Mapping and assessment of abiotic stresses in hot semi-arid ecosystem of western India using analytical hierarchy process and machine learning models

Abstract

Abiotic stress refers to non-living environmental factors that adversely affect crop growth, development, and productivity. Accurate mapping of abiotic stresses is essential for effective agricultural planning and resource management. This article introduces a novel approach for abiotic stress mapping by integrating various terrain, climatic, pedological and vegetation parameters using the analytical hierarchy process (AHP) and AHP-integrated machine learning (ML) models for the Pune district, a hot semi-arid ecosystem of western India. The abiotic stress map was generated through three key steps: first, datasets acquisition and processing, where relevant thematic layers were collected and pre-processed; second, AHP-based weightage assignment and consistency analysis, where thematic layers and their sub-classes were assigned, weights based on expert judgment, and consistency was verified; and third, generation of the abiotic stress map using the weighted sum approach. The final abiotic stress map was generated by integrating the reclassified weighted thematic layers. Furthermore, the result of AHP was used with other thematic layers to build AHP-integrated ML models. The generated map was validated using high-resolution Google Earth imagery at randomly selected locations to ensure robust verification. A receiver operating characteristic (ROC) curve was created with these selected points to confirm the model’s ability to effectively discriminate between different stress levels. The results indicate that high and very high-stress zones are predominantly located in the southern and southeastern parts of the district, mainly in Purandar, Baramati, Indapur, and Daund tehsils, where drought, shallow soils, and low annual rainfall (< 550 mm) are prevalent. The study demonstrates the potential of the AHP and combined ML models in abiotic stress mapping and identification of hot spots with reasonable accuracy and the findings of the study can be applied to develop combating strategies to reduce the impact of abiotic stress on agricultural systems.