Passive Microwave and Optical Indices-Based Approaches for Estimating Surface Conductance in Forest Ecosystems
Friday, 19 December 2014
The ability to monitor evapotranspiration (ET) from land surface is relevant for applications requiring spatially-resolved estimates of moisture availability over large areas. ET estimations from remote sensing data are generally based on parameterizations, such as canopy conductance(Gs) using optical vegetation indices. However, optical data presents some limitations related to the low temporal resolution and cloud contamination. Although characterized by coarser spatial resolutions, passive microwave sensors can be useful since they present shorter revisit times and are less affected by clouds and aerosols. In particular, microwave indices are known to be sensitive to vegetation moisture during growing season for forest ecosystems. In this work, we evaluate the performance of passive microwave frequency index (FI) and/or optical vegetation indices (VI) to retrieve ET over different forests under the Penman-Monteith (PM) method. Model results were validated in five FLUXNET sites over USA and Australia over three land covert type including deciduous broadleaf forest (DBF), evergreen needle leaf (ENF) and broadleaf forest (EDF). A subset of Gs values were then regressed against VIs, FI and a combination of FI and VI, and used to parameterize the PM equation for retrievals of ET (PM-Gs). The optical indices calculated from MODIS products were: NDVI, NDWI, and EVI. FI was calculated from AMSR-E passive microwave system. EVI and FI correlated well with Gs (coefficient of determination (R2) >0.5, root mean square error (RMSE)< 45 mm/s for EVI; and R2>0.5, RMSE < 47 mm/s for FI) for DBF. In general optical VI presents similar R2, but less RMSE. For evergreen forests, we found lower or poor relationships between vegetation indices and Gs. Finally, in terms of RMSE the coupled model (FI and EVI) resulted in a lower RMSE of 4-9% compared to independent relations (EVI-Gs and FI-Gs). Interestingly, these three models (PM-Gs) explained 80% of the variance of ET (RSME < 16 W/m2) for deciduous forests.