Temporal interpolation of land surface fluxes derived from remote sensing – results with an unmanned aerial system

Abstract

Abstract. Remote sensing imagery can provide snapshots of rapidly changing land surface variables, e.g. evapotranspiration (ET), land surface temperature (Ts), net radiation (Rn), soil moisture (θ), and gross primary productivity (GPP), for the time of sensor overpass. However, discontinuous data acquisitions limit the applicability of remote sensing for water resources and ecosystem management. Methods to interpolate between remote sensing snapshot data and to upscale them from an instantaneous to a daily timescale are needed. We developed a dynamic soil–vegetation–atmosphere transfer model to interpolate land surface state variables that change rapidly between remote sensing observations. The “Soil–Vegetation, Energy, water, and CO2 traNsfer” (SVEN) model, which combines the snapshot version of the remote sensing Priestley–Taylor Jet Propulsion Laboratory ET model and light use efficiency GPP models, now incorporates a dynamic component for the ground heat flux based on the “force-restore” method and a water balance “bucket” model to estimate θ and canopy wetness at a half-hourly time step. A case study was conducted to demonstrate the method using optical and thermal data from an unmanned aerial system at a willow plantation flux site (Risoe, Denmark). Based on model parameter calibration with the snapshots of land surface variables at the time of flight, SVEN interpolated UAS-based snapshots to continuous records of Ts, Rn, θ, ET, and GPP for the 2016 growing season with forcing from continuous climatic data and the normalized difference vegetation index (NDVI). Validation with eddy covariance and other in situ observations indicates that SVEN can estimate daily land surface fluxes between remote sensing acquisitions with normalized root mean square deviations of the simulated daily Ts, Rn, θ, LE, and GPP of 11.77 %, 6.65 %, 19.53 %, 14.77 %, and 12.97 % respectively. In this deciduous tree plantation, this study demonstrates that temporally sparse optical and thermal remote sensing observations can be used to calibrate soil and vegetation parameters of a simple land surface modelling scheme to estimate “low-persistence” or rapidly changing land surface variables with the use of few forcing variables. This approach can also be applied with remotely-sensed data from other platforms to fill temporal gaps, e.g. cloud-induced data gaps in satellite observations.