| Summary: | Despite advances in remote sensing–based gross primary productivity (GPP) modeling, the calibration of the Moderate Resolution Imaging Spectroradiometer (MODIS) GPP product (GPP<sub>MOD</sub>) is less well understood over rice–wheat-rotation cropland. To improve the performance of GPP<sub>MOD</sub>, a random forest (RF) machine learning model was constructed and employed over the rice–wheat double-cropping fields of eastern China. The RF-derived GPP (GPP<sub>RF</sub>) agreed well with the eddy covariance (EC)-derived GPP (GPP<sub>EC</sub>), with a coefficient of determination of 0.99 and a root-mean-square error of 0.42 g C m<sup>−2</sup> d<sup>−1</sup>. Therefore, it was deemed reliable to upscale GPP<sub>EC</sub> to regional scales through the RF model. The upscaled cumulative seasonal GPP<sub>RF</sub> was higher for rice (924 g C m<sup>−2</sup>) than that for wheat (532 g C m<sup>−2</sup>). By comparing GPP<sub>MOD</sub> and GPP<sub>EC</sub>, we found that GPP<sub>MOD</sub> performed well during the crop rotation periods but underestimated GPP during the rice/wheat active growth seasons. Furthermore, GPP<sub>MOD</sub> was calibrated by GPP<sub>RF</sub>, and the error range of GPP<sub>MOD</sub> (GPP<sub>RF</sub> minus GPP<sub>MOD</sub>) was found to be 2.5–3.25 g C m<sup>−2</sup> d<sup>−1</sup> for rice and 0.75–1.25 g C m<sup>−2</sup> d<sup>−1</sup> for wheat. Our findings suggest that RF-based GPP products have the potential to be applied in accurately evaluating MODIS-based agroecosystem carbon cycles at regional or even global scales.
|
|---|