: OFFC OF ASSOC DIR FOR ECOSYSTEMS Physical Scientist mailing and physical

Mail Stop 300, 12201 Sunrise Valley Dr

Reston VA 20192 US 703-648-4523 vwoltz@usgs.gov FGDC Biological Data Profile of the Content Standard for Digital Geospatial Metadata FGDC-STD-001.1-1999