Apart from pests and soil erosion, crops face many unseen challenges. For successful farming, it is essential to understand these factors, especially in places where the environment can be harsh, the weather unpredictable, the soil isn't ideal, or there is insufficient water. Such non-living environmental factors, like extreme temperatures, drought, poor soil quality, or even too much rain, are called abiotic stresses. Researchers note that about two-thirds of agricultural losses worldwide are due to soil-related issues alone.
Pune district in Maharashtra knows these challenges well. It's a hot, semi-arid region in the rain shadow of the Western Ghats. The hills block much of the rain, leaving the eastern side of the ghats, including Pune, much drier. Farmers here often deal with drought and intense heatwaves, making it challenging to grow food. Knowing exactly where these stresses happen could help farmers and people planning agricultural resources make smart decisions about what to plant, how to manage water, and how to protect their crops.
To help tackle this, researchers from the ICAR-National Institute of Abiotic Stress Management and the ICAR-National Bureau of Soil Survey & Land Use Planning developed a detailed map showing the areas in Pune most affected by abiotic stress.
Their approach combined different types of environmental information with machine learning to develop a detailed stress map. They gathered different maps of the same area, including one showing the kind of soil, another showing rainfall data, one for temperature, one for how healthy the plants look from a satellite, and one for the slope of the land. In all, the researchers collected data for ten different environmental factors that influence abiotic stress, using information from satellites and ground surveys.
Next, they used the Analytical Hierarchy Process, or AHP, which helps analyse complex situations by breaking down the factors into a hierarchical structure. The researchers ranked the ten environmental factors that cause abiotic stress in Pune in terms of their importance. The analysis showed that soil depth was the most crucial factor, followed by soil pH, land use (whether farmland or bare ground), and different plant health measures as seen from space. Each of the factors was then assigned weights based on their contribution to crop stresses.
Once they had these weights, they used a weighted sum by adding up all the maps, each multiplied by its weights, to create a single map showing the overall level of abiotic stress. This first map, based purely on the AHP weights, gave them a good starting point.
The researchers then used the AHP results, including the stress map, weights, and original environmental data, to train machine learning models. They used two popular types of models: Random Forest and Support Vector Machine. These models are good at finding complex patterns in data. By feeding them all the environmental information and the AHP results, the models learned to identify the areas most likely to be under high abiotic stress.
Their results showed that the Pune region, particularly in the southern and southeastern tehsils like Purandar, Baramati, Indapur, and Daund, where drought, shallow soils, and low rainfall are common, faced increased crop stresses.
The eastern part of Pune district has been historically drought-prone, but the constructions of dams, canals and wells have made agriculture less dependent on rainfall. |
To ensure accuracy, the researchers performed two kinds of checks. First, they did a visual check by visually comparing high-resolution satellite images from Google Earth for specific spots in Pune that their map identified as having very low, low, moderate, high, and very high stress. They found that the map's results matched the real-world conditions very well.
Second, they did a quantitative check using a standard scientific method called the Receiver Operating Characteristic (ROC) to see how well the model could distinguish between different categories, in this case, different stress levels. Their initial AHP map had a good AUC score of 0.784, but the maps created using the AHP-integrated machine learning models were even better, with AUC scores of 0.956 for Random Forest and 0.933 for Support Vector Machine. This showed that combining AHP with machine learning significantly improved the accuracy of the stress mapping.
This research builds on previous work by moving beyond just looking at one type of stress, like drought, and instead assessing multiple abiotic stresses together. This gives a more complete picture of the challenges crops face. While the study successfully created accurate stress maps, one limitation they noted was that some of the soil data they used wasn't as detailed as the satellite data. Future research could use higher-resolution soil information to make the maps even more precise. They also suggest expanding this approach to map abiotic stress across the entire hot semi-arid region of India, not just Pune.
The maps generated by this study are incredibly valuable for society, especially for the people of Pune who rely on agriculture. By clearly showing which areas are most vulnerable to abiotic stress, the maps provide essential information for making informed decisions, contributing to more sustainable farming practices and helping communities cope with the challenges of a changing environment.
This research article was written with the help of generative AI and edited by an editor at Research Matters.