Big data, small explanatory and predictive power: Lessons from random forest modeling of on-farm yield variability and implications for data-driven agronomy.

Context: Collection and analysis of large volumes of on-farm production data are widely seen as key to understanding yield variability among farmers and improving resource-use efficiency. Objective: The aim of this study was to assess the performance of statistical and machine learning methods to explain and predict crop yield across thousands of farmers' fields in contrasting farming systems worldwide. Methods: A large database of 10,940 field-year combinations from three countries in different stages of agricultural intensification was analyzed. Random effects models were used to partition crop yield variability and random forest models were used to explain and predict crop yield within a cross-validation scheme with data re-sampling over space and time. Results: Yield variability in relative terms was smallest for wheat and barley in the Netherlands and for wheat in Ethiopia, intermediate for rice in the Philippines, and greatest for maize in Ethiopia. Random forest models comprising a total of 87 variables explained a maximum of 65 % of cereal yield variability in the Netherlands and less than 45 % of cereal yield variability in Ethiopia and in the Philippines. Crop management related variables were important to explain and predict cereal yields in Ethiopia, while predictive (i.e., known before the growing season) climatic variables and explanatory (i.e., known during or after the growing season) climatic variables were most important to explain and predict cereal yield variability in the Philippines and in the Netherlands, respectively. Finally, model cross-validation for regions or years not seen during model training reduced the R2 considerably for most crop x country combinations, while for wheat in the Netherlands this was model dependent. Conclusion: Big data from farmers' fields is useful to explain on-farm yield variability to some extent, but not to predict it across time and space. Significance: The results call for moderate expectations towards big data and machine learning in agronomic studies, particularly for smallholder farms in the tropics where model performance was poorest independently of the variables considered and the cross-validation scheme used.

Estimating lime requirements for tropical soils: Model comparison and development.

Acid tropical soils may become more productive when treated with agricultural lime, but optimal lime rates have yet to be determined in many tropical regions. In these regions, lime rates can be estimated with lime requirement models based on widely available soil data. We reviewed seven of these models and introduced a new model (LiTAS). We evaluated the models' ability to predict the amount of lime needed to reach a target change in soil chemical properties with data from four soil incubation studies covering 31 soil types. Two foundational models, one targeting acidity saturation and the other targeting base saturation, were more accurate than the five models that were derived from them, while the LiTAS model was the most accurate. The models were used to estimate lime requirements for 303 African soil samples. We found large differences in the estimated lime rates depending on the target soil chemical property of the model. Therefore, an important first step in formulating liming recommendations is to clearly identify the soil property of interest and the target value that needs to be reached. While the LiTAS model can be useful for strategic research, more information on acidity-related problems other than aluminum toxicity is needed to comprehensively assess the benefits of liming.

The legacy effect of synthetic N fertiliser.

Cumulative crop recovery of synthetic fertiliser nitrogen (N) over several cropping seasons (legacy effect) generally receives limited attention. The increment in crop N uptake after the first-season uptake from fertiliser can be expressed as a fraction (∆RE) of the annual N application rate. This study aims to quantify ∆RE using data from nine long-term experiments (LTEs). As such, ∆RE is the difference between first season (RE1st) and long-term (RELT) recovery of synthetic fertiliser N. In this study, RE1st was assessed either by the 15N isotope method or by a zero-N subplot freshly superimposed on a long-term fertilised LTE treatment plot. RELT was calculated by comparing N uptake in the total aboveground crop biomass between a long-term fertilised and long-term control (zero-N) treatment. Using a mixed linear effect model, the effects of climate, crop type, experiment duration, average N rate, and soil clay content on ∆RE were evaluated. Because the experimental setup required for the calculation of ∆RE is relatively rare, only nine suitable LTEs were found. Across these nine LTEs in Europe and North America, the mean ∆RE was 24.4% (±12.0%, 95% CI) of annual N application, with higher values for winter wheat than for maize. This result shows that fertiliser-N retained in the soil and stubble may contribute substantially to crop N uptake in subsequent years. Our results suggest that an initial recovery of 43.8% (±11%, 95% CI) of N application may increase to around 66.0% (±15%, 95% CI) on average over time. Furthermore, we found that ∆RE was not clearly related to long-term changes in topsoil total N stock. Our findings show that the-often used-first-year recovery of synthetic fertiliser N application does not express the full effect of fertiliser application on crop nutrition. The fertiliser contribution to soil N supply should be accounted for when exploring future scenarios on N cycling, including crop N requirements and N balance schemes. Highlights: Nine long-term cereal experiments in Europe and USA were analysed for long-term crop N recovery of synthetic N fertiliser.On average, and with application rates between 34 and 269 kg N/ha, crop N recovery increased from 43.8% in the first season to 66.0% in the long term.Delta recovery was larger for winter wheat than maize.Observed increases in crop N uptake were not explained by proportionate increases in topsoil total N stock.

Risk aversion, impatience, and adoption of conservation agriculture practices among smallholders in Zambia.

Sustainable agricultural practices such as conservation agriculture have been promoted in southern Africa for nearly three decades, but their adoption remains low. It is of policy interest to unpack behavioural drivers of adoption to understand why adoption remains lower than anticipated. This paper assesses the effects of risk aversion and impatience on the extent and intensity of the adoption of conservation agriculture using panel data collected from 646 households in 2021 and 2022 in Zambia. We find that 12% and 18% of the smallholders were impatient and risk averse, respectively. There are two main empirical findings based on panel data Probit and Tobit models. First, on the extensive margin, being impatient is correlated with a decreased likelihood of adopting combined minimum-tillage (MT) and rotation by 2.9 percentage points and being risk averse is associated with a decreased propensity of adopting combined minimum tillage (MT) and mulching by 3.2 percentage points. Being risk averse is correlated with a decreased chance of adopting basins by 2.8 percentage points. Second, on the intensive margin, impatience and risk aversion are significantly correlated with reduced adoption intensity of basins, ripping, minimum tillage (MT), and combined MT and rotation by 0.02-0.22 ha. These findings imply a need to embed risk management (e.g., through crop yield insurance) in the scaling of sustainable agricultural practices to incentivise adoption. This can help to nudge initial adoption and to protect farmers from yield penalties that are common in experimentation stages.

Pathways to wheat self-sufficiency in Africa.

A growing urban population and dietary changes increased wheat import bills in Africa to 9% per year. Though wheat production in the continent has been increasing over the past decades, to varying degrees depending on regions, this has not been commensurate with the rapidly increasing demand for wheat. Analyses of wheat yield gaps show that there is ample opportunity to increase wheat production in Africa through improved genetics and agronomic practices. Doing so would reduce import dependency and increase wheat self-sufficiency at national level in many African countries. In view of the uncertainties revealed by the global COVID-19 pandemic, extreme weather events, and world security issues, national policies in Africa should re-consider the value of self-sufficiency in production of staple food crops, specifically wheat. This is particularly so for areas where water-limited wheat yield gaps can be narrowed through intensification on existing cropland and judicious expansion of rainfed and irrigated wheat areas. Increasing the production of other sources of calories (and proteins) should also be considered to reduce dependency on wheat imports.

Balancing indicators for sustainable intensification of crop production at field and river basin levels.

Adequate tools for evaluating sustainable intensification (SI) of crop production for agro-hydrological system are not readily available. Building on existing concepts, we propose a framework for evaluating SI at the field and river basin levels. The framework serves as a means to assess and visualise SI indicator values, including yield, water-use efficiency and nitrogen-use efficiency (NUE), alongside water and nitrogen surpluses and their effects on water quantity and quality. To demonstrate the SI assessment framework, we used empirical data for both the field level (the Static Fertilization Experiment at Bad Lauchstädt) and the river basin level (the Selke basin, 463 km2) in central Germany. Crop yield and resource use efficiency varied considerably from 1980 to 2014, but without clear trends. NUE frequently fell below the desirable range (<50%), exposing the environment to a large N surplus (>80 kg N ha-1). For the catchment as a whole, the average nitrate-N concentration (3.6 mg L-1) was slightly higher than the threshold of 2.5 mg L-1 nitrate-N in surface water. However, weather and climate-related patterns, due to their effects on transport capacity and dilution, influenced water quantity and quality indicators more than agronomic practices. To achieve SI of crop production in the Selke basin, irrigation and soil moisture management are required to reduce yield variability and reduce N surpluses at field level. In addition, optimum application of fertiliser and manure could help to reduce the nitrate-N concentration below the set water quality standards in the Selke basin. In this way, there is scope for increase in yields and resource use efficiencies, and thus potential reduction of environmental impacts at basin level. We conclude that the framework is useful for assessing sustainable production, by simultaneously considering objectives related to crop production, resource-use efficiency and environmental quality, at both field and river basin levels.