 | |||
Ecologist Dan Nepstad of the Woods Hole Research Center is engaged in an activity that might seem crazy for someone who cares about forests as much as he does. For the past two years, this veteran of tropical forest research has been stealing the rain over two and half acres of forest in the eastern Amazon. Strangely, no one seems to mind. None of his colleagues, including fellow Amazon researcher and remote-sensing expert Greg Asner of the Carnegie Institution and Stanford University—whose career in tropical forest research began with the environmental group The Nature Conservancy—has tried to stop him. In fact, if you ask Asner, he’ll say the whole thing is a great idea. Nepstad’s ‘grand theft water’ isn’t supporting an exclusive tropical resort or even a hydroelectric project. In fact, he has no need for the water at all. He just doesn’t want the forest to have it. Nepstad and Asner want to know how much drought the forest can take before it begins to show signs of stress, what those signs are, and whether any of them can be detected from space. |
  Dan Nepstad and his colleagues constructed a unique structure, similar to a greenhouse, to deprive a small plot of the Amazon Rainforest of rain. This seemingly bizarre activity was part of a study on the effects of drought on the forest. [video (4.5 MB Windows Media)] (Photograph courtesy Dan Nepstad, video courtesy Paul Lefebvre) | ||
“We started thinking about simulated drought experiments back in 1994, when the Amazon was coming out of a major drought caused by a severe El Niño, and the forest almost completely ran out of water,” Nepstad says. The fact that the Amazon experiences drought often comes as a surprise to people. It’s the rainforest, after all; doesn’t it rain all the time? “That’s one of the most fascinating things about the Amazon,” explains Nepstad. “The east and southeastern parts of the forest actually go months each year with little or no rain. The trees survive by tapping soil moisture as far down as 20 meters.” During strong El Niño events, wet-season rainfall decreases, making the dry season even drier. Under those conditions, even the deep-rooted trees begin to suffer. Nepstad is concerned that longer and more severe droughts hover on the Amazon’s horizon. Some scientists are predicting that El Niño events will become more frequent and severe as Earth’s climate warms. Large-scale deforestation and smoke from biomass burning interfere with local cloud formation and rainfall. Identifying the precise signals of a drought-stressed forest would benefit the region’s farmers, timber operators, fire planners, and conservationists. Being able to detect those changes from a satellite would be a huge advantage; the Amazon is enormous and in many places still remote and difficult or impossible to survey on foot. |
In the always-green Amazon Rainforest, signs of drought are rarely obvious. To simulate the severe droughts the Amazon can experience during El Niño events, scientists diverted rainfall from an experimental forest in Brazil and then observed the impacts from ground-level and from space. (Photograph courtesy 3rd LBA Scientific Conference) | ||
 |
Nepstad’s and Asner’s shared interests in forest ecosystems led both of their careers to Brazil’s Large-scale Biosphere-Atmosphere Experiment in Amazonia (LBA, for short), the largest cooperative international scientific project ever to study the interaction between the Amazon Forest and the atmosphere and ultimately, the climate. NASA’s LBA-ECO program is one of numerous participants in the effort. The scientists funded under the LBA-ECO program concentrate on the processes and effects of land use change, often using NASA satellite data to add a wider view to what they observe on the ground. “Through LBA,” Asner says, “Dan and I realized we had common research interests. I was working on how to use remote sensing to describe the structure and function of forest canopies, and he was working on the impact of drought, fire, and logging. We realized that the drought experiment he was planning was the perfect opportunity to try to find some field-based and remote-sensing indicators of drought stress.” |
Although the Amazon Rainforest is evergreen, many places in the east and southeast go months each year with little or no rain. Vegetation survives these natural droughts by tapping moisture deep in the soil. The drought simulation occurred at an experimental site in the Tapajós National Forest in Brazil, near where the Tapajós River joins the Amazon. (Map by Robert Simmon) |
It’s a Jungle out There |
  | ||
This perfect opportunity wasn’t going to just fall out of the sky, though. It would require tremendous creativity, ingenuity, and sweat from Nepstad and the local carpenters, self-educated engineers, and laborers he hired to help figure out how to exclude rainfall from a 100-meter-by-100-meter (1 hectare) plot of rainforest. |
|||
It might seem easier to just go out and find a location that was experiencing a naturally occurring drought and then compare it to a location that wasn’t. The trouble with that idea is that the plots need to be similar to each either other in as many ways as possible—from the number and kinds of trees, to topography and altitude, to soil type—so that the scientists could be sure any differences between them were due only to drought stress. In a place as diverse as the Amazon, that kind of similarity is rare. Once you found two such places, you might have to observe the two for years—perhaps a lifetime—in the hope that at some point, one would experience a drought and the other wouldn’t. “If you look at all the trees in a 1-hectare plot with a minimum diameter of say, 10 centimeters, at least 2/3 of those trees will be a single individual of a species. That’s how diverse the place is. Just to find two 1-hectare plots that had several species in common, we had to survey 22 hectares,” explains Nepstad. |
 Hand-crafted wooden walkways gave scientists access to the upper levels of the forest where they could measure the effects of drought on the rainforest canopy. (Photograph courtesy Paul Lefebvre) | ||
Then the sites faced another test. “The first big challenge at any potential site,” he continues, “was whether we would be able to dig a 10-meter pit in the ground so that we could measure soil moisture at various depths. I had to carry along an auger to test-drill at potential locations.” There was no guarantee that the ground would be suitable for digging. Almost the entire Amazon Basin was once covered with a vast lake, and the region’s soils are dense clay formed from sediments that settled to the bottom over hundreds of thousands of years. Asner laughs as he says he is glad his part of the project didn’t involve much digging. “In the wet season, the soil turns to a thick clinging mud that sticks to shovels and boots and everything. In the dry season, it turns to brick.” Asner faced challenges of his own, though. The biggest problem with remote sensing in the Amazon is the clouds. “Even if imagery is mostly clear,” says Asner, “it seems like there will always be a cloud in the area you want [to see]. At some locations you might get only one or two cloud-free observations per year, which isn’t much, but it turns out to be sufficient for land use change and selective logging studies. It works pretty well for drought stress, too, because the one to two cloud-free overpasses are in mid to late dry season, which is the most drought-stressed time of the year for the forests.” |
 A researcher descends into a 10-meter-deep pit to measure soil moisture. Normally, rainforest trees survive months of seasonal drought each year by drawing on water stored deep in the soil. During severe droughts, however, the soil can become so dry that even deeply rooted trees suffer. (Photograph courtesy Dan Nepstad) | ||
 | |||
In with the NewThe satellite observations for the study came from NASA’s first satellite-based hyperspectral remote sensor. The difference between a hyperspectral sensor and a multispectral sensor is that a multispectral sensor detects electromagnetic energy in a sampling of broad slices (groups of wavelengths) of the spectrum while a hyperspectral sensor detects hundreds of very narrow slices of the spectrum that are contiguous, meaning that one slice touches the next, leaving no gaps. Called Hyperion, the hyperspectral sensor is flying on the Earth Observing-1 satellite. Hyperion is one of several new sensors that NASA is testing in an effort to produce smaller, less expensive devices with more capabilities than its current generation of sensors. The current Landsat satellites, for example, whose observations have been the centerpiece of high-resolution land cover and land cover change mapping for years, only collect observations of 7 different spectral bands (the broad slices of wavelengths of electromagnetic energy described above) reflected or emitted from Earth. Hyperion detects 220. If a multispectral sensor can be compared to you standing at a paint counter in a home improvement store asking for white paint, a hyperspectral sensor would be the clerk asking, “Did you want antique white, or colonial white, or off-white, or eggshell white, or ...?” |
About 50 kilometers south of where the Tapajós River joins the Amazon River, the Tapajós National Forest lies between the Tapajós River in the west and the Santarém-Cuiabá Highway (BR-163) in the east. Large, rectangular clearings punctuate the forest beyond the highway. The Tapajós National Forest is a 6,000-square-kilometer protected area where international teams of scientists carry out research into how the Amazon responds to human and natural disturbances. (NASA image by Robert Simmon, based on data Landsat 7 data provided by the Global Land Cover Facility) | ||
 |
 | ||
With all those wavelengths to choose from, Asner had a much better chance of detecting the changes in pigment activity, leaf area, and carbon balance that he suspected would change when the forest got stressed by drought. “The problem with remote sensing of vegetation conditions in the Amazon—well, aside from the clouds—is how lush everything is,” says Asner. Most satellite-based indicators of vegetation describe vegetation “greenness,” which is a general characteristic of vegetation that results from leaf area, canopy cover, and architecture. Greenness indicators are based on the relative amounts of visible light and infrared light being reflected from the forest. Chlorophyll and other pigments in the plant leaves absorb visible wavelengths (except green), while the chemicals that make up the leaves’ cell structure reflect near-infrared light. The trouble is that because it is so green in the tropical rainforest, the signal can get “saturated,” which means that above a certain level of greenness, it all looks the same to the satellite. Asner decided to test the traditional vegetation indicators against some that used the new information provided by Hyperion to see which ones did the best job of detecting the changes in vegetation brought about by Nepstad’s simulated drought. In addition to two traditional greenness indicators, he tested three new indicators that could only be made from hyperspectral observations: one that was sensitive to the amount of a chlorophyll-helper pigment called xanthophyll, one that was sensitive to a pigment called anthocyanin, and one that was sensitive to the water content in the leaves of the forest canopy. |
A hyperspectral sensor like Hyperion observes energy in a series of adjacent, narrow slices of the electromagnetic spectrum. Plotted on this graph, Hyperion’s observations appear as a continuous black line. Multispectral sensors like Landsat detect a single, average value (red dot) for a block of wavelengths (pink rectangles). For wavelengths near 800 nanometers, for example, Hyperion detects several peaks and valleys in the energy reflected from a surface, while Landsat records a single measurement. (Graph by Robert Simmon, based on data provided by the EO-1 science team) | ||
Stealing the Rain |
  | ||
While Asner was getting familiar with the new kinds of data coming from Hyperion, and designing and refining computer programs to process the observations and calculate the different vegetation indicators, Nepstad had located a suitable pair of sites in the Tapajós National Forest, a managed forest south of Santarem, Brazil, where the Tapajós River joins the Amazon. He was busy figuring out how to temporarily thwart Mother Nature. |
|||
 | |||
“I just love that kind of thing,” Nepstad says of the challenge of devising a plan to prevent rain from reaching the rainforest. “Challenging” doesn’t seem like a strong enough word to describe the situation. “We came up with panels—roofing plates that you can get at local hardware stores—suspended on a wooden structure [about one and a half to two meters off the ground] and tilted to run off into gutters. Each panel was half a meter by 3 meters, and there were 5,600 of them, made out of essentially clear, greenhouse plastic. Then there were 1,700 meters of gutters--a mile of gutter! All that flows into a trench around plot. The trench is lined with plastic, and the water flows off site into a gully about 300 meters away.” |
Scientists prevented rain from reaching the ground in one area (drought plot) and allowed rain to fall normally in another (control plot). Although the forest looks uniform from above, species diversity is so high that scientists had to survey 54 acres of tropical forest in order to find two, 2.5-acre plots with enough tree species in common for a good comparison. (Image by Robert Simmon, based on IKONOS data copyright Space Imaging) | ||
 |
The LBA team built a system of panels, gutters, and trenches to steal the rain from the Amazon Rainforest. Clear greenhouse plastic wrapped around wooden frames deflected rain as it fell ... | ||
 |
... into a network of gutters ... | ||
 |
... that drained into a narrow trench, which carried the water off the site. (Photographs courtesy Dan Nepstad) | ||
To minimize damage to the forest during construction, the 1- to 2-meter-deep trenches around the plot were dug by hand—all 1,500 meters of them. “Of course, as we dug them out, they became great congregating places for every kind of snake you can imagine, boa constrictors, everything,” says Nepstad with a short laugh. |
|||
Snakes weren’t the only animal visitors. “With the volume of wood and other materials we were bringing in with trucks, we quickly created lakes in the road, and pretty soon we started seeing eyes in them.” The eyes belonged to caiman, a kind of crocodile. “A number of times the workers would be hauling wheelbarrows of dirt from the trenches out to a pile, and find jaguars sitting on the dirt pile, staring at them.” “In all we probably brought in a volume of wood equal to about thirty percent of what was there in the forest itself. I would say about 100 tons of wood was brought in from local, legal,” he stresses, “saw mills. The people who were bringing in all this wood thought we were crazy.” Aside from the trenches they dug to divert water from the site, they also had to dig out 11-meter pits so that they could measure soil moisture at various depths throughout the course of the experiment. “It took three people six weeks to dig each pit, and there were ten of them. At the height of the construction, we had about 45 people employed at the site, and it took about a year and half to set it all up.” The team spent about a year establishing a set of baseline measurements for 12 characteristics in each plot: rainfall, soil moisture, pre-dawn and mid-day leaf water potential (indicates water stress), litterfall (leaves, twigs, flowers, etc, dropped from the trees to the ground), litter decomposition, leaf area and canopy openness, photosynthesis, flower and fruit production, stem growth, stem respiration (the release of carbon dioxide back into the atmosphere as the tree consumes the sugars and starches it creates during photosynthesis), gases emitted from the soil, and solution chemistry (the different kinds of chemicals that leach into rainwater as it drips down through forest and onto the ground). |
 Prolific native wildlife frequently crawled in and out of the study area. This caiman may look threatening, but it’s only about 30 centimeters (12 inches) long. (Photograph courtesy Paul Lefebvre) | ||
Artificial El Niño Gets Underway |
  | ||||||||||||||||||||||||||
By January 2001, the experiment was officially underway. Between January 7 and May 31 about 1,368 millimeters of rain fell over the two sites. Over the course of the wet season, rainfall averaged 9.5 millimeters per day, and the structure Nepstad and his crew had constructed diverted about 50 percent of that from the site, bringing the average down to 4.7 millimeters of rain a day, simulating El Niño drought conditions. After the wet season ended, the panels were removed. As the dry season progressed, the effects of the rainfall diversion became obvious. By the end of the dry season in November 2001, the amount of soil water available to plants at depths between 0 and 11 meters in the “drought” plot was several hundred millimeters less than the control site, where Nepstad didn’t interfere with the normal rainfall. Leaf area was 17 percent lower than it was at the control site, and mid-day leaf water potential was 30 percent lower. |
|||||||||||||||||||||||||||
 |
Not surprisingly, all this stress affected tree growth. Net primary production (the total amount of carbon that winds up in trees as a result of what they take in during photosynthesis minus what they give off during respiration) was almost thirty percent less in the drought plot than in the normal plot; the mass of carbon in the control plot increased by 2.6 megagrams, while in the drought plot, it increased by only 1.9 megagrams. |
At the field experiment site, rainfall (top graph) began in January, peaked in May, and dropped off dramatically in July. Plant available water (bottom graph) increased and remained high at the control site (top line) throughout the rainy season, while at the drought site (bottom line), available water was already declining by mid-February, bottoming out at 400 millimeters by the end of the dry season in December. Hyperion collected data over the plots in July and November (grey lines). (Graph courtesy Asner et. al.) | |||||||||||||||||||||||||
The fact that drought interfered with the tree growth isn’t surprising. What is surprising, says Nepstad, is where in the tree this slow-down occurs. “We thought that early drought stress would show up first in leaves—that leaf area would decrease significantly and that litterfall would increase as leaves died and dropped off the trees,” said Nepstad. “Instead, we found only small decreases in leaf area, and litterfall actually decreased. It turns out that wood production is the most sensitive to drought stress. Trees just stop growing in diameter, which has important consequences for timber production.” The second of Nepstad’s two big surprises was which trees were most likely to die as a result of the drought stress. It seemed logical that a smaller tree would have a harder time in a drought than a large tree, since the smaller tree’s root system couldn’t reach as deeply into the soil for water as a larger tree’s could. Instead, says Nepstad, “the first trees to die are the big ones, probably because they are in the sun high in the canopy.” |
 Scientists were surprised to discover that their simulated drought decreased wood production more than it did leaf area. Scientists numbered some individual leaves (top image) to monitor leaf area and mortality. Scientists also measured the girth of tree trunks (bottom image) over the course of the experiment. (Photographs courtesy Paulo Brando) | ||||||||||||||||||||||||||
Asner’s analysis of the Hyperion data confirmed that the commonly used indicators of vegetation greenness and leaf area just weren’t sensitive enough to detect the small differences against a background of such lush vegetation. When he calculated net primary production (net carbon intake) based on the traditional greenness observations collected by Hyperion, the results suggested that the carbon content in the drought plot and the control plot were the same. |
Before the experiment, the scientists thought that bigger trees, with their larger root systems, would survive the drought better than smaller trees. Instead, they discovered that the big trees that extend above the forest canopy (pictured at left) were the first to die. (Photograph courtesy 3rd LBA Scientific Conference) | ||||||||||||||||||||||||||
The indicators that made use of the new hyperspectral information from Hyperion were much more successful at detecting the changes in carbon content brought about by drought stress. When Asner factored in observations of xanthophyll pigment activity (increased xanthophyll activity is a sign that a tree is using light efficiently), anthocyanin pigment activity (reddish anthocyanin pigments are most visible in newly formed leaves and buds), and canopy water content, the satellite-based calculations of net carbon intake came very close to matching the growth that had been measured on the ground. |
Based on field measurements, net primary production (NPP) at the drought site was 73 percent less than the control site in 2001 (bottom row of table). Rows 2-4 show the ratio of NPP at the two sites estimated from different types of satellite observations in July, November, and for the entire year. Estimates based on hyperspectral observations of chlorophyll-helper pigments and canopy water content matched the ground-based measurements more closely than did the estimate based on greenness alone. (Table courtesy Asner et al.) |
Carbon Emissions at Stake |
 | ||
Asner’s results from the analysis of satellite data held some surprises as well. According to Hyperion, anthocyanin pigments in the drought plot were 60 percent higher than the control plot at the end of the rainy season in July. Then by the end of the dry season, they were 40 percent lower. “If anthocyanin is an indicator of leaf turnover and growth, why would we see an increase in the drought plot compared to the normal plot at the end of the wet season?” Asner wonders. He is eager to get to work on additional experiments that will help him understand the question. |
|||
 | |||
Asner is both surprised and incredibly pleased that the pigment and water absorption indicators were so successful in detecting the impacts of drought from space. As an unfunded—but, in his mind, absolutely critical—side project, Asner and some of his colleagues have been analyzing every single Hyperion image collected over the Amazon to document just how variable drought is across the region. “Drought in the Amazon is amazingly variable,” he says. That variability could be one factor in why studies of the carbon cycle in the Amazon sometimes report the forest as a carbon dioxide sink while in another location, it will seem to be a source. The issue is more than just an interesting puzzle for scientists to solve. Asner’s and Nepstad’s results suggest that even if mature Amazon forests are a net sink for atmospheric carbon, the process is extremely sensitive to even moderate drought stress. “What’s at stake here is about 70-80 billions tons of carbon--about a decade’s worth of human emissions—in the trees in the Amazon. Our results show that as temperatures and drought increase, much of that carbon may return to the atmosphere because growth slows way down and the biggest trees experience mortality for years afterward,” says Nepstad. The early signs of drought stress are not likely to be detectable with the current generation of multi-spectral satellite sensors, whose observations of vegetation are confined mostly to the visible part of the electromagnetic spectrum. The 220 wavelengths observed by Hyperion provide an almost mind-boggling number of new options. Hyperspectral sensors are the future of remote sensing,” says Asner, his voice filled with excitement. “It might sound unreal, even like science fiction,” he continues, “but we are talking about observing the vibrations, rotations, and interactions of molecules—like water molecules or pigments in tree leaves—from space. I think the most significant thing about this study is that it opens a whole new path for understanding how the forest is responding to the climate system.”
|
Scientists used the amount of light reflected from the control and drought plots at the beginning and end of the 2001 dry season to identify satellite-based signals of drought. The top graph shows reflectance measured by Hyperion at wavelengths between 450 and 1,250 nanometers in July (blue lines) and November (orange lines). The forest reflected very little visible light (450 to 700 nanometers) because that is the type of light plants absorb for photosynthesis. The lower graph shows the visible reflectance in greater detail. Dashed lines show wavelengths used for satellite-based estimates of carbon uptake: green for vegetation greenness, purple for light-use efficiency, and red for the activity of anthocyanin, a chlorophyll-helper pigment. (Graph by Robert Simmon, based on data from Greg Asner et al.) | ||