Remote Sensing Data Assimilation by Forcing Method in Simulation of Silage Maize Yield Using AquaCrop Model

An essential part of agricultural plans for maintaining and developing performance at the regional scale is the timely and accurate estimation and prediction of crop yield prior to harvesting using crop growth models. Modeling dynamic changes during crop growth helps researchers to plan crop management strategies to improve its yield. These models contain several parameters that should be calibrated according to the characteristics of the study area. Lack of spatial/geographic components in these models and parameter uncertainties lead to errors in the estimated outputs. Remote sensing data assimilation can be useful for solving this problem and evaluating the spatial variability in the lands, especially at the regional scale. Remote sensing can estimate the values of input variables of crop growth models such as the Leaf Area Index (LAI), canopy cover, biomass, and soil characteristics.

To achieve accurate crop yield, it is possible to use crop growth models. To this end, the AquaCrop model parameters were estimated and the model was calibrated with measuring and sampling different requied information of model in the crop growing stages and prior to cultivation over agricultural silage maize fields at the regional scale. To calibrate the Aquacrop simulation model through assimilation of remote sensing (RS) data, fCover biophysical variable was extracted from pixel-based RS data by developing GPR-PSO algorithm. Besides, to simplify the Aquacrop model, and to identify more sensitive parameters, the combined sensitivity analysis Morris and EFAST algorithms were employed. Finally, by assimilating the biophysical variable extracted by RS into the Aquacrop model, these more effective parameters were estimated through the forcing method, and compared the results with the results of no application of RS data. In order to calibrate the Aquacrop model, field sampling of soil (before planting) and crop during the growing season of silage maize, digital hemispherical photography (DHP) as well as measurement by destructive method for comparison, was performed in the fields of Qhale-Nou county located in the south of Tehran, in the summer of 2019.

The results of assimilation of RS data in Aquacrop model compared to no application of RS data in this model showed that considering data assimilation of RS data leads to the increase in model calibration accuracy. As the results suggest, the output yield for the model with data assimilation was estimated with R2 values of 0.89 and 0.88 for calibration and evaluation, respectively. The superiority of RS data assimilation into the model as opposed to not its incorporating was also verified by improving the accuracy with increases in R2 values by 0.14 and 0.15 and decrese in Relative RMSE (RRMSE) values of 4.12 and 5.17 percent and RMSE of 2.5 and 2.4 ton/ha for calibration and evaluation, respectively. So, compared to RS data assimilation and without assimilation is associated with improving the model calibration process with RS data assimilation.

The present study employed estimated fCover values obtained via RS data as observed state variables fed as input to the AquaCrop model for means of estimating the most effective parameters identified (via sensitivity analysis). The findings of this procedure indicate that RS data assimilation as a forcing method for model parameters estimating can increase the overall accuracy of the model. Moreover, the correlation between simulated and observed values was higher for the case of RS data assimilation as opposed to not incorporating such data. As these results suggest, RS data assimilation into the AquaCrop model can prove more successful and attain higher accuracies as opposed to not incorporating such data. Furthermore, this process of data assimilation can be used for estimating biophysical variables and calibrating crop growth models at the regional scale, with less time complexity and lower costs and more updated information.