Analysis of Spectral Measurements in Paddy Field for Predicting Rice Growth and Yield Based on a Simple Crop Simulation Model

Abstract

Spectral reflectance of differentially-managed rice canopies was measured over an entire growing season and analyzed with special attention to linking remotely sensed information with a simple growth model. The fraction of absorbed photosynthetically active radiation(fAPAR), which is often used as a key variable in simple process models, was well correlated with spectral vegetation indices(VI). VIs, such as NDVI and SAVI, were derived from the ratio of reflectance at two wavelengths(R660 nm and R830 nm)and a new VI, termed the normalized difference ND[R1100 nm, R660 nm], was derived from the difference of R1100 nm and R660 nm divided by their sum. These close relations between fAPAR and VIs were expressed by exponential formulae with different parameters for the periods before and after heading. These indices became less sensitive to fAPAR when fAPAR was larger than 0.4. The use of R1100 nm and R1650 nm with R660 nm and R830 nm in multiple regression significantly improved the prediction accuracy of fAPAR. A close linear relation was found between a spectral ratio R830 nm/R550 nm and leaf nitrogen content during the ripening period although it was not the case before heading. Results suggested that R830 nm/R550 nm was effective for estimation of leaf nitrogen content when the paddy field was regarded as a big leaf. The total amount of leaf nitrogen was well correlated with ND[R1100 nm, R660 nm];nevertheless, the sensitivity was lost when the total amount of leaf nitrogen was greater than 3 g m^-2. Multiple regression analysis showed that a combination of four spectral bands R550 nm, R830nm, R1650 nm and R2200 nm was useful for estimation of the total amount of leaf nitrogen. Remotely-sensed nitrogen variables would be a potential model parameters in a simple model. A real-time recalibration module based on a simplex algorithm was developed and proved effective in linking the remotely-sensed fAPAR with a simple model. This approach was also useful for inferring the physiological parameters such as radiation use efficiency for each rice canopy without destructive sampling. The re-parameterization and/or re-initialization with remotely-sensed information was demonstrated to be a practical and effective approach, especially for operational purposes.