Browsing by Subject "Multi-model-simulation"

Now showing 1 - 1 of 1

Combining remote sensing and crop modeling techniques to derive a nitrogen fertilizer application strategy
(2020) Röll, Georg; Graeff-Hönninger, Simone
The crucial question in this thesis was how can remote sensing data and crop models be used to derive a N fertilizer strategy that is capable to lower the environmental side effects of N fertilizer application. This raised the following detailed objectives: The first objective (i) how N content determination via spectral reflectance is influenced by different leaves and positions on the leaf was investigated in Publication I. Different wheat plants were cultivated under different N levels and under drought stress in two hydroponic greenhouse trials. Spectral reflectance measurements were taken from three leaves and at three positions on the leaf for each plant. In total, 16 vegetation indices broadly used in the literature were calculated based on the spectral reflectance for each combination of leaf and position. The plant N content was determined by lab analyses. Neither the position on the leaf nor leaf number had an impact on the accuracy of plant N determination via spectral reflectance measurements. Therefore measurements taken at the canopy level seem to be a valid approach. However, if other stress symptoms like drought or disease infection occur, a differentiation between leaves and positions on the leaf might play a more crucial role. Publication II dealt with the second objective on (ii), how to incorporate leaf disease into the DSSAT wheat model to enable the simulation of the impact of leaf disease on yield. An integration of sensor information in crop growth models requires the update of model state variables. A model extension was developed by adding a pest damage module to the existing wheat model. The approach was tested on a two-year dataset from Argentina with different wheat cultivars and on a one-year dataset from Germany with different inoculum levels of septoria tritici blotch (STB). After the integration of disease infection, the accuracy of the simulated yield and leaf area index (LAI) was improved. The Root mean squared error (RMSE) values for yield (1144 kg ha−1) and LAI (1.19 m2 m−2) were reduced by half (499 kg ha−1) for yield and LAI (0.69 m2 m−2). A sensitivity analysis also showed a strong responsiveness of the model by the integration of different STB disease infection scenarios. Increasing the modeling accuracy even further a MM approach seems to be suitable. Assembling more models increases the complexity of the simulation and the involved calibration procedure especially if the user is not familiar with all models. To avoid these conflicts, Publication III evaluated the third objective (iii) if an automatic calibration procedure in a MM approach for winter wheat can eliminate the subjectivity factor in model calibration. The model calibration was performed on a 4-yr N wheat fertilizer trial in southwest Germany. The evaluation mean showed satisfying results for the calibration (d-Index 0.93) and evaluation dataset (d-Index 0.81). This lead to the fourth (iv) objective to use a MM approach to improve the overall modeling accuracy. The evaluation of a fertilizer trial showed an improved modeling accuracy in most cases, especially in the drought season 2018. Based on the combination of a MM approach and the incorporation of sensor data, a Nitrogen Application Prescription System (NAPS) was developed. The initial NAPS setup requires long term recorded data (yield, weather, and soil) to ensure proper MM calibration. After calibration, the current growing season conditions are required (weather, management information) until the N application date. Afterward, the NAPS incorporates remote sensing information and generated weather for running future N application scenarios. The selection of the proper amount of N is determined by economic and ecological criteria. Furthermore, in order to account for differences in in-field variabilities and to deliver a N prescription site-specifically, the NAPS concept has to be applied on a geospatial scale by adjusting soil parameters spatially. The NAPS concept has the potential to adjust the N application more economically and ecologically by using current sensor data, historical yield records, and future weather prediction to derive a more precise N application strategy. Finally, this concept exhibits the potential for reconciliation of the issue of an economic, agricultural production without harming the environment.