Plants act as an effective ‘biotic pump’ that recycles surface water back to the atmosphere through evapotranspiration (ET). Two major processes are involved in this recycling: leaf releases water to the surrounding air via its stomata (i.e., transpiration); and canopy directly intercepts rainfall that is subsequently vaporized (i.e., rainfall interception loss, Ei; Figure 1). Transpiration, which is known to be the primary ET component, has been extensively studied and seen significant progresses. The other vegetation-involved ET flux, Ei, has received much less attention and is often regarded as a non-beneficial water use, since it reduces soil and river water needed for plant or societal use. However, a knowledge of the size of rainfall intercepted by plants is important for forest planning and management policy making particularly in arid regions.
Figure 1. Canopy interception loss of rainfall.
Up to date, it remains unknown how much rainfall is intercepted by canopies from catchmental to global scales, and how this rainfall partitioning has been changing over time. This Ei flux cannot be directly measured by Earth observing satellites. In fact, Ei can be measured for individual trees, as the difference between gross rainfall collected above the canopy or at a neighbouring open land, and the sum of the throughfall and stemflow sampled simultaneously on the forest floor. Unfortunately, such field measurements are not helpful for the global mapping of Ei, because canopy capacity to intercept rainfall varies greatly across species, and depends on rainfall characteristics (intensity and duration) that also vary across space and time.
Our study, recently published at Nature Communications (https://doi.org/10.1038/s41467-022-35414-y), proposed an innovative framework to isolate Ei using global flux network of ET measurements and a physics-informed hybrid machine learning algorithm (known as hybrid model). Eddy covariance towers measure water fluxes at the ecosystem level as opposed to the tree level, but they just observe the overall ET flux composed of Ei, leaf transpiration and soil evaporation. The core concept for separating Ei is that, Ei occurs only during or shortly after a rain event when the canopy is wet. During this particular period, Ei is a primary ET component compared to transpiration and soil evaporation. In this sense, the Ei flux can be estimated as the excess ET occurring during rain events relative to a hypothesized value in the absence of rainfall (Figure 2).
The hybrid model is a powerful tool to estimate the hypothesized value without rainfall occurrence. This model is an advanced version of machining learning model, but has additionally represented some key physical processes and conserved surface energy balance. It has proved to outperform pure machine learning model. In our framework, two hybrid models are built on different inputs: one trained with wet conditions, while the other trained with dry conditions (that is, not seeing the wet spells when Ei occurs). The difference between their ET estimates is regarded as an estimate for the Ei flux. Importantly, this framework allows for upscaling the tower-based Ei estimates to the global scale, with the geo-spatial information of predictors available from Earth observations and climate reanalysis.
Figure 2. The conceptual framework for separating Ei from tower-observed latent heat flux (LE). Cyan bars show hourly precipitation (P), blue and red curves show the estimated LE with and without accounting for rainfall occurrence, respectively.
With this hybrid model framework, our study provides the first observationally constrained global estimate of Ei covering the last two decades. This product helps figure out the exact role of Ei in the terrestrial water cycle. Globally, we find that about 1.25×103 km3 of rainfall every year is directly intercepted and evaporated by canopies, which accounts for 8.5% of total land precipitation (P). Although this ratio (Ei/P) is not a big global value, it is however can be higher than 15% over some dry or cold regions. In this regard, canopy interception of rainfall has non-trivial influence on local hydrology.
There is a known trade-off between rainfall interception loss and rainfall available for recharging soil reservoirs, rivers and lakes. The former provides critical water resources for sustaining plant productivity and meeting societal water needs. The local knowledge of Ei thus informs more effective water management measures, particularly against the background that large-scale afforestation or reforestation practices are viewed as key natural solutions to combat climate change. The rainfall interception by newly planted trees represents a sizable net loss of locally available water. In dry regions, an extra 15% loss of rainfall from planted trees would pose an additional threat to the survival of vulnerable ecosystems. This could be a significant driver of observed decline in available surface water over some intensely afforested drylands, and must be accounted for in formulizing future afforestation programs.
In addition, we find that it is rainfall characteristics (frequency and intensity), rather than previously proposed vegetation cover, that primarily determines rainfall partitioning to Ei (indicated by the Ei/P ratio). Canopy has greater potential to intercept rainfall in drizzle and light rain conditions than during storms. This is because during heavy rain events, after reaching the maximum capacity of canopy water storage, the remaining rainfall would become throughfall or stemflow and no longer contribute to Ei. During the past 20 years, observations show that global rainfall remain almost unchanged but it has shifted to be less frequent and more intense. The shift of rainfall characteristics has induced a significant decreasing trend of Ei since 2000. In the warming future, state-of-the-art climate models project a robust global intensification of rainfall extremes and decrease of rainfall frequency. Hence, the observed decreasing trend of interception loss should continue in future, further increasing soil moisture and runoff.
Our observation-based Ei product has now been publicly available at https://doi.org/10.5281/zenodo.7309030. This Ei product, as a new input to the hydrology community, has the potential to benchmark model representation of Ei and constrain estimates of other surface water fluxes.