Abstract
Monitoring forest soil carbon dioxide efflux (FCO2) is important as it contributes significantly to terrestrial ecosystem respiration and is hence a major factor in global carbon cycle. FCO2 monitoring is usually conducted by the use of soil chambers to sample various point positions, but this method is difficult to replicate at spatially large research sites. Satellite remote sensing is accustomed to monitoring environmental phenomenon at large spatial scale, however its utilisation in FCO2 monitoring is under-explored. To this end, this study explored the potential of LANDSAT-8 to estimate FCO2 with the specific aims of deriving land surface temperature (LST) from LANDSAT-8 and then develop FCO2 model on the basis of LANDSAT-8 LST to account for seasonal and inter-annual variations of FCO2. The study was conducted over an old European beech forest (Fagus sylvatica) in Czech Republic. In the end, two kinds of linear mixed effect models were built; Model-1 (inter-annual variations of FCO2) and Model-2 (seasonal variations of FCO2). The difference between Model-1 and Model-2 lies in their random factors; while Model-1 has 'year' of FCO2 measurement as a random factor, Model-2 has 'season' of FCO2 measurement as a random factor. When modelling without random factors, LANDSAT-8 LST as the fixed predictor in both models was able to account for 26% (marginal R-2 = 0.26) of FCO2 variability in Model-1 whereas it accounted for 29% in Model-2. However, the parameterisation of random effects improved the performance of both models. Model-1 was the best in that it explained 65% (conditional R-2 = 0.65) of variability in FCO2 and produced the least deviation from observed FCO2 (RMSE = 0.38 pmol/m(2)/s). This study adds to the limited number of previous similar studies with the aim of encouraging satellite remote sensing integration in FCO2 observation.
NSTL
Elsevier