Satellite Data Assimilation to Improve Regional Crop Yield Estimates

The AquaCrop crop growth model of the Food and Agriculture Organization was recently integrated into NASA's Land Information System Framework. This allows unprecedented crop estimation and satellite data assimilation (DA) experiments at regional scales. Satellite DA aims to combine a model and observations to reduce the uncertainties in crop estimates. This thesis assimilates the Copernicus fraction of vegetation cover (FCOVER) product to update the canopy cover (CC) and biomass, and consequently winter wheat yield estimates, in the Piedmont Region of Italy between 2017 and 2023. After calibrating the crop parameters, testing the model, and developing the DA routines, three model ensembles (modes) were generated by perturbing the model in various ways to estimate the model forecast uncertainty. Forcing data and state variables were perturbed in all modes; mode 1 included perturbation bias correction, mode 2 did not, nor did mode 3, which included an additional variation of parameters. Next, the FCOVER observations for winter wheat were assimilated with an ensemble Kalman filter. The results were compared with other satellite products and field surveys of yield. The ensemble mode 3 with the most degrees of freedom led to the best model-only simulations compared to reference data, and came with the strongest DA updates. The DA improved CC by design and enhanced the model's dry above-ground biomass production; however, yield estimates showed no clear improvement for all ensemble modes. The DA increments to CC were constrained by a potential upper boundary (CCpot), either as a result of exceeding physical boundaries or due to the asynchrony between the model's sowing date and the observations, and this in turn limits the updates to yield. On balance, the most promising DA results were obtained for the ensemble mode 2. Further studies are needed to understand how to address the uncertainty of the planting date or crop stages in general, and investigate the joint assimilation of soil moisture retrievals to overcome the degradation due to vegetation DA updates during the information propagation within the model.