Agriculture has been the backbone of our nation's economy. The sector employs more than 50% of the labour force in India, and the country ranks second in the world in terms of farm output. However, the sector is also heavily dependent on the weather, especially with the effects of climate change accelerating changes. Too much rain, not enough rain, scorching heat – it all affects how much food you'll be able to grow. Knowing how much yield you can expect before planting or early in the growing season could significantly boost production and returns. Stakeholders can manage resources like water and fertiliser more efficiently and make smarter decisions about what to plant.
Traditionally, predicting crop yield relied on experienced farmers or agricultural experts observing the fields throughout the season. They examine everything, from soil type and seed quality to plant growth, water availability, and weather conditions. This is time-consuming and can still be uncertain. Researchers have been exploring how artificial intelligence (AI) can enhance the accuracy and timeliness of these predictions by learning from data to inform decisions and predictions.
In a recent study, researchers from Sardar Vallabhbhai National Institute of Technology, Surat, Arba Minch University, Ethiopia and ICAR-Indian Institute of Water Management, Bhubaneswar, examined how a specific type of AI, known as ANFIS (Adaptive Neuro-Fuzzy Inference System), could be utilised to predict the yield of several important crops grown in the Nashik region of Maharashtra, India. This area is considered semi-arid, meaning it's relatively dry for a significant portion of the year, making farming particularly sensitive to weather fluctuations.
The Nashik region in Maharashtra, where this study took place, is known for its vineyards and has been described as "The Wine Capital of India", producing around 10,000 tonnes of grapes per year. |
The researchers aimed to determine whether ANFIS could accurately predict the yield of five key crops — kharif rice, sorghum, maize, groundnut, and sugarcane — using only weather information. They focused on five main weather factors: average rainfall, minimum and maximum temperatures, average relative humidity, and average evaporation. These are all things that significantly impact how well crops grow.
To train their ANFIS models, they used historical data from records of past weather conditions and the corresponding crop yields for each of these crops in the Nashik region. They gathered data from 1987 to 2020 from several weather stations. However, even with 33 years of data, they found that they didn't have quite enough to train the AI models to the desired accuracy.
So, they used mathematical techniques called normal and inverse normal distributions to synthesise more data. These mathematical methods enabled the researchers to create numerous more realistic yet artificial examples based on the patterns in the real data. This provided the AI with more practice problems to learn from, enabling it to understand the relationship between weather and yield better.
Once they had enough data (both real and synthesised), they split it into two parts: 70% was used to train the ANFIS models, teaching them the patterns between weather and yield. The remaining 30% was kept separate to test the models, seeing how well they could predict yields for data they hadn't seen before.
ANFIS is somewhat akin to combining two distinct types of AI. One part is similar to a neural network, which excels at learning complex patterns from data. The other part is based on a concept called fuzzy logic. Fuzzy logic is a way of thinking that mimics how humans think, using concepts like high temperature or low rainfall rather than just strict numbers. It enables the system to create if-then rules, such as "IF rainfall is low AND temperature is high, THEN the expected yield is low." The neural network part helps the fuzzy logic part automatically create and fine-tune these rules based on the data.
The researchers found that their ANFIS models worked well, predicting the crop yields with “acceptable accuracy.” They used several statistical measures to assess the accuracy of the predictions, examining factors such as the closeness of the predicted yields to the actual yields. They found that the rules created by ANFIS accurately captured the relationship between the climate variables and the yield for each crop.
They found that the model predicted sugarcane yield with the highest accuracy. The model for sorghum had the lowest accuracy compared to the others. The researchers suggested this might be because sorghum is a crop that needs less water and is less dependent on the specific climate factors they included compared to the other crops studied.
The researchers also pointed out some limitations. As you add more input variables (like more weather factors or other things), the fuzzy logic system in ANFIS can become very complex, making the training process slow. Additionally, like any AI, it requires adequate data validation to prevent overfitting, where the model becomes overly proficient at predicting the training data but struggles to predict new data. A significant limitation of this study is that it only utilised climate data. While climate is crucial, other factors, such as soil type, the amount of fertiliser used, irrigation practices, and pests, also significantly impact yield. Including these could further improve prediction accuracy.
Despite these limitations, the study demonstrates that utilising ANFIS with climate data is a promising approach for predicting seasonal multi-crop yields early. By incorporating multiple crops and climate factors, the new research enhances the practicality of AI models for real-world farming, where farmers often cultivate several crops and yields are influenced by a combination of weather conditions. They also successfully employed data synthesis to overcome the common problem of insufficient historical data. The study represents a significant step towards utilising innovative technology to help feed a growing world, particularly in regions facing the challenges of a changing climate.
This research article was written with the help of generative AI and edited by an editor at Research Matters.