Phenology Laboratory / UNESP AND RECOD / UNICAMPImages from one of the cameras installed in the Cerrado savannah, with six plant species identified in white. Images at top, for example, indicate peroba-do-cerrado (Aspidosperma tomentosum) on the left, and pequi (Caryocar brasiliensis)Phenology Laboratory / UNESP AND RECOD / UNICAMP
Phenology, one of the oldest branches of natural science, studies plant and animal life cycle events and their relationship to climate. “It is artisanal work that begins with marking trees on the ground and then observing the emergence of leaves, buds, flowers and fruits every month,” says Professor Patrícia Morellato, coordinator of the Phenology Laboratory at the Institute of Biosciences, Universidade Estadual Paulista (Unesp) in Rio Claro. In traditional phenology, the collected data are correlated with seed-scattering by fruit-eating animals or pollinating insects that use floral resources. To go further and analyze the influence of climate on plants, we need a systematic field study that lasts three to five years on average.
“It’s tedious work that involves several people over a long period of time,” says Morellato. Since 2010 she has coordinated an innovative remote phenology project—referred to as e-phenology—in tropical areas, with funding from FAPESP and Microsoft Research Institute. A digital camera was installed atop an 18-meter tower in a Cerrado savannah area in Itirapina, in inland São Paulo State, and software and other tools were developed for remote observation and analysis of the information collected. The project partners are the Unesp Phenology Laboratory in Rio Claro, and the Reasoning for Complex Data Laboratory (RECOD) at the Institute of Computing, University of Campinas (Unicamp), where Professor Ricardo Torres is also participating in the research.
Beginning in August, five cameras will be set up in different vegetation types, such as campo, Cerrado, Caatinga, semideciduous forest and Atlantic Forest. “We’re going to do a study to assess the value of remote phenological monitoring of different vegetation types, as compared to traditional observation,” Torres says. The idea proposed by the researchers for a new phase of the project—an extension of the current project—is to use images obtained from Unmanned Aerial Vehicles (UAVs) to cover a much larger vegetation area. “We want to bring up new questions to analyze the impact of climate changes on tropical forests,” he says. The project has opened up the field of remote phenological research in tropical areas of South America. “There aren’t many cameras in the tropics, and to date there are no publications on this topic,” Morellato notes. One of the principal research groups that uses digital cameras and other remote monitoring technologies in temperate areas is headed by Professor Andrew Richardson of Harvard University.
PHENOLOGY LABORATORY / UNESP AND RECOD / UNICAMPAn 18-meter tower and digital camera have been installed in Itirapina, in the Cerrado savannah of inland São Paulo StatePHENOLOGY LABORATORY / UNESP AND RECOD / UNICAMP
The USA National Phenology Network, which is funded by the National Science Foundation (NSF), began monitoring the impact of climate on plant, animal and landscape phenology in 2007. Researchers from the network recently published a paper confirming that the Spring of 2012 began nearly a month earlier than the historical average since 1900. The phenomenon—termed false spring because a large percentage of vegetation was unable to achieve its expected growth for the season—was the result of warm weather that arrived early. According to the study, plants flowered very early, leaving several species vulnerable to cold waves that were still occurring. The regions most heavily affected included the Corn Belt states of Iowa, Illinois, Indiana and Michigan. The researchers concluded that, because of global warming, early springs are likely to become the new reality in the US.
Several remote phenological research groups are scattered across the globe, in countries such as Japan, the Netherlands, Australia, Canada and the United Kingdom. In Japan, for instance, computer-connected webcams set up around the country for generalized monitoring are also being used to observe plant cycles. Modern phenology owes a major debt to Swedish botanist Carl Linnaeus, who systematically recorded flowering times and the exact climatic conditions when flowering occurred for 18 locations in Sweden over the course of many years during the 18th century.
Data-checking
The starting point for the e-phenology project are data on more than 2,000 plants, obtained in monthly observations in a 260-hectare section of Cerrado vegetation in Itirapina beginning in late 2004. The vegetation is sampled in 36 transects—bands of land each measuring 25 x 2.5 meters that were demarcated earlier by the researchers—distributed in four different environments, two on the edge and two in the interior of the Cerrado. Each month, four researchers go into the field to individually sample each of the more than 2,000 plants. “We assess budding, leaf fall, flower and fruit,” explains Bruna Alberton, who took part in the field research during her master’s studies in remote phenology and continues to do so currently in connection with her doctoral studies.
The handwritten data are transferred to a worksheet, and only then are they entered into a computer in table form. “The use of remote technology will enable us to obtain an estimate of variations in the phenological pattern of leaf change over time without needing to go into the field,” says Torres, who coordinates the research at the RECOD lab. And the database being created will facilitate faster data-checking. The tower has a digital camera that takes five photos per hour between 06:00 and 18:00, as well as a climate station.
Leaf analysis, one of the methods used to study plant phase changes, was chosen to validate the use of remote monitoring technology. “We were able to demonstrate that, just as in temperate climates, remote monitoring of plants in the tropics produces results compatible with on-the-ground observations,” Torres says. Since tropical regions host many more species, there is more complexity involved, both in recognizing patterns and in understanding how climate affects phase change. “We’re working on validating the camera data against the on-the-ground phenological observation data,” says Morellato.
In addition, since the tropics do not have well-defined seasons, the phase changes are subtle. In temperate climates it is much easier to perceive changes, such as color changes in tree leaves or leaf fall. “Studies have already shown that in temperate regions, plant phenological cycles are affected by climate changes, but until we launched the e-phenology project, there were no initiatives that focused on understanding these cycles in tropical regions,” Torres notes.
Information on plant color is extracted using the RGB (red, green and blue) color channels in the image captured by the digital camera. “The analysis and processing of images to extract the color percentages is done at Unicamp, with assistance from the Unesp researchers,” says Torres. Since there are two distinct seasons in the Cerrado, dry and wet, it is possible to see the color variation in these channels over the course of a specific period of time. In the dry season, for instance, the leaves take on hues ranging from brown to russet, so there is an increase in the red channel.
The research group at Unicamp developed algorithms and new techniques based on image processing and computational vision to automatically identify individuals in the image. “Referenced against a given species of interest, our programs can determine which other areas of the image contain individuals of the same species, taking into account phenological patterns and using machine learning techniques,” Torres says. “Remote phenology has several applications, because it is possible to monitor species as a model in an area with different degrees of fragmentation, on a much larger scale than is done today,” comments Morellato. Since the meteorological data are also daily, it is possible to conduct coordinated analyses, which did not happen before. “We are currently working on showing how leaf changes are related to climate changes.”
Field app
A new representation of the phenological aspects of plants in images is revealed in scientific articles scheduled for presentation at international conferences. “We decode the information from several images taken over a long period of time into a single image—a technique called phenological visual rhythm,” says Torres. “The idea is that, instead of processing the entire collection of images to extract cyclical patterns, we can use a simpler, more compact representation to offer equivalent information.”
Torres, with the collaborative efforts of researchers working on interface design at the Institute of Computing have developed a smartphone app that will enable biologists to record their field observations directly onto a cell phone instead of on paper. “The data will be in digital format from the beginning, which will facilitate its inclusion in our database,” Torres comments. Various strategies for the app will be tested in the field this semester to determine which is most effective. Another avenue of research, known as time series prediction, identifies, among other things, the cut-off point of a change. “We’re trying to find out if the green variation in a species is related, for example, to climatological features such as rainfall,” he says. Also partnered on the project are the Federal University of Rio Grande do Sul, the Pontifical Catholic University of Minas Gerais and the University of Lorraine in France.
Project
E-phenology: the application of new techniques to monitor plant phenology and track climate changes in the tropics (No. 2010/52113-5); Grant mechanism Regular Line of Research Project Award – Global Climate Change Research Program – FAPESP/Microsoft agreement; Coord. Leonor Patrícia Cerdeira Morellato/Unesp; Investment R$331,023.44 (FAPESP).
Scientific article
ALMEIDA, J. et al. Applying machine learning based on multiscale classifiers to detect remote phenology patterns in cerrado savanna trees. Ecological Informatics. On-line version, July 4, 2013.
