A single hyperspectral image can detect growth rate variation within and among tropical tree species

Background/Question/Methods

Single remotely sensed images, particularly information-rich hyperspectral images, can measure key chemical and structural properties of tree canopies.  However, multiple, well-aligned, high-resolution images are needed to measure dynamic rates of trees, such as growth and mortality, which are critical for landscape-scale monitoring of forest dynamics and managing forests for production and conservation. Due to the high investment of acquiring a single image, obtaining multiple images of the same location is difficult. However, chemical and structural properties measured by a single image are related to tree dynamic rates such that a single hyperspectral image could potentially discriminate individual trees with different growth rates.  We tested this hypothesis in an 11-year old reforestation species trial in the Azuero Peninsula of Panama.  We measured growth of 20 species planted in pure species blocks where canopies could be identified in a high resolution (2 m) hyperspectral image (480 – 2500 nm) from the Carnegie Airborne Observatory.  We used linear discriminant analysis and multiple regression on all possible two-wavelength ratios to develop models that described variation in growth rates within and among species. On the same trees, we measured canopy structure and leaf chemical properties that may be mediating growth rate detection in hyperspectral images.

Results/Conclusions

We developed a model of 14 two-wavelength ratios that explained over 70% of the variation in standardized growth rates among trees.  This analysis did not include species identity, indicating a general relationship between growth rate and image reflectance.  Within-species, selected narrowband ratios predicted pixel membership in either low, medium and high growth sites with a range of 71-100% accuracy.   The wavelengths selected in the model suggest that leaf pigment content, nitrogen, and possibly water content may be driving the relationship between reflectance and growth rates.  Preliminary analyses of leaf property data indicate consistent differences between high and low growth sites that may provide a link between the strong relationship quantified between reflectance and growth rates.  If the detection of growth from a single hyperspectral image can be replicated at other sites, hyperspectral aerial images promise to be an important tool for dynamic global vegetation models and precision forestry.