Machine learning for optimizing complex site-specific management

Abstract

Despite the promise of precision agriculture for increasing the productivity by implementing site-specific management, farmers remain skeptical and its utilization rate is lower than expected. A major cause is a lack of concrete approaches to higher profitability. When involving many variables in both controlled management and monitored environment, optimal site-specific management for such high-dimensional cropping systems is considerably more complex than the traditional low-dimensional cases widely studied in the existing literature, calling for a paradigm shift in optimization of site-specific management. We propose an algorithmic approach that enables farmers to efficiently learn their own site-specific management through on-farm experiments. We test its performance in two simulated scenarios-one of medium complexity with 150 management variables and one of high complexity with 864 management variables. Results show that, relative to uniform management, site-specific management learned from 5-year experiments generates $43/ha higher profits with 25 kg/ha less nitrogen fertilizer in the first scenario and $40/ha higher profits with 55 kg/ha less nitrogen fertilizer in the second scenario. Thus, complex site-specific management can be learned efficiently and be more profitable and environmentally sustainable than uniform management.

Nitrogen and the future of agriculture: 20 years on

“…Traditional, replicated field studies with two or three treatment variables are not up to the task of optimizing the 10-20 management factors (including selection of the optimal hybrid or cultivar) that affect NFE, crop yield, profit, and other key performance indicators. Machine learning and other artificial intelligence approaches can play an important role in developing optimized, tailored, and site-specific management solutions (Saikai et al 2020), particularly once it becomes possible to move seamlessly from data to prescriptive analytics and automated decision making (Smith 2020). The rapid spread of GPS-enabled smartphones presents a unique opportunity for directly reaching 2 billion people in smallholder farming households, provided the smartphone apps and messages are scientifically sound, give actionable advice, and utilize feedback mechanisms to enable rigorous testing and continuous improvement (Fabregas et al 2019).…”

Achieving the nutrient reduction objective of the Farm to Fork Strategy. An assessment of CAP subsidies for precision fertilization and sustainable agricultural practices in Germany

“…Lastly, research on machine learning develops algorithms which enable site-adapted management of land, including fertilization, by considering many environmental and managerial factors. An experiment finds that one algorithm was able to reduce N input by about one quarter after five learning years (Saikai et al, 2020). Experiments, models and algorithms highlight that nutrient losses into the environment can be reduced through digital precision fertilization.…”

“…For example, algorithms can be guided by different objectives such as reducing fertilizer input, minimizing environmental impacts or improving fertilization profitability. The latter objective demands maximum yields in relation to financial input which is likely to induce negative environmental impacts (see e.g., Saikai et al, 2020;Tzachor et al, 2022). Therefore, to contribute to the nutrient objective of the Farm to Fork Strategy, digital precision fertilization must clearly aim at fertilizer reduction and nutrient loss minimization.…”

The Influence of Aerial Hyperspectral Image Processing Workflow on Nitrogen Uptake Prediction Accuracy in Maize

“…Statistical optimization techniques are being recognized for their importance towards the increased adoption of precision agriculture technologies [54], in part because of the advantages they offer in handling the specificity and high-dimensional nature of big data in agriculture. This dimensionality problem is only being exacerbated with advanced sensor technologies such as hyperspectral imagery [5], highlighting the need for robust, straightforward protocols aimed at delivering optimized machine learning models.…”