Climate change is no longer a distant concern for agriculture—it is a present and escalating challenge. Across the globe, extreme weather events such as heat waves, droughts, floods, and cold snaps are becoming more frequent and more intense. These changes are already reshaping agricultural productivity and placing unprecedented pressure on global food security.

The Rise of Extreme Weather Events

Over the past few decades, extreme weather events have increased in both frequency and severity. A systematic review by Cogato et al. (2019) highlights how agriculture is uniquely vulnerable to these events due to its direct dependence on climatic conditions. Unlike other sectors, even short-term anomalies—such as a heat wave during flowering or unexpected flooding before harvest—can result in severe yield losses.

Heat waves, in particular, pose a major threat. Research by Schlenker and Roberts (2009) demonstrated that crop yields respond nonlinearly to temperature, meaning that once critical thresholds are crossed, yields decline sharply. This finding suggests that even modest increases in extreme heat days can have disproportionately large impacts on agricultural output.

Impacts on Crop Yields Across Regions

The effects of extreme weather are not uniform; they vary across crops, regions, and agroecological systems. Lesk et al. (2016) analyzed global data on weather disasters and found that major droughts and heat events have significantly reduced global crop production, with some regions experiencing losses exceeding 10% during extreme years.

In Europe, Webber et al. (2018) showed that drought stress affects crops differently. While maize yields are highly sensitive to water shortages, winter wheat shows greater resilience in certain regions. This divergence highlights the complexity of climate impacts and the need for crop- and region-specific adaptation strategies.

Food Security at Risk

Declining agricultural productivity has direct implications for food security, particularly in developing and climate-vulnerable regions. Reduced yields can lead to higher food prices, increased reliance on imports, and heightened risks of malnutrition. A recent review focusing on Thailand by Waqas et al. (2024) underscores how climate change threatens smallholder farmers through yield instability, income loss, and increased production risks.

These challenges are compounded by socio-economic factors such as limited access to technology, inadequate infrastructure, and weak institutional support, making adaptation even more difficult for vulnerable farming communities.

Looking Ahead: Adaptation and Resilience

Addressing the impacts of extreme weather on agriculture requires a multifaceted approach. Climate-smart agricultural practices—such as drought-tolerant crop varieties, improved irrigation management, early warning systems, and diversified cropping systems—can help build resilience. At the policy level, investments in agricultural research, climate services, and farmer education are essential.

As extreme weather events continue to intensify, the relationship between climate and agriculture will become even more critical. Understanding these impacts and acting proactively is not just an agricultural priority—it is a global food security imperative.

References

Cogato, A., Meggio, F., De Antoni Migliorati, M., Marinello, F. 2019. Extreme weather events in agriculture: A systematic review. Sustainability, 11(9), 2547.

Lesk, C., Rowhani, P., Ramankutty, N. 2016. Influence of extreme weather disasters on global crop production. Nature, 529(7584), 84-87.

Schlenker, W., Roberts, M.J. 2009. Nonlinear temperature effects indicate severe damages to US crop yields under climate change. Proceedings of the National Academy of sciences, 106(37), 15594-15598.

Waqas, M., Naseem, A., Humphries, U.W., Hlaing, P.T., Shoaib, M., Hashim, S. 2024. A comprehensive review of the impacts of climate change on agriculture in Thailand. Farming System, 100114.

Webber, H., Ewert, F., Olesen, J.E., Müller, C., Fronzek, S., Ruane, A.C., Bourgault, M., Martre, P., Ababaei, B., Bindi, M. 2018. Diverging importance of drought stress for maize and winter wheat in Europe. Nature communications, 9(1), 4249.

Water is at the heart of agriculture, ecosystems, and climate resilience. One of the most important—but difficult—variables in water management is potential evapotranspiration (PET), which represents the atmospheric demand for water from soil and vegetation. Accurate PET estimates are essential for irrigation planning, drought assessment, and hydrological modeling.

Traditionally, PET has been estimated using empirical equations. But with the rise of artificial intelligence, a key question emerges: Can deep learning do better?

Why PET Estimation Matters

PET influences how much water is available in rivers, reservoirs, and agricultural fields. Over- or underestimating PET can lead to poor water-allocation decisions, inefficient irrigation, and inaccurate drought forecasts—especially in regions with complex climate conditions.

The Nakdong River Basin (NRB) in South Korea is one such region. It spans diverse topography and climate zones, making PET estimation particularly challenging.

Traditional Methods: Useful but Limited

For decades, empirical models like Priestley–Taylor (P–T) and Hargreaves–Samani (H–S) have been widely used because they are simple and require limited data. However, simplicity comes at a cost.

When these models were applied to long-term meteorological data (1973–2024) from 13 stations in the NRB, their performance was poor. Large errors and negative efficiency scores revealed that these traditional equations struggle to represent the basin’s complex climate behavior. In short, what works elsewhere does not always work locally.

Enter Deep Learning

Deep learning (DL) models excel at learning complex, nonlinear relationships—exactly what climate variables tend to have. In this study, two DL approaches were tested:

• A Long Short-Term Memory (LSTM) network, designed to capture temporal patterns in time-series data

• A hybrid CNN–Bidirectional LSTM model with an attention mechanism, which combines spatial feature extraction, bidirectional temporal learning, and adaptive focus on important inputs

• Both models were trained using multiple combinations of meteorological inputs to test how robust they remain when data availability changes.

A Clear Winner Emerges

The results were striking. Deep learning models dramatically outperformed empirical methods, producing far more accurate and reliable PET estimates.

Among them, the hybrid CNN–BiLSTM with attention delivered the best performance. It not only reduced estimation errors but also consistently captured daily variability and seasonal cycles of PET. Even more importantly, the model generalized well across the basin, with strong performance at individual meteorological stations—despite differences in local climate.

This suggests that hybrid DL architectures are particularly well-suited for heterogeneous river basins, where climatic behavior varies from one location to another.

Why the Hybrid Model Works So Well

The strength of the hybrid model lies in its design:

• CNN layers extract meaningful patterns from meteorological inputs

• Bidirectional LSTM layers learn temporal dependencies both forward and backward in time

• Attention mechanisms focus the model on the most influential features during PET estimation

Together, these components allow the model to “understand” climate dynamics in a way that fixed empirical equations cannot.

What This Means for Water Management

Accurate PET estimation is critical for:

• Irrigation scheduling

• Drought monitoring

• Climate impact assessments

• Hydrological and land-surface modeling

This study shows that deep learning is not just an academic exercise—it offers real, operational benefits for water-resource planning in South Korea and other regions with complex climates.

Looking Ahead

As climate variability increases, reliance on oversimplified models becomes riskier. Hybrid deep learning approaches provide a powerful alternative—one that adapts to local conditions, handles nonlinear processes, and performs well even under data constraints.

The future of hydrology is not about replacing physical understanding, but about combining it with intelligent data-driven tools. In the Nakdong River Basin, that future is already taking shape.

Complete Study: 

https://doi.org/10.3390/w18010032