Integrating crop growth simulation and remote sensing to improve resource use efficiency in farming systems

This study investigated the scope and constraints for integrated use of mechanistic crop growth simulation models and earth observation techniques. Integration of highquality crop growth models and information derived from earth observations can contribute to improved use of resources, reduced crop production risks, reduced environmental degradation, and increased farm income. In the past, both, simulation modelling and remote sensing have been shown to be valuable tools in separate applications in agriculture. Crop growth simulation has made valuable contributions to yield forecasting, proto-typing crop varieties, generation of input-output coefficients for improved agricultural production technologies and to management decision support systems at field level. Likewise, remote sensing techniques have been successfully applied in classification of arable crops and in quantification of vegetation characteristics at different spatial and temporal scales. The starting point of this study was the hypothesis that integration of both techniques would lead to improvements in the dynamic simulation of the crop-soil system and thus contribute to improvements in management decision support systems for environmentally sound agricultural production. Thus far, mutually beneficial linkages have been limited to land use classification via remote sensing (choice of adequate model) and quantification of crop growth and development curves using e.g. estimates of leaf area indices derived from remote sensing images for model calibration under (usually) favourable growth conditions. Only a few studies have considered the potentials of remote sensing for model initialization of growth and development characteristics of a specific crop. In this thesis these potentials have been extended to a more continuous approach, in which remote sensing information is not only used in model initialization, but also in model calibration in the course of the simulation run, so-called run-time calibration. During such a run-time calibration procedure, simulated values of e.g. leaf area index (LAI) and canopy nitrogen status (CNS) are replaced by values estimated from remote sensing images acquired at different stages in the course of the growing period. LAI and CNS are important controlling variables in models for arable crops such as wheat, potato and maize. This run-time calibration procedure has been performed for a full crop growth cycle, for optimal as well as sub-optimal growth conditions. This approach enables spatial differentiation in crop growth simulation, as variations in crop status, resulting from differences in growth conditions, lead to differences in remote sensing signals. The relationships between near and remote sensing observations at leaf, plant and canopy level have been investigated and the effects of variations in estimated values of LAI and CNS used in run-time calibration of dynamic crop growth simulation models on final model results (e.g. crop yield) have been analyzed. Results from potato trials in the Netherlands show that leaf nitrogen contents derived from near sensing observations can be up-scaled to plant and canopy nitrogen status by taking into account the vertical nitrogen distribution in the crop. A vertical nitrogen extinction coefficient (kN) of 0.41 resulted in an accuracy increase of the relation between leaf nitrogen (g N m-2 leaf) and SPAD readings (a near sensing technique at leaf level), with a correlation coefficient (r2) of 0.91. Remote sensing observations integrate nitrogen contents over canopy depth and do not require adjustment for vertical nitrogen gradients, if canopy nitrogen status is expressed in total nitrogen content per unit of soil surface. The red edge position (an index derived from remote sensing observations) could be related to canopy nitrogen content (g N m-2 soil) with a correlation coefficient (r2) of 0.82. Leaf area indices of potato (Netherlands) and maize (Argentina, France, USA) crops, for use in run-time calibration, were also accurately derived from field, airborne and spaceborne remote sensing platforms. Introducing LAI values derived from RS in the simulation model and concurrently adjusting CNS by retaining leaf N-concentrations, led to more accurate simulation results for CNS than without such adjustment. The different crops, and the range in environmental conditions, soil fertility status and management practices that were examined in the different case-studies in this thesis, have demonstrated the broad applicability of mechanistic simulation models integrated with remote sensing information Winter wheat fields, wheat phenological stages (emergence, flowering) and management operations (harvest) were successfully identified on the basis of information from optical and radar remote sensing data in a case-study in Southeastern France. Timing of these phenological stages and management operations is important in model calibration as they mark the length of the crop growth period and of the grain-filling period, which are co-determinants of grain yield. At flowering, Cband radar backscatter from the soil is maximally reduced by canopy moisture content. This characteristic was successfully used to estimate regional wheat flowering dates. Integration of RS data in the (point-based) crop growth simulation model allowed its spatial application for prediction of wheat production at regional scale. The estimated value was in agreement with regional yield statistics. This integration thus allows expansion of the application area of valuable research tools, as up-scaling has become feasible. Introduction of remote sensing-based estimates of LAI and CNS in the course of the growing seasons into dynamic simulation of the growth of potato and maize resulted in improved simulation accuracy for aerial crop characteristics, as well as for variables that could not be directly observed by remote sensing, such as soil inorganic nitrogen contents. The degree of success and robustness of the integrated approach depends on the timing, accuracy and number of remote sensing observations available for resetting the relevant state variables in the course of the simulation period. Simulation accuracy was positively correlated with the number of observation dates from remote sensing. Remote sensing observations around flowering had more impact on calculated final grain yield (FGY) for maize than earlier or later observations. The investigations reported in this thesis have shown that the accuracy of predictions of dynamic and mechanistic crop growth simulation models significantly improves through integrating earth observation-derived information as input for the models and for their run-time calibration. Such integration not only yields more accurate estimates of crop bio-physical variables, such as leaf area index and canopy nitrogen status, but also contributes to improved prediction at regional scales. Such models, producing reliable, site-specific predictions of crop performance and crop requirements are thus effective tools in the development of environmentally-friendly production methods and in optimizing the use of our natural resources. Further research should focus on the scope for estimating additional crop variables of interest for integration in simulation modelling through remote sensing. Management interventions may be triggered by various crop characteristics, such as: 1) canopy temperatures derived from thermal remote sensing systems as an indicator for water stress, 2) canopy discolouring derived from optical remote sensing systems as an indicator for nutrient shortages and 3) canopy architecture derived from radar remote sensing images as an indicator for water and nutrient supply. Remote sensing is also a valuable technique to identify spatial patterns of crop performance and crop status within arable fields. Moreover, remote sensing allows identification of patterns that may be related to specific diseases or special events, such as outbreaks of phytophtera in potato, or lodging in grain crops. This study has demonstrated that a decision support system for crop and soil management based on the integration of crop growth simulation modelling and remotely sensed data is within reach. In addition, nitrogen uptake, its vertical distribution within the crop, and the inorganic nitrogen content of the soil can be simulated more accurately with such an integrated system. Such a decision support system can be used for fine-tuning of fertilizer regimes thus contributing to more environmentally sound and sustained agricultural production.