Drone-based mapping of canopy biochemistry with leaf spectral library
Abstract
Leaf spectral library is a spectral dataset marked by species. Coupled with drone, it has potential for species specific management in heterogenous urban ecosystem. Due to canopy architecture affects its spectral information, however, leaf data could not be applied to the canopy. Thus, scaling up from leaf to canopy is necessary. Here, we build leaf spectral library and up-scale it to canopy level. Then we map 6 canopy biochemicals using drone-based imaging system. Study area is the largest park in suwon, South Korea. To build leaf spectral library, we target 10 representative species in the park. We collect leaf reflectance and the biochemicals consisting of chlorophyll (Cab), carotenoid (Car), water (Cw), LMA, C and N. Coupled with reflectance, our library includes 133 samples for pigments, 73 for C and N and 101 for water and LMA. Seasonally it covers early summer to late autumn. At leaf scale, we apply partial least square (PLS) regression to pairs of reflectance and biochemical. PLS estimates biochemicals with acceptable root mean squared error(RMSE) and coefficient of determination(R2) for Cab(RMSE: 2.60 ㎍/㎠, R2: 0.96), Car(RMSE: 0.54 ㎍/㎠, R2: 0.89), C(RMSE: 4.94 g/㎡, R2: 0.80), N(RMSE: 0.22 g/㎡, R2: 0.87), Cw(RMSE: 14.80 g/㎡, R2: 0.96) and LMA(RMSE: 12.48 g/㎡, R2: 0.69). To up-scale the library, we combine SAIL model and PLS regression. SAIL is used to simulate canopy reflectance using reflectance, transmittance and canopy architectural scenarios. As leaf transmittance is hard to acquire immediately, however, we generate leaf transmittance using PROSPECT inversion. As PLS shows good performance in extracting biochemicals from reflectance, we use PLS to retrieve biochemicals from simulated reflectance. First, we set canopy structural scenarios varying leaf area index and leaf average inclination angle. For each scenario, SAIL model simulates a set of canopy reflectance and PLS regression is applied to them. Then, the mean value which has the lowest relative standard deviation is chosen. To evaluate our method, we compare field measured data to biochemical maps directly. Though we used modeled leaf transmittance, our research suggests a way to use leaf spectral library for plant management.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMB060.0006C
- Keywords:
-
- 0410 Biodiversity;
- BIOGEOSCIENCES;
- 0466 Modeling;
- BIOGEOSCIENCES;
- 0480 Remote sensing;
- BIOGEOSCIENCES;
- 1922 Forecasting;
- INFORMATICS