Fig 6: Example of generating (x,y) leaf coordinates pairs for American beech at BART
 
● We use remote time-lapse cameras mounted level with sunlit treetops to take weekly leaf photos through growing season.
● From which we measure (x,y) coordinates pairs of petioles and tips of 75 leaves ((x,y) of 150 points).
● We compute species MLA from the (x,y) pairs.
 
3.3. SPECTRAL INDICES
We compute several spectral indices such as near-infrared reflectance of vegetation (NIRv) and their derived gray-level co-occurrence matrices for each crown.
 
3.4. SITE-SPECIES COMBO 
From these data, we construct site-species trait combos and test for species difference and trait co-variations.
 
4. RESULTS & DISCUSSION
● Early successional tuliptrees (LITU, Fig 7) with a tower architecture (Fig 1) have more vertical leaves (Fig 8a) and higher rugosity (Fig 8b) than mid-successional black oak (QUVE) and red oak (QURU).
● Late-successional mesic sugar maple (ACSA) with a dome architecture has the most horizontal leaves (Fig 8a) and the lowest rugosity (Fig 8b), resulting in higher NIRv reflectance (Fig 9).
● We suggest that more vertical leaf angles allow lower leaves to receive more sunlight, an advantage for light harvesting in early-successional species to grow, but a disadvantage for shading out neighbors in late-successional species such as ACSA.  
●  Our APAD50 metric (Fig 8c & Fig 7) indicates that LITU has more "top-heavy" profile - leaves clustered at the top 50% of the crown, which fits with its production-oriented strategy focused on optimizing light harvesting at the expense of casting shade or conserving water. ACSA is more "bottom-heavy" with most leaves in the lower half of its crown.
● These ecological adaptations of crown architecture, expressed through economic trade-offs among crown traits (Fig 9), have underappreciated impacts upon NIR reflectance and forest functioning.