(3)INESC TEC - C-BER
In order to present an idea of the benefits of using artificial intelligence (AI) in agriculture, we explore three use cases. These use cases do not cover all possible AI applications in agriculture, but they can demonstrate its usefulness in this fundamental economic area. The three use cases are: water management, production estimation and human resources management.
Water management: If we think that: (1) the agriculture sector is, according to the United Nations, an activity that requires around 70% of all the water we consume worldwide; (2) some areas of the globe are feeling the impact of the global warming, namely having less availability of water for agriculture [1]; (3) and the optimised use of the water influences the quality of the production. All these notions show us the importance of optimising the use of water in agriculture. Within the scope of the Smart Farming project, INESC TEC has developed an irrigation system for a vineyard that allows to manage the irrigation according to the intended hydric stress that is expected to generate in the plants. The amount of hydric stress in the plant conditions the alcohol content of wine that is produced with these grapes. In order to produce the irrigation management system, it was necessary to have sensors at different depth in the soil. We also considered the predawn leaf water potential (PLWP), measured using the Scholander Pressure Chamber. However, due to the cost of obtaining PLWP measures, a regressor to estimate the PLWP was developed [4]. These estimations are used in the optimisation method to define the amount of water for irrigation over the next seven days. Genetic algorithms were used as optimisation method.
Production estimation: agricultural production is strongly sensitive to climate. Current climate change can disrupt the environment and change pest and plagues behavior, causing large inter-annual variations in crop production that do not match the constant growth in demand. Moreover, crop production is affected by farmer decisions such as irrigation, fertilization, and selection of crop seeds. Unpredictable inter-annual variations in crop production are a major threat to farmers, industry, and society. Therefore, there is a strong demand for predictive models that can improve the production efficiency, optimise the production plan/operations and support commercial strategies [1,2]. There is still a great lack of operational methods, particularly to predict the weather and production at farm level. Crop simulation models and data–driven models are the basis of the most popular approaches for crop yield prediction. The crop simulation models are very complex and expensive in terms of time and biophysical data requirements, thus hindering its operation. Data-driven predictive models of yield predictions are built empirically, and do not require a deep knowledge on biophysical mechanisms that produced the data; they are inexpensive and already proved to be extremely efficient methods [1, 2, 3]. Therefore, during the last few years, several Machine Learning techniques such as regression trees, random forest, support vector machines and deep learning have been successfully applied to forecast crop yield. Most of these predictive methods explore climate data (e.g., NWP) and crop related information (e.g., phenostages) to make predictions. Moreover, these predictions can then be used as input to a mathematical optimisation model that can find the optimum production plan [2,3].
Human resources management: in order to hire experienced workers, producers are starting to offer higher salaries and benefits to their staff, like health insurance. This issue raises a concern that was not previously in mind of producers: are my workers doing the job properly? Is the effort proportional to the salary and benefits? Such concerns require productivity indicators of each worker, which paves the way to a new area in agriculture and fruticulture, related with the need to implement technologies for workers’ monitoring. INESC TEC researchers believe that workers’ monitoring might be of high benefit for the employee and the workers themselves, if we consider that not only productivity indexes can be estimated, but also health related indicators, important to ensure well-being and good working conditions, and support more informed decision-making. INESC TEC is already tackling this issue with a national project named AgWearCare, which aims to apply this monitoring concept to retrieve specific indicators in this area, together with WiseCrop solution. A system like this could have a wide range of advantages, such as understanding the distance made by each worker in a vine harvest, identifying the worker posture or even detecting exposure to extreme working conditions or high levels of human effort (combining wearable devices with advanced data processing and artificial intelligence methodologies). Like all amazing technologies, there are always disadvantages – to what extent would workers allow being monitored by their employee, and to provide them access to their performance during working hours? This is a huge issue that not only has an ethical and data protection concerns, but also labor laws and policies that should be taken in consideration. It might be a huge burden to the workers to expose that much to their employee, in a way that the latter would be aware of their movements, activities and ways of working in agriculture and fruticulture scenarios. It is vital to find a balance between these two approaches, in order to be able to collect enough productivity indicators to support decision-making without compromising the worker privacy – and, more than that, without losing workers’ trust.
References
[1] Gornall J, Betts R, Burke E, et al. (2010) Implications of climate change for agricultural productivity in the early twenty-first century. Philos Trans R Soc Lond B Biol Sci.: 365(1554):2973-2989. doi:10.1098/rstb.2010.0158.
[2] MS Sirsat, J Mendes-Moreira, C Ferreira, M Cunha (2019) Machine Learning predictive model of grapevine yield based on agroclimatic patterns. Engineering in Agriculture, Environment and Food 12 (4), 443-450.
[3] Yoosefzadeh-Najafabadi M, Tulpan D, Eskandari M (2021) Application of machine learning and genetic optimization algorithms for modeling and optimizing soybean yield using its component traits. PLOS ONE 16(4): e0250665. https://doi.org/10.1371/journal.pone.0250665
[4] AA Fares, F Vasconcelos, J Mendes-Moreira, C Ferreira (2021) Predicting Predawn Leaf Water Potential up to seven days using Machine Learning, EPIA 2021, pp. 39-50.