Seeing below-ground with satellites and trees?
Two papers in today's post, both of which do something similar in very different ways. One of the challenges of hydrology is that direct observations of the things you care about can be difficult to make - whether that be a process like groundwater recharge, or the effective (plot or ecosystem-scale) water use of plants. In both cases, adding more than one kind of indirect observation into the mix can help provide more insight - or as modelers might call it, "constraints".
In the first of these studies, Yaojie Liu used sapflux, climate and soil moisture data along with to invert plant hydraulic models and estimate hard-to-measure plant hydraulic parameters. This required some clever manipulation of the models to make them behave themselves in the inversion process, but lead to some cool results - for example estimates of the P50 - the water potential in a plant at which 50% of xylem are blocked with air bubbles. Normally we have to measure this on a stem or branch section - so to get estimates at a "whole plant" level is really cool.
In the second of these studies, Simone Gelsinari explored whether incorporating satellite estimates of evapotranspiration improved groundwater model predictions, and in particular if there were more benefits doing this through data assimilation or through more conventional calibration.
Simone found that including ET in the model calibration worsened the fit of the model predictions to water table levels (see panel B below), but improved the mdoel estimation of ET fluxes (see panel A below) - suggesting the model might be closer to "getting the right answer for the right reasons" if this extra information was included. This was most important at sites where the vegetation was likely using groundwater, meaning ET was closely coupled to groundwater levels.
Including more data sources in modeling estimates can complicate parameter estimation and modeling strategies - the inversion methods, data assimilation and calibration approaches used in these studies are complex and labour intensive - but even these fairly early attempts illustrate some of the power of data-fusion approaches to learn and make predictions about plant-water systems.