Exploring the agricultural parameter space for crop yield and sustainability

Abstract

Achieving global agricultural sustainability while maximizing yield and yield quality is a key issue for which we should consider all options. We here show that agricultural ‘classical’ management options are likely underexplored: we estimate that the most intensely researched sites cover only a fraction of a percent of available options. Since some of these untested options may prove important for yield, yield quality and sustainability, we suggest the use of distributed trials, exploring a broader range of options; this should occur in parallel to employing new technologies. Agroecosystem management relies on a rather limited suite of management factors, mainly fertilization regime and application of pesticides, tillage, and crop rotation. Given the formidable challenge to achieve agricultural sustainability, while feeding a growing number of people, new approaches have been proposed, and are currently being actively explored. These include microbiome engineering (Mueller & Sachs, 2015) or core microbiome inoculation (Toju et al., 2018), and the use of advanced technological options (drones, robotics) among others (Walter et al., 2017). However, have we fully explored the ‘classical’ options? Each of these few major management factors mentioned earlier may at first sight seem monolithic, but upon closer scrutiny it is clear that each factor harbors a bewildering range of treatment levels. Importantly, these factor levels can essentially be linked in a virtually combinatorial fashion, with perhaps only a few combinations excluded on first principles (e.g. nitrogen (N)-fixing plants and certain levels of N fertilization). This means there is actually a rather wide range of factor combinations. We provide an estimate of possible treatment combinations by reviewing the existing literature. Specifically, for each of the major management practices (tillage, crop rotation and fertilization) we identified meta-analyses or synthesis papers, which we used to subdivide these practices into different categories and then specific levels (Fig. 1; Table 1; for more detail also see Supporting Information Table S1). We thus estimate over 285 000 different management combinations. This calculation did not consider cover crops or other aspects of agroecosystem diversification, and only coarsely represented pesticide use. We also did not consider crop-specific chemical fertilization rates other than recommended, and higher or lower than recommended (e.g. we did not include fertigation). We also did not consider factorial combinations of levels within categories. We thus believe that this estimate is rather conservative. However, some of the combinations are perhaps to be excluded from first principles or because they are locally not feasible (e.g. because of unavailability of technical means). Either way, it is clear that this number is going to be in the hundreds of thousands. The agriculture we see today is the result of perhaps 10 000 years of trial and error and continuous learning. Nevertheless, it is also clear that there is a lot of room for improvement in terms of optimizing sustainability and yield, and that the scientific method is required to establish such improvement (Cui et al., 2018). Therefore, it is helpful to ask: how many of the factor combinations we estimated earlier have actually been formally scientifically tested at any one site or region? To ascertain this, we turned to perhaps one of the best-researched agroecosystems, Rothamsted, site of the longest-running agricultural experiments (Rothamsted Research, 2006). Rothamsted has overall tested 314 treatment levels, which we calculated as the sum of Broadbalk, Park Grass and Hoosfield experiments (see Table S2). Most agricultural experiment stations cover a lot fewer treatment combinations, especially in terms of long-term experiments necessary to ascertain key sustainability parameters. Thus, we believe this number of over 300 to be an absolute maximum estimate. This mean