Proof of Concept
Introduction
PAGGAIA is an ambitious project aiming to increase the overall agriculture production by improving the soil quality of a given field. An important aspect of our work was to make our solution scalable and extendable. The involvement of farmers helped us to better understand the issue at hand in addition to getting critical data that can only be precisely provided by them, to implement our project. Nevertheless, we managed to design a system that proves our concept. The proof of concept for our project aims at combining physicochemical and agronomic data from the field with the lab results in order to showcase if our project as a whole, will be applicable in the real world.
Design & Flow
Here we demonstrate a real world example based on the PAGGAIA approach, including our experiments, to prove that our project is likely to work in a relevant context.
Step 1 - Farmers seek an optimal Agriculture Production
A major issue in the field of Agriculture is the low efficiency of crop inputs including fertilizers, tillage, pesticides, and irrigation water. To a great degree, a determining factor is the lack of important data regarding soil physical and chemical properties in large and isolated areas. Research on the existing literature data has concluded that there is insufficient information regarding the combination between precision agriculture tools and other technological advancements. Among these are: Machine Learning (ML), Next-Generation Sequencing, and Unmanned Aerial Vehicles.
Step 2 - Soil sampling
We determined that following a strict protocol leads to the avoidance of cross-contamination of the samples. Soil samples were manually collected, while changing gloves before the collection of samples from different areas of the field, while also using a new falcon tube for each sample. The soil samples were taken at a depth of 0-30 cm from the surface using soil sampling tools. The samples were then preserved in the freezer in optimal conditions.This procedure can be automated by using aerotransportable equipment, such as specially designed soil-sampling drones, with a custom-made soil sampling mechanism attached.
The field we visited had been cultivated with industrial tomatoes (Solanum lycopersicum L.) and more specifically HEINZ 1015 hybrids and impregnated with B. subtilis bacterial strains.
Step 3 - Data Sources
The temperature and the precipitation data were recorded by a weather station Weatherhub (TFA, Wertheim-Reicholzheim, Germany) which was placed in the experimental area. The variables used were the average, the maximum and the minimum temperature of each month and the total precipitation of each month. Physical and mineral properties of the soil of the experimental site at the depth of 0-30 cm were determined as described in Table 1.
| Parameters | Method |
|---|---|
| Sand (%) | Bouyoucos, 1962 |
| Silt (%) | |
| Clay (%) | |
| Soil Texture | |
| pH | pH-meter |
| Electrical Conductivity (mS cm-1) | ISO 11265:1994 |
| Total salts (%) | calculation |
| Organic Matter (%) | ISO 14235:1998 |
| Total Nitrogen (mg g-1) | ISO 11261:1995 |
| Available K (cmol+ k-1) | atomic absorption spectrometry |
| Available Ca (cmol+ kg-1) | |
| Available Mg (cmol+ kg-1) | |
| Available P (mg kg-1) | ISO 11263:1994 |
| Fe-DTPA (mg kg-1) | DTPA |
| Cu-DTPA (mg kg-1) | |
| Zn-DTPA (mg kg-1) | |
| Mn-DTPA (mg kg-1) | |
| Available B (mg kg-1) | Page, 1986 |
Step 4 - Next-Generation Sequencing (NGS) and Data Analysis
Next-Generation Sequencing (NGS) of 16S rRNA gene was performed in our lab, but can also be performed directly at the field using portable DNA sequencing devices plugged into a laptop, which can provide rapid, real-time and actionable results at a low cost. The resulting metagenomic data, together with the physicochemical and agronomic data are being analysed in order to get the desired dataset for our Machine Learning model. After this step, the input of the ML model is ready.
Step 5 - Training a Machine Learning model
The model is trained using the analysed data from the previous step and is able to provide the desired output which helps the farmers to increase the quality of the soil and thus, the agriculture production.
ML has already been used in agriculture research over the years and that reflects the meaningful contribution of applying that technology in agriculture. In a previous work, a plant-oriented ML model using Bayesian Ridge algorithm was developed in order to predict the Total Soluble Solids (°Brix) of industrial tomatoes during the cultivation period. The dataset used, contained weather and soil data as well as data from previous cultivation periods1. Another °Brix was developed using Bayes Automatic Relevance Determination (ARD) algorithm in order to reduce the environmental impact and the cost of cotton cultivation by reducing the irrigation times2. Lastly, it is important to mention that adding microorganisms to soil in order to create variability between the plots and to find which one is the best under these circumstances was tested before3. The scalability of the model is based on the possibility of the model to resample weather data when needed (monthly average, 15-days average, 10-days average etc.). Moreover, 18 different classifiers were tested in order to find the one which provides the optimal results and will be used for the model. This process will be repeated each time the dataset updates since new relations between the variables may be created. Last but not least, correlation between the variables is extracted, aiming at finding variables which are important to the models and features that do not affect the model and could be removed to make the model faster and more accurate.
Step 6 - Report
The output of the model, a report in a specific format, is shared with the farmer through a smartphone application. The final report consists of two parts:
- First part: Results of the soil analysis (collected samples) which include statistics for the variables used such as, the weather and the microorganisms
- Second part: Finite number of steps for the farmers to follow, as suggested tailor-made interventions to the soil, in order to improve the soil quality and the crop yield, according to the needs of each cultivation area
Step 7 - Previous Work
The development of a ML model for soil prediction advice will help farmers improve their field and their crop yield as well as improve quality features. Similar models have been designed in the previous years. More specifically, we were based on models for °Brix prediction and reduction of irrigation water were used12. In addition, previous experiments in Maize and industrial tomato proved that microorganisms changed the agronomic characteristics of the cultivation34. This way we could have variability between different plots in the same field, instead of having the same field experiment in different locations.
References
-
Kasimatis, C.-N., Psomakelis, E., Katsenios, N., Katsenios, G., Papatheodorou, M., Vlachakis, D., Apostolou, D., Efthimiadou, A., 2022. Implementation of a decision support system for prediction of the total soluble solids of industrial tomato using machine learning models. Comput. Electron. Agric. 193, 106688.
-
Leonidakis, D., Psomakelis, E., Kasimatis, C.N., Katsenios, N., Kakabouki, I., Roussis, I., Mavroeidis, A., Efthimiadou, A., 2021. Development of Decision Support System Based on the Bayes ARD Algorithm for Irrigation of Cotton. Bull. Univ. Agric. Sci. Vet. Med. Cluj-Napoca Hortic. 78, 112.
-
Katsenios, N., Andreou, V., Sparangis, P., Djordjevic, N., Giannoglou, M., Chanioti, S., Kasimatis, C.-N., Kakabouki, I., Leonidakis, D., Danalatos, N., Katsaros, G., Efthimiadou, A., 2022. Assessment of plant growth promoting bacteria strains on growth, yield and quality of sweet corn. Sci. Rep. 12, 11598.
-
Katsenios, N., Andreou, V., Sparangis, P., Djordjevic, N., Giannoglou, M., Chanioti, S., Stergiou, P., Xanthou, M.-Z., Kakabouki, I., Vlachakis, D., Djordjevic, S., Katsaros, G., Efthimiadou, A., 2021. Evaluation of Plant Growth Promoting Bacteria Strains on Growth, Yield and Quality of Industrial Tomato. Microorganisms 9, 2099.