The Malagasy rosewood species Dalbergia maritima has a long history of unsustainable exploitation for its beautiful, burgundy-colored heartwood. As currently circumscribed, D. maritima has a wide geographic distribution in eastern Madagascar and exhibits significant morphological, ecological, and genetic variation, suggesting it may comprise more than a single entity. Multivariate analyses of leaf, flower, and inflorescence characters as well as eco-geographic features reveal several morphologically well delimited entities with distinct habitat preferences and/or geographic ranges, which are consistent with results from recent phylogenomic and population genomic studies of Malagasy Dalbergia. Based on these findings, we describe and illustrate two new species from southeastern Madagascar comprising material previously assigned to D. maritima, viz. D. pseudomaritima, characterized by paniculate inflorescences and small, broadly elliptic to orbicular, glabrous leaflets, and D. razakamalalae, distinguished by racemose inflorescences with large flowers, and narrowly ovate to narrowly elliptic, glabrous leaflets. Dalbergia maritima is consequently re-circumscribed to include only populations from east-central Madagascar, within which we recognize two subspecies, D. maritima subsp. maritima, with glabrous leaves, inflorescence axes, and gynoecia, occurring in littoral forest habitats, and D. maritima subsp. pubescens, with indument on these structures, and growing in evergreen humid forest farther inland. Photos are provided for each taxon, along with line drawings for the two new species. Provisional IUCN Red List assessments indicate that all three species are Endangered, D. maritima and D. razakamalalae mainly because of selective logging for trade in their high-quality heartwood, and D. pseudomaritima primarily because of habitat degradation due to land clearing and fire for subsistence agriculture, which has important implications for their conservation and sustainable management.

1. Methods for collection / processing the data:

    Morphological Measurements
        We assessed 13 leaf and leaflet characters (Supplementary Material 3), along with
        11 inflorescence and flower characters (Supplementary Material 4). Continuous and
        discrete characters (see Manuscript Table 1), which were measured several times on
        a given collection, were recorded as sample medians.
       
    Ecological Characteristics
        We assessed 17 potentially relevant ecological characteristics (Supplementary
        Material 5) from available spatial raster or vector data for Madagascar (see
        Manuscript Table 2).
       
        We used R (R Core Team 2020) version 4.0.2 and the elevatr package (Hollister et
        al. 2020) version 0.3.1, the terra package (Hijmans 2021) version 1.0.10, and the
        fasterRaster package (Smith 2020) version 0.6.0 to download high-resolution
        elevation data (3 arc seconds, ca. 90 m resolution) and to perform raster
        calculations.
       
        All rasters were projected to Universal Transverse Mercator (UTM) zone 38S and
        re-sampled to the resolution of the highest-resolved raster (30 m) when needed.
        We then extracted the ecological characteristics of the selected collections,
        resulting in an ecological dataset for multivariate analysis.
               
    Multivariate analyses
        Commented R code documenting the extraction of ecological characteristics from
        occurrence data as well as all panels of Figure 3 is available in
        SupplementaryMaterial_2.R.

2. Instrument- or software-specific information needed to interpret the data:

    This is the R session info documenting the R and R package versions used

    > sessionInfo()
    R version 4.0.2 (2020-06-22)
    Platform: x86_64-apple-darwin17.0 (64-bit)
    Running under: macOS Catalina 10.15.7

    Matrix products: default
    BLAS:   /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
    LAPACK: /Library/Frameworks/R.framework/Versions/4.0/Resources/lib/libRlapack.dylib

    locale:
    [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

    attached base packages:
    [1] stats     graphics  grDevices utils     datasets  methods   base    

    other attached packages:
      [1] XML_3.99-0.8     rgdal_1.5-27     cleangeo_0.2-4   maptools_1.1-2   rgeos_0.5-8      ggrepel_0.9.1  
      [7] ggforce_0.3.3    ggplot2_3.3.5    rmapshaper_0.4.5 concaveman_1.1.0 tmap_3.3-2       alphahull_2.2  
    [13] ConR_1.3.0       raster_3.5-2     sp_1.4-6         terra_1.4-22     ape_5.5          cluster_2.1.2  
    [19] adegenet_2.1.5   ade4_1.7-18      tabplot_1.4.1    ffbase_0.13.3    ff_4.0.5         bit_4.0.4  


3. Describe any quality-assurance procedures performed on the data:

    Some missing values in the measurement dataset were imputed using class means before
    subjecting it to multivariate analyses (PCA, PCoA), see R functions impute and ordplot.
    
    Missing data code is "NA".


4. People involved with sample collection, processing, analysis and/or submission:

    Simon Crameri, Nivohenintsoa Rakotonirina