Water plays a central role in the dynamic world of ecosystems. The sensitivity of vegetation to drought, especially, hinges upon understanding water limitation. However, in trying to unravel this complex relationship, the scientific community has often been hamstrung by the immense variety of vegetation types, different climates, and the unpredictability of root zones. Our recent study, published in the scientific journal New Phytologist, sheds light on some of these open questions, but it also highlights where the knowledge gaps remain.
Deep learning illuminates plant water stress across biomes
By employing deep neural networks, we have taken a closer look at how vegetation responds when faced with drought conditions. We have been able to isolate and measure a water stress factor (fET) that indicates evapotranspiration (ET) reductions during drought. Notably, our results show varying ET responses to water stress. Indeed, the range is broad from the rapid decline of fET in some savannah and grassland sites to the subtle reduction in most forests. Precisely, we observed pronounced decreases in fET in savannahs and grasslands, at times plummeting to 10% of the rate observed under water-rich conditions. By contrast, most forests exhibited only slight reductions in fET, even when faced with significant water deficits.
But here’s the catch: in most arid sites, after an initial drop, the relationship between fET and the cumulative water deficit seems to level off (Figure 1). But why?
When initially presented with these findings, we wrestled with the question: was this due to increased xylem resistance at these sites, or could these locations access deeper underground water reserves?
The answer, or at least a part of it, was found in previous field studies, which revealed that vegetation, especially in drier areas, can sustain ET during drought due to groundwater or deeper soil moisture access. At the same time, plants also strategically adjust their stomatal closure based on the progression of water deficits. However, many conventional land surface models don’t account for these intricacies, leading to an incomplete understanding of water stress and its impacts.
Towards a global water potential network
And this brings us to a critical gap. Despite their undeniable importance, field studies and measurements, particularly those focusing on water potential (Figure 2, Figure 3), are not easily accessible and not commonly co-located with ecosystem-scale measurements. This lack of data poses challenges for researchers aiming to refine global understanding of water limitation in terrestrial ecosystems.
Imagine the breakthroughs possible if researchers had access to a standardized and comprehensive database of water potential measurements across various biomes! Our exploration into ET responses to drought using deep learning has shown promising results. But the road ahead requires collaboration, shared resources, and a concerted effort to fill the data gaps.
The scientific community would greatly benefit from a global network of water potential measurements, giving us the tools to better understand our planet’s ecosystems. This is what PSInet, an initiative to create a global water potential network, is doing.
ABOUT THE WRITER:
Francesco Giardina is a researcher based at ETH under the guidance of Sonia Seneviratne. He works at the interface of land-atmosphere interactions, ecohydrology and remote sensing. He is particularly passionate about applying novel machine learning techniques to understand plant responses to water stress under changing climatic conditions. For more information: Francesco Giardina | ETH Zurich