Browsing by Person "Zare, Hossein"

Now showing 1 - 1 of 1

Assimilation of leaf area index data from satellite remote sensing to improve the forecasting power of crop models
(2024) Zare, Hossein; Streck, Thilo
In face of the threats to agricultural production arising from climate change, natural hazards, and suboptimal management, there is a need to better understand crop responses to these conditions. Numerous process-based crop models of varying complexity have been developed to estimate crop responses. However, the predictive power of these models is reduced by uncertainties originating from model inputs, weather, parameter estimation, and model structure. This research addresses these uncertainties by integrating observational data, specifically from remote sensing, into the simulations. The focus is on assimilating remotely sensed Leaf Area Index (LAI) into different crop models, acknowledging the diverse scales of input, weather, parameter, and model uncertainty. The dissertation has three primary objectives: i) Examine the assimilation of LAI into a simple crop model (PILOTE) with minimal input requirements. ii) Evaluate the assimilation of remotely sensed LAI into various process-based crop models and their multi-model ensemble. iii) Analyse the assimilation of remotely sensed LAI into the most suitable crop model and assess the impact of model calibration on the results. The presented research study utilized observational data from wheat fields over nine years in two regions of southwest Germany (Kraichgau and Swabian Jura). The Particle Filtering method was employed for data assimilation (DA), and LAI data was derived from Landsat and Sentinel-2 satellites using an empirical radiative transfer model called the Choudhury model. Weather uncertainty was addressed through the use of the MARKSIM downscaled weather generator. In the first phase, the PILOTE model was utilized, and both in-situ and remotely sensed LAI values were assimilated. The results indicate that assimilating remotely sensed LAI considerably improves yield predictions, similar to those based on measured LAI. Weather uncertainty appeared as a major contributor to prediction uncertainty. The contribution of LAI data assimilation to improving the PILOTE model predictions suggest that regional calibration is crucial, as the PILOTE model inherently lacks the capability to account for regional variability. In the second part of this thesis, the focus was shifted to the exclusive assimilation of remotely sensed LAI into different the process-based crop models: CERES, GECROS, and SPASS, as well as the multi-model ensemble consisting of all models. Prior to DA, all models were calibrated using measured data. To address uncertainties related to model inputs, sowing date, nitrogen fertilizer application, soil hydraulic parameters, and weather data were treated as random variables. The findings reveal that weather data, followed by soil parameters, contributes the highest level of uncertainty to the predictions. The uncertainties in weather data consistently led to underestimated yield predictions across all models. Notably, assimilating LAI into the multi-model ensemble emerged as the most effective strategy, producing the most promising yield predictions with reduced bias and uncertainty. Models showed different responses to LAI assimilation, with CERES showing minimal impact of DA, while SPASS and GECROS demonstrated meaningful improvement. The correlations between errors of modeled LAI and yield error were found the key criteria for selecting a model for DA. Models with high correlation like SPASS are more suitable for LAI assimilation. In the third and final phase, the SPASS model was employed to assess the influence of different calibration scenarios on DA. These scenarios were designed based on the availability of data: calibration to yield only, calibration to phenology and yield, calibration to LAI and yield, and calibration to phenology, LAI, and yield. The results show that the assimilation of LAI and weather data remarkably reduce overall uncertainty in crop yield predictions. The findings underscore that scenarios involving the calibration of the model to phenology data consistently yield superior predictions for crop yield. This highlights that, given the set of SPASS model parameters used for winter wheat calibration, additional field-based LAI data does not necessarily enhance the calibration quality. The study also finds that uncertainties associated with weather ensembles exert a more substantial influence compared to those resulting from the calibration process. This underlines the paramount importance of accounting for variations and discrepancies in weather forecasts when evaluating yield uncertainty. In conclusion, this doctoral thesis demonstrates that assimilating remotely sensed LAI into different crop models enhances crop yield prediction and mitigates uncertainties related to inputs, weather, calibration, and model complexity. It underscores the importance of analyzing correlations between assimilated variable errors and yield prediction errors for model and variable selection. Also, although LAI assimilation improves yield prediction, it is not necessary that the crop model is calibrated to LAI. Weather uncertainty is the most influential factor among the various sources of uncertainties. Improving medium-term weather forecasting will lead to significant progress in predicting crop yields.