Quantifying Surface Energy Flux Estimation Uncertainty Using Land Surface Temperature Observations

Abstract:

Remote sensing with thermal infrared is widely recognized as good way to estimate surface heat fluxes, map crop water use, and detect water-stressed vegetation. When combined with net radiation and soil heat flux data, observations of sensible heat fluxes derived from surface temperatures (LST) are indicative of instantaneous evapotranspiration (ET). There are, however, substantial reasons LST data may not provide the best way to estimate of ET. For example, it is well known that observations and models of LST, air temperature, or estimates of transport resistances may be so inaccurate that physically based model nevertheless yield non-meaningful results. Furthermore, using visible and near infrared remote sensing observations collected at the same time as LST often yield physically plausible results because they are constrained by less dynamic surface conditions such as green fractional cover. Although sensitivity studies exist that help identify likely sources of error and uncertainty, ET studies typically do not provide a way to assess the relative importance of modeling ET with and without LST inputs. To better quantify model benefits and degradations due to LST observational inaccuracies, a Bayesian uncertainty study was undertaken using data collected in remote sensing experiments at Maricopa, Arizona. Visible, near infrared and thermal infrared data were obtained from an airborne platform. The prior probability distribution of ET estimates were modeled using fractional cover, local weather data and a Penman-Monteith mode, while the likelihood of LST data was modeled from a two-source energy balance model. Thus the posterior probabilities of ET represented the value added by using LST data. Results from an ET study over cotton grown in 2014 and 2015 showed significantly reduced ET confidence intervals when LST data were incorporated.