Particle swarm optimization for assimilation of remote sensing data in dynamic crop models

Agricultural monitoring is of growing importance due to an increasing world population, slowing growth of agricultural output and concerns regarding food security. Remote sensing and dynamic crop modeling are powerful tools for yield prediction and frequently applied in literature. A large question arising in this context is the assimilation of remote sensing data into the model process. We present a novel technique employing Particle Swarm Optimization in an updating scheme flexibly incorporating different sources of uncertainty in both the model simulation and remote sensing observations. We tested the technique with the AquaCrop-OS model for winter wheat yield prediction by updating canopy cover obtained from remote sensing datasets. Preliminary results showed that the new method can outperform both a simple replacement update and a Kalman filter approach. It succeeded in removing the bias from field-level yield predictions and reducing the RMSE from 1.32 t/ha to 0.89 t/ha.