| Popis: |
Process-based grass models (PBGMs) are widely used for predicting grass growth under potential climate change and different management practices. However, accurate predictions using PBGMs heavily rely on field observations for data assimilation. In data-limited areas, performing robust and reliable estimates of grass growth remains a challenge. In this paper, we incorporated satellite-based MODIS data products, including leaf area index, gross primary production and evapotranspiration, as an additional supplement to field observations. Popular data assimilation methods, including Bayesian calibration and the updating method ensemble Kalman filter, were applied to assimilate satellite derived information into the BASic GRAssland model (BASGRA). A range of different combinations of data assimilating methods and data availability were tested across four grassland sites in Norway, Finland and Canada to assess the corresponding accuracy and make recommendations regarding suitable approaches to incorporate MODIS data. The results demonstrated that optimizing the model parameters that are specific for grass species and cultivar should be targeted prior to updating model state variables. The MODIS derived data products were capable of constraining model’s simulations on phenological development and biomass accumulation by parameter optimization with its performance exceeding model outputs driven by default parameters. By integrating even a small number of field measurements into the parameter calibration, the model’s predictive accuracy was further improved - especially at sites with obvious biases in the input MODIS data. Overall, this comparative study has provided flexible solutions with the potential to strengthen the capacity of PBGMs for grass growth estimation in practical applications. |
Full Text from ScienceDirect