WaterTalks: Geophysical Approaches for Quantifying Watershed Structure and Function by Susan Hubbard

Well, I have to say it's a pleasure to be here. I've heard such great things about Water for so many years, and it's fun to see old friends—I shouldn't say old, existing friends and colleagues—and meet new people. I heard a great series of lightning talks from the Ecohydrology group today, so thank you for inviting me and welcoming me here. I'm going to talk about some of the work in my group, which has really been focused over the last 15-20 years on developing geophysical methods to understand how subsurface systems function, but as we're also trying to understand terrestrial environments in the context of climate change, we're working from bedrock to canopy using geophysical methods. I'd like to recognize some of my colleagues shown there who are working closely with me in this geophysical group. I also mention two different teams here; you'll recognize that our group's geophysical work is done in the context of a couple big projects, and it's a really important part of advancing science to interface with our biology, geochemistry, and climate colleagues.

So what I thought I'd do is start out very simply: what is the motivation for trying to understand terrestrial ecosystems and watersheds? I'm sure this is rudimentary for many of you, but I wanted to start at a high level. Then I'll spend a few moments talking about geophysical work, hydrogeophysics in particular, and give a very brief overview without digging deeply into the methodologies. I'll spend a few moments talking about two different field study sites where I'll be giving examples: one in the Arctic in Barrow, Alaska, and moving to other parts of Alaska, looking at high-latitude permafrost systems, and another at the headwater river catchment in the Colorado River Basin. These are two different types of systems where we're trying to develop our methods. What I really want to get across are two different methods we are developing, moving a bit away from the hydrogeophysical work we've been doing to develop ways of integrating different types of data that sample different properties over different support scales, giving a holistic picture of these environments and how they behave as they're threatened by a variety of perturbations. In the Arctic case study, there are short-term perturbations like freeze-thaw cycles and longer-term warming and permafrost thaw; in the Colorado River Basin, we'll look at perturbations like floods, droughts, and episodic weather events. I won't spend a lot of time on that, but this is the direction of our research.

Starting at a very simple high level: why are we interested in understanding ecosystem science and particularly permafrost work? It really has to do with the carbon cycle. This shows carbon cycle 1.0 with the current magnitude of different fluxes across our Earth system. Before the Industrial Revolution, our system was in balance—what was put into the atmosphere was taken up by different parts of the Earth system. That has changed as we've brought up older carbon from geological reservoirs for combustion, leading to a buildup of carbon in the atmosphere. The Earth still plays a large role in taking up a lot of this carbon, both the terrestrial land sink and the ocean sink. There's a vulnerability at the heart of the work I'm going to talk about. As CO2 builds up, UV light is trapped, causing warming. A big uncertainty is the role of these land systems in continuing to serve as a large carbon sink. These plots show CO2 going up over time to the end of this century, with different IPCC models projecting ocean uptake and terrestrial ecosystem uptake. It's surprising that we are at a pretty early stage in understanding how the terrestrial environment will continue to function throughout the 21st century—will it serve as a sink or give off carbon? The bottom line is we really don't know.

That's part of the motivation for the ecosystem work, particularly the Arctic work. We're also interested in how changes in climate and associated attributes like early snowmelt, floods, droughts, and extreme weather influence the way watersheds function and deliver water, nutrients, and contaminants down gradient. This is a very difficult question, and something we're focusing on in our Colorado River Basin study. Basically, we're trying to understand ecosystem feedbacks to climate and climate influences on watershed function. We're recognizing that this is not simple—many biogeochemical processes are microbially mediated, so we have to understand the biological component in a very heterogeneous and dynamic template. We also have to bring climate-scale models down to scales where they can interface with a watershed and provide realistic information about precipitation and temperature. Today I'm going to talk about using geophysics to help bridge across some of these scales. It's by no means a silver bullet, but one piece in the arsenal to understand our connected system across different space and time scales. I'll try to get at both using geophysics to look at the structure of these watersheds and ecosystems, as well as their functioning over time in response to perturbations.

I'll talk about two large team-based projects. I'm at Berkeley Lab, primarily funded by the Department of Energy, which funds large team-based projects with iterations between experiments, observations, and modeling, driven by developing predictive frameworks. We have two projects I'm quite involved in. I lead the project called Genomes to Watershed, where we're trying to understand from fundamental biology how it cycles biogeochemistry, how it changes with environmental stress, and what that means for biogeochemical cycling, developing models to scale from that fundamental biology out to the larger system scale. On the other hand, we have another project called Next Generation Ecosystem Experiment, or NGEE Arctic, which goes from watershed to ecosystem up to climate. I'll speak about both. They are driven by where we are in science: we have very little in terms of computational tools to take us across the spatial scales we need, and we're still at an early stage understanding how much is enough. DOE recognizes this; these are large projects, multi-institutional, about $10 million a year for 10 years, to get at big questions like whether the vulnerable Arctic ecosystem will remain a carbon sink through the 21st century.

I wanted to say a little about the methods we're using. I won't spend too much detail on geophysics—Tony, you're here, and we have a great geophysics group at this institution. We've learned in the last decade that geophysics can provide great information about the critical zone or near subsurface, but there's quite a bit of work to take a geophysical signature as measured in the field into something useful. We measure things like electrical conductivity, dielectric constant, or seismic velocity, and there's a lot of work to translate that into understanding state, stocks, flows, residence times, and reactions in a heterogeneous template. Geophysics are nice because they provide spatially distributed information non-invasively, but they are indirect. There are several components to get from A to B: collecting the right data for the target of interest, developing petrophysical relationships that link geophysical observables to properties like water content or salinity, and finally inversion or integration approaches. I spend a lot of time on joint inversion or integration approaches, combining different types of data—geophysics, point measurements, above and below ground—to honor their sensitivity and spatial resolution and give a holistic picture of the subsurface.

There have been many advances in hydrogeophysics in the last 10-15 years, improving our understanding. One powerful approach is using geophysics in a time-lapse sense, collecting data at the same spatial location over time. As the system is dynamic with fluid movement and changes in geochemistry, differencing these data removes the geology and images what's going on, similar to medical imaging. This has been used for monitoring moisture, plume movement, and even estimating microbially mediated processes. For example, using time-lapse spectral induced polarization to estimate iron concentration. In conjunction with remediation-based technologies where we perturb biogeochemistry, geophysics provides great information about how complex biogeochemical reactions happen in the context of structured material.

I want to turn to examples of how we've used these data in two different systems, starting with the Arctic. Why work in the Arctic? This system contains a huge amount of organic carbon locked up in permafrost. Plants live and die, and in temperate environments organic carbon degrades rapidly, but in the Arctic degradation is very slow, so organic material gets buried in permafrost. Permafrost soils store almost as much organic carbon as soils in the rest of the world and more than twice as much as is currently in the atmosphere. The worry is that Arctic temperatures are changing rapidly, and as permafrost thaws, there could be a huge release of carbon into the atmosphere through microbial decomposition. In a simplified cross-section, we have permafrost—ground frozen for two or more consecutive years—with an active layer on top, typically half a meter or less, that freezes and thaws each year. When it thaws, conditions are right for microbial communities to degrade organic carbon, respiring CO2 and methane. With warming, there will be more permafrost thaw, a larger active layer, and microbes exposed to bioavailable organic carbon, potentially causing a huge pulse to the atmosphere. On the other hand, warming can also lead to plant growth that takes up CO2 and changes in land surface, like vegetation and albedo. From my perspective, this is a real tipping point question. The Arctic is one of the most vulnerable land systems, similar to the tropics, but in the tropics most biomass is above ground, while in the Arctic most carbon is below ground.

Why is it so hard? There's a huge range of scale and process complexity. Organic carbon is degraded by microbes inside soil aggregates, yet climate models have pixel sizes on the order of 30 km by 30 km. It's a grand challenge to think about how to scale processes from their native scale up to where we can simulate and parameterize climate models. Here we have an Arctic soil aggregate from Barrow, and from a microbe's perspective, carbon and water are distributed in the aggregate; the organism needs coincidence of all these to carry out respiration. These are the processes at small scales, yet when we go to the field, we have much different scales. I'll take us from the microbial scale up to measurement scales and finally to climate model scales to set the stage for the challenge.

Here is the land surface in an ice-wedge polygon region in Barrow, Alaska. At the local scale, we see ice-wedge polygons. Each subsequent figure encapsulates the previous in a yellow box. This region has low topography, ice-wedge polygons, drained thaw lake basins, and water distributed on the surface. Our conceptual model is that as permafrost degrades, it changes microtopography, which changes water distribution, which changes above-ground biology (vegetation) and below-ground biology (redox). Stepping out to landscape scales, ice-wedge polygons recede and thaw lake basins come forward. Finally, at climate model scales, we have the northernmost town in the US in the Arctic, with very cold continuous permafrost. We have to take what we understand about that microbe and how it's influenced by variable vegetation, redox, microtopography, hydrology, and geochemistry at meter scales, and parameterize models working at much larger scales. It's not just spatial scale and heterogeneity; processes are extremely complex. Here are different reaction networks for different compartments: permafrost degradation influences water distribution, which influences vegetation; soil methane production is a function of hydrology and geochemistry; and in the Colorado River Basin, different microbial communities mediate different reactions.

How are we going about studying this? We're a few years into the NGEE Arctic project. My role is to look at heterogeneity and scaling aspects, trying to understand how to represent these systems using different measurements. Geophysics is my main tool, but we also use stochastic methods. We start by collecting nested-scale experiments, wanting to move up to larger scales and parameterize pixels, but also coming down to local scales to understand what's going on as a function of microtopography. We work in an ice-wedge polygon environment. Ground freezes and thaws, and when it freezes, it can crack, and water gets in, forming ice wedges that bump up the land. When ice wedges thaw, the ground collapses, forming troughs and peaks. Some work has focused on how peaks, troughs, and rims of individual polygons influence biology or geochemistry, but not much looks at ice-wedge polygon landforms as a whole to scale up to model-appropriate parameterizations.

We start very small. We work with microbiology colleagues, pull cores from the ground, CT scan them to understand vertical distribution of mineral soils, ice, and water. We use different energies of CT scanning to estimate organic fraction, which is useful for constraining larger geophysics. We have about 100 cores from this site. We do laboratory studies, like freeze-thaw experiments to understand water distribution and the formation of saline striations and monolayers. We work with microbiology colleagues to understand where microbes might sustain life during winter in pockets of water, particularly saline-rich water. We've done a lot of geophysics—electrical, seismic, radar, induced polarization—collecting point measurements from the ground and from the air, constrained by many point measurements. The data are astonishingly beautiful from this site. We collect data over various times during the year, with autonomous systems measuring below and above ground, and we go out at certain times. We're interested in how systems respond to perturbations, both press and pulse. This is an extremely dynamic environment: hard cold winter, thaw with snowmelt, a short growing season, and freeze-up again. We collect data throughout the year, some autonomous, and take big campaigns collecting geochemistry, hydrology, biology, and geophysics together.

I'll show a few examples of that data, starting with more traditional approaches using one geophysical method to estimate one parameter. For example, estimation of snow at the land surface using ground-penetrating radar (GPR). GPR sends waves into the ground that bounce back at the base of the snow. We estimate snow thickness and snow water equivalent. Snow is important as a thermal insulator and influences how long the active layer stays unfrozen and microbial respiration, as well as spring runoff. Here, GPR-based estimates of snow thickness compared to point measurements show we can do a great job over about a kilometer-by-kilometer block. If we add LIDAR measurements of microtopography and GPR data, we can fill up the whole site and estimate snow thickness very well.

We've used electrical methods to tell us about active layer properties and permafrost properties. Electrical methods send a current into the ground and measure potential difference to estimate electrical conductivity, from which we interpret other properties. Along a 500-meter transect, the active layer is a thin top part; we see ice wedges, ground ice, and some saline unfrozen permafrost. We've used this to estimate active layer thickness and moisture content, comparing with point measurements and getting good information. We've also used seismic methods, particularly surface seismic waves, to estimate deeper permafrost structure. This is a longer-term question: how competent is deeper permafrost and how will it behave? We found that in this region, conceptualized to have very thick continuous permafrost, there are actually big sections (20 meters or so) that are unsaturated and not frozen due to saline incursion. This changes our conceptual model—if we only thaw through the first couple of meters, we could have devastating mass wasting and carbon cycle issues.

What we really want to do is develop new methods that don't use method-by-method, property-by-property approaches, but use all data together to understand how the system is structured and functions. I'm excited about two methods. The first is ecosystem zonation or functional zonation. As we go through these data sets, we realize each geophysical data set can be a proxy for something we care about that influences biology: active layer thickness, soil moisture, soil temperature, amount of ground ice. My conceptual model was to look at these geophysical attributes as a suite and see if certain suites spatially clump in the subsurface, recognizing they are proxies for things we care about. At the same time, I was interested in how land surface properties clustered spatially, because our hypothesis was that microtopography controls hydrology, which controls above and below-ground biology. I used geophysical data to look at spatial clustering in the subsurface and LIDAR-based metrics to understand land surface distribution. I found that there were certain clumps of attributes both above and below ground that were spatially coherent. This opened up our conceptual model to move beyond one compartment, one parameter, to get zones in the environment.

At this site, we took LIDAR data and used a watershed delineation approach to delineate polygons—about 2,000 ice-wedge polygons. Then we went into areas with intensive measurements, including geophysics and microbiology, to understand what controls variability: is it the trough, rim, and center features at the polygon scale, or can we find regions with unique distributions of property suites important for system function? We looked at statistics of variability as a function of polygon type (high-centered, low-centered, flat-centered) and as a function of center, rim, and trough. The analysis told us that polygon type—the larger-scale feature—controlled most of the variability, not the small-scale features. That helps with scaling. So we used this information to estimate functional zones in the landscape. Instead of 2,000 polygons, we have basically three zones (red, green, blue), each with a unique distribution of properties like active layer thickness, soil moisture, geochemistry, and soil temperature. This is exciting because it allows us to move beyond local scale to understand how different subsystems within a landscape behave differently.

Let's see if this works. We come back to a transect crossing a low-centered polygon, high-centered polygon, and transitional polygon. We understand the parameter distributions in each zone. Using geophysical transects, we define zonation. When we look at microbiology samples and chamber-based measurements of CO2 and methane flux from the ground surface, we see completely different microbiomes and greenhouse gas signatures in different regions. This is a nice approach to move beyond very small scales and start estimating parameters at scales relevant for larger models.

The second thing I'm excited about is, once we have these ecosystem zones and see hot spots of activity, how do these different regions behave as a function of perturbation? We've started developing sensing networks that autonomously sense below and above ground. For example, we use electrical resistance tomography laid out in different hot spots, collect soil-based measurements of temperature and moisture, and image the ground surface with cameras capturing RGB and near-infrared data to track snow, inundation, and vegetation growth. These data stream in daily, giving a beautiful picture of the system breathing, freezing, thawing, and vegetation growth. We're finding great correspondences between when the near surface starts to thaw and what's happening below ground, and when it thaws, what's happening above ground—from permafrost to canopy. We're starting to see dynamic relationships between soil moisture, active layer freeze state, salinity, and vegetation greenness. For example, correlation coefficients between electrical conductivity in the first half meter and greenness are on the order of 80-90% during the growing season, showing a strong connection between above and below ground. This pushes us to consider using above-ground canopy information as a biosensor of soil processes. We're extending this with drone-based methods. I see a push in terrestrial sensing that will change rapidly in the next 5-10 years.

I'll briefly talk about the watershed case study. The Arctic project focused on ecosystem feedbacks to climate via microbial respiration. Another big problem is understanding how climate- or weather-driven perturbations change watershed function. Watersheds are extremely complicated, with vegetation, soils, vadose zone, capillary fringe, groundwater, hyporheic zone, and surface waters. There's been work on ecosystems and climate, but not much on carbon or nutrient migration in the subsurface. The project I lead, Genomes to Watershed, focuses on developing multiscale simulation approaches that capture how a microbiome functions and evolves with environmental stresses, and what difference that makes to predicting larger system behavior. We started at a site called Rifle, Colorado, a floodplain next to the Colorado River, where we've done work on uranium biogeochemistry and moved into nitrogen and carbon. The team, including Jill Banfield, Harry Beller, Carl Steefel, Owen Brodie, Tetsu Tokunaga, Ken Williams, and others, has done a fantastic job understanding the value of genome-informed reactive transport models. We improved predictions of carbon export to the river by including metagenomic and metatranscriptomic information to refine reaction networks. We found deeper respiration in the vadose zone, particularly in the capillary fringe, and learned about the role of hot spots and hot moments. This environment is dominated by snowmelt delivering an oxygenated pulse of water, so we use it as a natural laboratory to study how the subsurface responds to that pulse, coupling biology, hydrology, geophysics, and geochemistry.

We are now moving to a larger site in the Upper Colorado River Basin, a headwater catchment. Here we need to understand not only the complicated subsurface system but also vegetation and how earlier snowmelt changes the coupling of ecohydrology and what it delivers down gradient. This is a newer push to go up to the watershed scale. We've been developing reactive transport models in the age of genomics, and I'll show examples of geophysics feeding into that. Where we're going is developing scale-adaptive models that recognize sub-compartments within watersheds—like subalpine versus floodplain—that may have different reaction rates and residence times and behave differently to perturbations like flood, drought, and earlier snowmelt. We're developing a scaling schema and model framework that recognizes these different subsystem contributions to the aggregated watershed response, so we can predict how the watershed aggregates responses and delivers water, nutrients, carbon, and contaminants down gradient.

In the first few years, we used geophysics to understand hot spots and hot moments. For example, using spectral induced polarization in the floodplain to identify naturally reduced zones—areas where floods wash fines and woody material overbank, which get buried and serve as a carbon source. We mapped these with geophysics and put them into the model to explore how important parameterization of these hot spots and the hot moment of spring runoff was for predicting biogeochemical cycles and export to the Colorado River. We've also developed inversion frameworks that consider different data types. For example, Fun Tran, a postdoc in my group, used time-lapse streaming ER data with other measurements to estimate moisture and temperature in the shallow zone important for microbial processes. This is the first inversion approach that coincidentally considers thermal, hydrological, and electrical raw data. In the first few years, we've done a nice job developing methods and understanding carbon cycling, subsurface hot spots and hot moments, what we can get from geophysics, and how to build models. The genome-informed reactive transport models improved predictions of biogeochemistry in that floodplain. Now we're moving up to the Upper Colorado River Basin, which is big. The idea of having detailed genomics data in many places or installing many wells is challenging. So we've been thinking about ways to conceptualize the site to focus on a few subsystems, like a hillslope or floodplain, and zoom into these using adaptive mesh refinement techniques. This allows us to telescope into the system where and when we need high resolution to predict the cumulative aggregated response of the watershed, including hydrology and biogeochemistry, to a perturbation.

We're just starting this work, but I'm excited. We're taking the measurements and approaches developed at Rifle and in the Arctic out to the watershed scale. We're using different types of ground-based and remote sensing data—primarily from Heruka and WRI in my group—to quantify watershed zonation based on maps that tell us where to set up intensive above- and below-ground monitoring. Then we assimilate this into models using weather data, aerial platform data, and below-ground measurements to understand and predict how the system functions. That's really all I wanted to communicate. We have two projects: one focused on Genomes to Watershed, one on Watershed to Climate. There are many interesting things going on across both. I tried to select a subset of the work I'm most involved in on the geophysics side. I think there's a lot of work to be done, but exciting directions moving beyond one parameter, one compartment to take a holistic look and identify functional zones in the landscape, and then watch bedrock-to-canopy processes across them. We also have other intensive field observatories with nice infrastructure and welcome collaboration in the lab or in the field. Thank you very much.

No, actually. We've taken our measurements from some of our intensive sites out towards the marine system in different directions, but only inland some kilometers. It's a really good question. We're also moving to the Seward Peninsula around Nome, Alaska, which is a less continuous permafrost region—a space-for-time analog, though other variables like topography and bedrock will make a difference. I think the permafrost long-term question is interesting. Most of our NGEE Arctic project is focused on soil and feedback, but there's a lot of work to be done. We and others are starting geomechanical modeling to think about permafrost degradation, and there's a frontier in coupling geomechanics with geochemistry, microbiology, and hydrology to think about future looks. We are on a peninsula surrounded by marine water, with marine spray coming in, and we've drilled down to confirm it's briny.

We haven't done that. We have more work to do on the deep permafrost.

Not so much. I showed that to highlight the work of a microbiologist in our group. We have folks doing freeze-thaw column experiments similar to what you're doing with hydrology and biogeochemistry. She is partnering with us to look at how the microbial community changes with freeze-thaw as a function of vertical gradients. By the way, she also showed that the microbiology community varied laterally as a function of geophysically identified functional zones, and they also vary significantly vertically as a function of zones we see in geophysics. So there's another frontier in understanding the spatial and temporal variability of these communities in the natural environment. The geological template holds value for that. Thank you.

Not too much. We've recognized, like you, this spring burp. As soon as we can get out there, it's really hard working conditions—freezing and dark. We have chamber-based measurements and eddy flux towers at these sites. The eddy flux tower is broken down for the winter. During this last year, we saw that as soon as we get thaw of the very thin layer of ice, we see this big spring burst. From some of our geophysics, you can see that these processes continue long after ice forms. Our team now recognizes we need to do more wintertime measurements; it's just hard to do. But I think it's important for the whole scientific community to figure that out, because as we and others are seeing, it could be a pretty decent fraction of the annual cycle.

That's a great question. We are considering these functional zones as suites of properties important to the process we're interested in. Here we're interested in soil respiration and fluxes. But when we go into the watershed, it's a different story—we'll look at slope, aspect, vegetation, soil water content, bedrock, and we're interested in residence times and different biogeochemical cycles. So it comes back to your conceptual model about what parameters you think are controlling the processes that are important, and then looking at the data through that lens. Thanks for highlighting that.

That's a big question. How do we constrain the model? In this project, we're just starting. Carl Steefel and Reed Maxwell are leading a component called the scale-adaptive approach. That means we can use adaptive mesh refinements to zoom in not just in resolution but also in physics where we need to—we can change physics in certain areas. First, we need a conceptual model about which subsystems of that environment might be behaving differently, from our watershed zonation aspect or from knowledge about life zones or biomes where we know there are big fluxes. Right now, we're building the big watershed model with adaptive mesh refinement capability. Within each subcompartment identified as high priority for spanning residence time and reaction rates, we're also developing submodels—a hillslope model, a floodplain model—and thinking about using those as scaling motifs. Within these more localized models, we use our geophysics, isotope data, and point measurements to parameterize the model. Even at that local scale, we develop the model and challenge it with data. For example, the genome-enabled reactive transport model is a certain type where Owen Brodie and colleagues developed an eco-trait model, where the microbial community functions as a function of environmental parameters and responds accordingly, with feedback between environmental parameters and the evolution and behavior of the community. That's an example of coupling an eco-trait model with a conventional reactive transport model. In each subsystem, we have different model formulations and parameterizations. In the eco-trait model, we parameterize with metagenomic and metatranscriptomic informed reaction networks. If we're looking at how Mancos Shale releases metals under different water or redox conditions, we might use more conventional core-based sorption data. So we have these submodels within the larger watershed model, and depending on the processes involved, we work at the local scale to understand and fix those models, then embed them within the larger model.

We've taken a variety of approaches to go from the raw geophysical signature through inversion and then understand the petrophysical response in the laboratory. For example, how does spectral induced polarization respond to gas content versus what's happening at the pore-mineral interface? For those measurements, we're still at such an early stage that you absolutely need laboratory experiments because there's so much ambiguity. For more established approaches like electrical responses to moisture or salinity, we can use field-based collocated measurements to help constrain them. Most of my approaches and our team's approaches rely on a Bayesian framework for integrating data. We have different types of data, but we understand their relationships, so they are all honored within this stochastic framework, and we end up with estimates of the properties we're interested in as well as their uncertainty distributions. That's been really nice because you can always say, for example, with geophysical data, you trust your well data—it's direct but sparse—and you complement that with geophysics, which is spatially dense but indirect. In terms of an estimation framework, that has worked really well for us. But it's a combination: for petrophysics, we use laboratory experiments and collocated field experiments to figure that out. Thank you.