AI and drones transform how farmers tackle climate-driven crop loss

Artificial intelligence is rapidly becoming a frontline tool for global agriculture as farmers face the mounting challenges of climate extremes. A new review published in Plants provides a detailed assessment of how machine learning, deep learning, ensemble models, and explainable AI are shaping crop yield predictions under unpredictable weather conditions.

The study, titled “Recent Trends in Machine Learning, Deep Learning, Ensemble Learning, and Explainable Artificial Intelligence Techniques for Evaluating Crop Yields Under Abnormal Climate Conditions,” systematically reviews advances made between 2019 and 2024, synthesizing insights from around 60 selected research papers across major databases to map how artificial intelligence is transforming precision agriculture.

How are AI models being applied to crop yield prediction?

The review identifies machine learning and deep learning as the dominant approaches in yield forecasting. Among machine learning techniques, Random Forest and Support Vector Machines emerged as the most frequently applied algorithms, offering solid predictive capacity across diverse datasets. On the deep learning side, Artificial Neural Networks and Convolutional Neural Networks dominate, particularly where image-based sensing data is available.

The study also highlights ensemble methods, with stacking techniques showing especially strong results. By combining multiple models, ensemble learning reduces bias and variance, delivering more stable predictions across different climate scenarios. Importantly, explainable AI is beginning to be integrated into these systems, offering pathways to clarify how models reach their conclusions. This emerging focus on interpretability is seen as critical to building trust among farmers, agronomists, and policymakers who rely on these predictions for decision-making.

The research underscores that feature selection plays a vital role. Stepwise feature selection often outperformed models where additional variables were indiscriminately added, proving that targeted inputs are more effective than data overload. This finding suggests that precision in data collection matters more than sheer volume.

What role do sensors and climate variables play in predictions?

A significant portion of recent work has focused on integrating remote sensing and imaging technologies into predictive models. Hyperspectral and multispectral imaging, especially when deployed via drones, has become the most common data source for monitoring crop growth and estimating yields. These imaging systems capture detailed spectral signatures, allowing algorithms to track subtle variations in plant health and soil conditions.

The review asserts that environmental variables are critical determinants of yield outcomes. Temperature, precipitation, soil fertility, and soil moisture consistently appear as the most influential factors in prediction models. The authors categorize these into three main groups: crop-related, soil-related, and temperature-related variables. Together, they form the foundation for accurate and context-specific forecasting.

The ability of AI to integrate multiple environmental drivers makes it particularly valuable in abnormal climate conditions. As weather patterns become less predictable, conventional models struggle to capture variability. By contrast, AI systems can adjust dynamically, learning from new data streams to refine forecasts. However, the study warns that extreme climate events still expose weaknesses, with performance often declining when models are tested outside their trained conditions.

What are the challenges and future directions for AI in agriculture?

While the review finds that AI models generally achieve strong accuracy, often with root mean square errors ranging between 0.2 and 0.5 tons per hectare and R² values from 0.4 to 0.9, serious limitations remain. The foremost challenge is generalization. Models trained on specific crops, regions, or climate conditions often lose accuracy when applied elsewhere, limiting their scalability.

The authors stress the urgent need for larger and more diverse datasets that cover a wider range of crops and geographic locations. Such datasets would strengthen domain adaptation and improve model resilience to out-of-sample scenarios. Another priority is uncertainty quantification. Current AI systems tend to present forecasts as definitive outputs, yet farmers and policymakers require clear indicators of confidence and risk levels in order to make informed choices. Ensemble methods and explainable AI could play a major role in addressing this gap.

The study also identifies a critical shortfall in interpretability. While deep learning models can outperform simpler methods, their “black-box” nature undermines adoption in practice. Explainable AI is still in its infancy but offers promise for revealing which variables most influence predictions. This transparency is essential not only for farmer trust but also for regulatory acceptance and policymaking.

Implications for Global Agriculture

The ability of AI to synthesize complex environmental and crop-related data makes it uniquely suited to supporting food security at a time of growing uncertainty. However, the technology’s effectiveness will depend on progress in several areas: expanding data diversity, improving model adaptability, enhancing interpretability, and embedding uncertainty measures.

For farmers, these advances could mean more reliable planning for planting, irrigation, and harvesting. For policymakers, AI-based forecasts could guide national food security strategies, ensuring resilience against climate shocks. For researchers, the challenge is to balance technical performance with transparency and usability.

The next stage of development, as the authors envision, will not only determine how well AI predicts yields but also how effectively those predictions translate into real-world agricultural resilience. The integration of explainable and ensemble methods, supported by large-scale data collaboration, could mark the turning point for making AI a dependable ally in the global effort to safeguard food supplies against the threat of climate change.