Multi-objective assessment of three remote sensing vegetation products for streamflow prediction in a conceptual ecohydrological model

Abstract

This study assesses the implications of using three alternate remote sensing vegetation products in the simulation of streamflow using a conceptual ecohydrologic model. Vegetation is represented as a dynamic component in this model which simulates two response variables, streamflow and one of the following three vegetation attributes: Gross Primary Productivity (GPP), Leaf Area Index (LAI) or Vegetation Optical Depth (VOD). Model simulations are performed across 50 catchments with areas ranging between 50 and 1600km2 in the Murray-Darling Basin in Australia. Moderate Resolution Imaging Spectroradiometer (MODIS) LAI and GPP products, passive microwave observations of VOD and streamflow are used for model calibration and/or validation. Single-objective model calibration based on one of the vegetation products (GPP, LAI and VOD) shows that GPP is the best vegetation simulating product. On the contrary, LAI produces the best streamflow during validation when the optimized parameters are applied for streamflow estimation. To obtain the best compromise solution for simultaneous simulation of streamflow and a vegetation product, a multi-objective optimization is applied on GPP and streamflow, VOD and streamflow and LAI and streamflow. Results show that LAI and then VOD are the two best products in simulating streamflow across these catchments. Improved simulation of VOD and LAI in a multi-objective setting is partly related to the higher temporal resolution of these datasets and inclusion of processes for converting GPP to net primary productivity and biomass. It is suggested that further development of these remote sensing products at finer spatial and temporal resolutions may lead to improved streamflow prediction, as well as a better simulation capability of the ecohydrological system being modeled.