Linking seasonal foliar traits to VSWIR-TIR spectroscopy across California ecosystems

Autor(es)

Meerdink, Susan K. ; Roberts, Dar A. ; King, Jennifer Y. ; Roth, Keely L. ; Dennison, Philip E. ; Amaral, Cibele H. ; Hook, Simon J.

• Tipo de documento: Artigo Científico
• Fonte: REMOTE SENSING OF ENVIRONMENT; v. 186, p. 322-338, DEC 1 2016.
• Citações Web of Science: 15

Resumo

Vegetation traits provide critical information on ecosystem function that can be used to assess the effects of disturbance, land use, and climate change. Recent studies have demonstrated the use of spectroscopy to predict vegetation traits accurately and efficiently. To date, most spectroscopic studies have utilized data from the Visible Short Wave Infrared spectrum (VSWIR) or, occasionally, the Thermal Infrared spectrum (TIR), but not in combination. This study focuses on VSWIR and TIR synergy to evaluate the ability to predict leaf level cellulose, lignin, leaf mass per area (LMA), nitrogen, and water content across seasons. We used fresh leaves from sixteen common California shrub and tree species collected in the 2013 spring, summer, and fall seasons. The 284 samples exhibited a wide range of leaf traits as determined by standard analytical procedures: 4.2-27.3% for cellulose, 2.6-22.5% for lignin, 34.7-388.9 g/m(2) for LMA, 0.45-3.81% for nitrogen, and 20.2-76.9% for water content. For each leaf trait, partial least squares regression (PLSR) models were fit using different portions of the spectrum: VSWIR (0.35-2.5 mu m), TIR (2.5-15.4 mu m), and Full spectrum (0.35-15.4 mu m). We also fit PLSR models using spectra resampled to simulate three airborne sensors: the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS; 0.4-2.5 mu m), the Hyperspectral Thermal Emission Spectrometer (HyTES; 7.5-12 mu m), and the Hyperspectral InfraRed Imager (HyspIRl; 0.4-12 mu m). The majority of best performing models used the Full spectrum, demonstrating the value of combining TIR and VSWIR spectra for leaf trait prediction. Sensor simulated PLSR models created with the entire data set yielded validation R-2 and root mean square error of prediction (RMSEP) values as follows: R-2 = 0.70 and RMSEP = 13.1% for cellulose, R-2 = 0.50 and RMSEP = 17.7% for lignin, R-2 = 0.56 and RMSEP = 18.3% for LMA, R-2 = 0.56 and RMSEP = 18.1% for nitrogen, and R-2 = 0.89 and RMSEP = 5.7% for water content. General models successfully captured the variability among all seasons and leaf forms for cellulose and water content, while the other leaf traits were better modeled with season or leaf form -specific models. This study successfully captured the large seasonal and geographical variation in leaf traits across California's diverse ecosystems, supporting the possibility of using HyspIRl's imagery for global mapping efforts of these traits. (C) 2016 Elsevier Inc. All rights reserved. (AU)

• Processo FAPESP: 10/51718-0 - Utilizacao de imagens hiperespectrais de alta resolucao espacial no reconhecimento remoto de especies da flora do sudeste do brasil: desenvolvimento metodologico e aplicacao potencial na geobotanica
• Beneficiário: Cibele Hummel Do Amaral
• Modalidade de apoio: Bolsas no Brasil - Doutorado