Estimating Ecosystem Respiration Using Satellite Remote Sensing

SUMMARY Here we report a method for estimating ecosystem respiration (R e) by exclusive use of remotely sensed land surface temperature (T r) from NASA's moderate resolution imaging spectroradiometer (MODIS) sensors. A combination of daytime and nighttime MODIS T r data was used to model an 8-day mean value of land surface temperature (T r(mean)), which was then used in a modified Arrhenius-based Lloyd and Taylor [1994] respiration model to simulate 7 years of ecosystem respiration of a temperate deciduous forest. These model outputs were then compared with respiration estimates of the site measured by eddy covariance (EC) method. Measured and modeled values were highly correlated (R 2 = 0.87, RMSE = 0.64 — mol CO 2 m-2 s-1). This study demonstrates the potential of estimating " per-pixel " ecosystem respiration by exclusive use of remotely sensed data from the existing space-based sensors. Conventionally, temperature sensitivity of ecosystem respiration (R e) is expressed by an exponential function of soil temperature or air temperature (T s and T a respectively) known as Q 10 (the factor by which respiration rate increases with every 10 o C increment of temperature). Recent studies showed that, for the forested flux-tower sites in the USA, a reasonably strong Q 10 based exponential relationship (R 2 = 0.6) exists between the16-day average R e and 16-day composite values of radiometric land surface temperature (T r) from NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) sensor [1]. But the principal limitations of Q 10 based R e estimation method are its constant temperature response relationship to respiration, and its site specificity [2]. In this paper we examine whether a more robust physiological based modeling framework, originally proposed for soil respiration only, can be utilized to estimate R e from MODIS T r. Our study areas was Morgan Monroe State Forest (MMSF) in south central Indiana (39.3232 o N, 86.4131 o W). For this study, we used eddy covariance (EC) estimates of R e from the flux tower at MMSF, and T r data of the flux tower footprint from MODIS. The EC system was located at the top of the 46 m tall tower and consisted of a three-dimensional sonic anemometer (Sampling frequency was 10 Hz with fluxes calculated hourly. The flux values (F C) were then subjected to quality control, including outlier rejection, and a u * ” 0.3 m s-1 (u* is the friction velocity) criterion to …