Impact of Using Different Remotely Sensed Variables on Wheat Grain Yield and Nitrogen estimation with the DSSAT-CERES-Wheat Model in a data assimilation scenario

Abstract

The combination of remote sensing and crop growth models has become an effective tool for yield estimation and a potential method for grain quality estimation. The study evaluated whether the calibration of the DSSAT model, with seasonal data of Leaf Area Index (LAID), and/or crop aboveground biomass (CWAD, kg[dw] ha -1 ) with or without canopy nitrogen (N) content (CNAD, kg[dw] ha -1 ), assimilation variables that could be derived from satellite data, would lead to an improvement in the estimation of wheat grain production (HWAD, kg[dw] ha -1 ) and N content in the grain (HN%D, %), through a procedure that involved the generation and use of synthetic data. This experiment's initial conditions and parameters were established using soil and weather data collected from a field experiment conducted in Rieti during the 2022-2023 wheat season as part of the “Pris4veg: Development of algorithms for the retrieval of plant functional traits from PRISMA data in agricultural and forest ecosystems” project. Additionally, eleven different simulation treatments were created, varying N fertilisation rates ranging from 0 to 300 kg ha -1 by 30 kg ha -1 increments. The experiment used simulated data from different fertilization rates, as observed data, to calibrate the model under a 150 kg N ha -1 fertilisation rate. This procedure was repeated using, as observed data: i) LAID, ii) LAID+CNAD, iii) CWAD, iv) CWAD+CNAD, and v) LAID+CWAD+CNAD. The comparison of each calibrated second run with the first, defined as the “truth”, was performed to assess which among LAID and CWAD variables, alone or in combination with CNAD or all of them together, was more effective in calibrating the DSSAT model for wheat grain yield and protein estimation. Moreover, the procedure was conducted under different meteorological and soil conditions, selecting three crop seasons: 2011-2012, the wet season; 2014-2015, median season; and 2022-2023, the wet season, based on different precipitation conditions over a period of ten years. Regarding soil conditions, we changed the soil texture composition from the actual field soil condition. Finally, nine calibration procedures were performed to cover different combinations of meteorological conditions and soil types. After the model calibration was performed using the different sets of synthetic observed dynamic variables, the goodness of fit between simulated and observed grain yield and nitrogen values was determined by computing the root mean square error (RMSE) for each meteorological and soil scenario combination. The study revealed that the calibration of the DSSAT model using seasonal observations of crop biophysical variables, which can be retrieved from remote sensing in a data assimilation scenario, does indeed improve the estimation of wheat grain yield and N content as compared to the baseline open-loop DSSAT simulation. Moreover, it showed that, in most cases, the use of biomass and canopy N content observations, which are variables typically retrieved from hyperspectral sensors such as PRISMA, improve the estimation accuracy, in addition to the use of only LAI observations, which are typically available from multispectral sensors such as Sentinel-2. We intend to conduct further investigation into this aspect by utilizing measurement data.