Premium
Remote sensing‐based estimation of annual soil respiration at two contrasting forest sites
Author(s) -
Huang Ni,
Gu Lianhong,
Black T. Andrew,
Wang Li,
Niu Zheng
Publication year - 2015
Publication title -
journal of geophysical research: biogeosciences
Language(s) - English
Resource type - Journals
eISSN - 2169-8961
pISSN - 2169-8953
DOI - 10.1002/2015jg003060
Subject(s) - environmental science , evergreen , moderate resolution imaging spectroradiometer , water content , radiometer , deciduous , atmospheric sciences , data assimilation , spectroradiometer , remote sensing , hydrology (agriculture) , meteorology , geography , ecology , reflectivity , geology , engineering , biology , satellite , geotechnical engineering , aerospace engineering , physics , optics
Soil respiration ( R s ), an important component of the global carbon cycle, can be estimated using remotely sensed data, but the accuracy of this technique has not been thoroughly investigated. In this study, we proposed a methodology for the remote estimation of annual R s at two contrasting FLUXNET forest sites (a deciduous broadleaf forest and an evergreen needleleaf forest). A version of the Akaike's information criterion was used to select the best model from a range of models for annual R s estimation based on the remotely sensed data products from the Moderate Resolution Imaging Spectroradiometer and root‐zone soil moisture product derived from assimilation of the NASA Advanced Microwave Scanning Radiometer soil moisture products and a two‐layer Palmer water balance model. We found that the Arrhenius‐type function based on nighttime land surface temperature (LST‐night) was the best model by comprehensively considering the model explanatory power and model complexity at the Missouri Ozark and BC‐Campbell River 1949 Douglas‐fir sites. In addition, a multicollinearity problem among LST‐night, root‐zone soil moisture, and plant photosynthesis factor was effectively avoided by selecting the LST‐night‐driven model. Cross validation showed that temporal variation in R s was captured by the LST‐night‐driven model with a mean absolute error below 1 µmol CO 2 m −2 s −1 at both forest sites. An obvious overestimation that occurred in 2005 and 2007 at the Missouri Ozark site reduced the evaluation accuracy of cross validation because of summer drought. However, no significant difference was found between the Arrhenius‐type function driven by LST‐night and the function considering LST‐night and root‐zone soil moisture. This finding indicated that the contribution of soil moisture to R s was relatively small at our multiyear data set. To predict intersite R s , maximum leaf area index (LAI max ) was used as an upscaling factor to calibrate the site‐specific reference respiration rates. Independent validation demonstrated that the model incorporating LST‐night and LAI max efficiently predicted the spatial and temporal variabilities of R s . Based on the Arrhenius‐type function using LST‐night as an input parameter, the rates of annual C release from R s were 894–1027 g C m −2 yr −1 at the BC‐Campbell River 1949 Douglas‐fir site and 818–943 g C m −2 yr −1 at the Missouri Ozark site. The ratio between annual R s estimates based on remotely sensed data and the total annual ecosystem respiration from eddy covariance measurements fell within the range reported in previous studies. Our results demonstrated that estimating annual R s based on remote sensing data products was possible at deciduous and evergreen forest sites.