AI in Fertilization: Variable Rate Application

Artificial Intelligence in fertilization, particularly through Variable Rate Application (VRA), represents a transformative advancement in precision agriculture.

1. By leveraging machine learning algorithms and real-time data from soil and crop sensors, AI-driven VRA systems can tailor fertilizer application rates to the specific needs of different field zones.
2. This approach optimizes nutrient use, enhances crop yields, and reduces environmental impact by preventing over-fertilization and minimizing runoff.
3. The integration of remote sensing technologies, such as NDVI, further refines these applications by providing accurate, up-to-date assessments of crop health and soil conditions.

As a result, AI-powered VRA not only improves the efficiency and sustainability of fertilization practices but also offers significant economic benefits for farmers through more precise input management and increased productivity.

Let's look closer at specific use cases of VRA of fertilisers for potatoes, grapevine and maize in Belgium, Italy and South Africa respectively.

Smart Fertilisation for Potato Farming

This study has been recently published by the researchers from Belgium ????. They investigated the benefits of variable rate nitrogen (VR-N) application in potato farming by integrating visible and near-infrared spectroscopy and remote sensing data to optimize nitrogen use and increase crop yield.

The methodology involved the use of an online vis-NIR spectroscopy sensor to measure various soil properties, and Sentinel-2 satellite imagery to obtain the normalized difference vegetation index (NDVI). Soil data and NDVI were combined to create fertility management zones (MZs), which guided the variable application of nitrogen fertilizers. The experiment compared VR-N to uniform rate (UR) treatments across different MZs, adjusting nitrogen levels according to soil fertility.

Key findings indicated that VR-N led to a reduction in nitrogen use by 50% in high-fertility zones (VR-H) and a 25-50% increase in low-fertility zones (VR-L), resulting in an 8.1% increase in potato yield for VR-H zones and overall higher yields in VR-ML and VR-L zones. The VR-N treatment provided a relative gross margin of 374.83 €/ha over the UR treatment, demonstrating both economic and environmental benefits.

Farmers and agronomists can apply these findings to enhance potato yields and reduce environmental impacts through precision nitrogen management.

Technologies used:

• Online vis-NIR spectroscopy sensor
• Sentinel-2 satellite imagery
• Differential Global Positioning System (DGPS)
• Partial least squares regression (PLSR) models
• K-mean clustering analysis
• ArcGIS software for spatial analysis

Spatially Variable Rate Fertilizer Management in a Sicilian Vineyard Using Sentinel-2 Satellite Data

The next study, conducted by a consortium of researchers from Italy ???? and Portugal ????, aimed to optimize nitrogen fertilizer application in two Sicilian vineyard plots using Sentinel-2 satellite data to create spatially variable rate fertilization maps.

The methodology involved using Sentinel-2 satellite imagery to calculate Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) for assessing vegetative vigour and leaf water content. These indices were used to produce management zones and determine the optimal fertilization time. The study focused on two vineyard plots, one with Syrah and the other with Nero d’Avola grape varieties. The data were processed using ArcGIS software, and temporal and spatial variability was analyzed.

Key findings showed that the optimal fertilization time for both plots was determined to be 12 April 2021. The spatially variable rate fertilization approach helped in reducing production costs and minimizing environmental impact. The study highlighted the significant temporal and spatial variability in vegetative vigour and leaf water content in the two vineyard plots. Applying the recommended variable rate fertilization improved the overall efficiency and sustainability of vineyard management.

Viticulturists and vineyard managers can practically apply these results to enhance vineyard management practices, optimize fertilizer use, and improve grape yield and quality.

Technologies used:

1. Sentinel-2 satellite imagery
2. Normalized Difference Vegetation Index (NDVI)
3. Normalized Difference Water Index (NDWI)
4. ArcGIS software
5. Statistical data analysis

Agronomic Management, Soil and Remote Sensing Variables in Maize Farming

Finally, the study, conducted and recently published by researchers in South Africa ????, aimed to evaluate the predictive accuracy of machine learning (ML) models for maize yield predictions and identify key yield-limiting factors in a data-intensive farm management (DIFM) setting.

The research employed multiple linear regression (MLR), multilayer perceptron (MLP), decision tree (DT), and random forest (RF) models, using data from 2019/2020 and 2020/2021 maize seasons. The models were trained and tested on datasets consisting of crop management, soil properties, and NDVI data. The process involved preprocessing data, splitting it into training and testing sets, and applying the respective ML algorithms.

Key findings revealed that the RF model achieved the highest predictive accuracy with R2 values of 0.69 and 0.80, and low MAPE values of 5.4% and 8.4% for the two seasons. Feature importance analysis showed that urea application was the most critical factor influencing yield variability. The study also indicated that incorporating NDVI data significantly improved model performance.

The integration of remote sensing data, such as NDVI, with machine learning models significantly enhances the predictive accuracy and efficiency of yield predictions. The use of Random Forest models, which incorporate both soil and crop management data, allows for precise identification of critical factors like urea application rates that influence yield variability, thereby enabling more effective and site-specific fertilization strategies.

The results of this research can be practically applied by farmers and agronomists to optimize input management and improve yield predictions.

Technologies and Hardware Used in this Study:

• Python Keras libraries
• Google Colaboratory
• NDVI remote sensing data
• Machine learning algorithms (MLR, MLP, DT, RF)

References for "AI in Fertilization: Variable Rate Application"

??What's next in AI in Fertilization?

In the next 'AI in Fertilization' edition, we will continue exploring the cases of utilizing satellite imagery and NDVI (Normalized Difference Vegetation Index) to monitor crop health and guide fertilization strategies.

?What do you think about this topic - would you like to suggest anything else?

Please, share your thoughts with us in the comments below ??

Wishes of proper variability of fertilizer application,

Maryna Kuzmenko, Ph.D ????, Chief Inspiration Officer at Petiole Pro Community

#fertilizers

Photo credit for the cover:

Qaswar, M.; Bustan, D.; Mouazen, A.M. Economic and Environmental Assessment of Variable Rate Nitrogen Application in Potato by Fusion of Online Visible and Near Infrared (Vis-NIR) and Remote Sensing Data. Soil Syst. 2024, 8, 66. https://doi.org/10.3390/soilsystems8020066