The TerrA-P project: towards a global monitoring system for terrestrial primary production

Most land surface models (LSM) require, inter alia, inputs of temperature and moisture to generate predictions of gross primary production (GPP).  But air temperature measured at an arbitrary height in (or above) the canopy may offer only a poor estimate of leaf temperature.  Differences between leaf and air temperature have been shown to vary temporally and spatially and, due to depressed transpiration, may be especially pronounced under conditions of low soil moisture availability.  The Sentinel-3 satellite program offers modellers estimates of the land surface temperature (LST) which for vegetated pixels can be adopted as the canopy (or leaf) temperature.  But retrieving plant-available moisture remains problematic and to date remote-sensing tools lack the ability to penetrate beyond the upper soil-layer to the root zone.  Could remotely-sensed estimates of LST offer a parsimonious LSM input by uniting information on leaf temperature and hydration - avoiding the need for explicit modelling of soil moisture effects?  In a modelling experiment, we hypothesised that agreement with flux-derived GPP estimates would be stronger for simulations forced with LST versus gridded meteorological air temperature and superior performance would be most evident in dry summers.

Using a first-principles, process-based, light use efficiency model (the P-model) that requires only a handful of input variables (not including soil moisture), we generated alternate GPP simulations for comparison with eddy-covariance inferred estimates available from flux sites within the Integrated Carbon Observation System.  Remotely-sensed temperature and greenness (the fraction of photosynthetically active radiation absorbed by vegetation, fAPAR) data were input from Sentinel-3 sources.  Pre-processing steps included interpolation and smoothing before averaging to ten-day timesteps.  Gridded air temperature data were obtained from the European Centre for Medium-Range Weather Forecasts.  We chose the years 2018-2019 to exploit the natural experiment of a pronounced European drought.  For each site, timesteps were assigned a drought index (Standardised Precipitation-Evapotranspiration Index) using a 30-year time-series of climatic water balance.  Unusually dry conditions, for a given site, were characterised as those having SPEI < -1.5.

Overall, simulated GPP showed good agreement with flux-derived estimates, but the experimental effect on simulated GPP was modest and the hypothesis found only partial, biome-dependent support.  During dry conditions, simulations forced with LST performed better than those with air temperature for shrubland, grassland and savannah sites.  For certain sites, we found pronounced early-season deltas with simulations consistently exceeding flux-derived GPP.  That finding was not general to whole biomes or both years.  We speculate that these deltas arose, in part at least, from fAPAR values inflated by neighbouring vegetation not incorporated in the flux-tower’s footprint. 

This study advances the prospect for LSMs that will require as few parameters as possible and rely, as far as is practical, on remotely-sensed input data.  In subsequent steps, we envisage further experiments to assess (i) the desirability of adopting a seasonally weighted diurnal average LST versus the single morning overpass employed here and (ii) whether the Sentinel-3 LST pixel values can be usefully disaggregated to distinguish vegetation and bare ground components.