Review: Development of an in situ observation network for terrestrial ecological remote sensing: the Phenological Eyes Network (PEN)
- First Online:
- Cite this article as:
- Nasahara, K.N. & Nagai, S. Ecol Res (2015) 30: 211. doi:10.1007/s11284-014-1239-x
- 926 Downloads
The Phenological Eyes Network (PEN), which was established in 2003, is a network of long-term ground observation sites. The aim of the PEN is to validate terrestrial ecological remote sensing, with a particular focus on seasonal changes (phenology) in vegetation. There are three types of core sensors at PEN sites: an Automatic Digital Fish-eye Camera, a HemiSpherical SpectroRadiometer, and a Sun Photometer. As of 2014, there are approximately 30 PEN sites, among which many are also FluxNet and/or International Long Term Ecological Research sites. The PEN is now part of a biodiversity observation framework. Collaborations between remote sensing scientists and ecologists working on PEN data have produced various outcomes about remote sensing and long-term in situ monitoring of ecosystem features, such as phenology, gross primary production, and leaf area index. This article reviews the design concept and the outcomes of the PEN, and discusses its future strategy.
KeywordsRemote sensing Phenology Ground truth Biodiversity Vegetation index
Satellite remote sensing is a useful tool for monitoring terrestrial ecosystems (e.g., Muraoka and Koizumi 2009; Muraoka et al. 2013a). A single image taken by a satellite shows the spatial distribution of ecosystems, and a series of such images over time can reveal the dynamics (temporal changes) of the ecosystems. However, because these images are just electromagnetic signals and are often contaminated by various types of background noise, they only indirectly describe information useful to ecologists. Many remote sensing scientists and ecologists have focused on how to interpret satellite data in an ecological context, such as biodiversity, canopy structure, and productivity (e.g. Horning et al. 2010; Jones and Vaughan 2010).
List of Phenological Eyes Network sites as of 2014
Site information (name, country, period, sensors, institute)
Alice Holt, UK. 2009–, ADFC. Forest Research and U Edinburgh
AIST Complex#7, JPN. 2004 only. Sky ADFC only
Bordeaux, France. ADFC only
EGAT Tower, Thailand. 1999–2002. Film camera only
Fuji-Hokuroku, JPN. 2005–, ADFC, HSSR, SP. NIES and AIST
Fuji-Yoshida, JPN. 2009–, ADFC. FFPRI
Gwangneung, Korea. 2009–, ADFC, HSSR. Kangwon National U
Hawaii Volcano Thurston, USA. 2012–, ADFC. U Hawaii
Kranzberger, Germany. 2010–, ADFC. Technical U Munich
Kiryu Experimental Watershed, JPN. 2004–, ADFC, HSSR. Kyoto U
Biyala, Kafr El Sheikh, Egypt. 2011–2014, GWC. U Tsukuba
Karuizawa, JPN. 2009–, ADFC. Shizuoka U
Lambir Hills, Malaysia. 2009–, ADFC
Mase Flux site, JPN. 2005–, ADFC, HSSR. NIAES and U Tsukuba
Moshiri Mixed Forest, JPN. 2010–, ADFC. Hokkaido U
Mt. Tsukuba, JPN. 2008–, ADFC. U Tsukuba
Poker Flat Research Range, Alaska. 2011–, ADFC. JAMSTEC & IARC
RIHN, JPN. 2005–2006. Sky ADFC only
Sugadaira, JPN. 2005–, ADFC. U Tsukuba
Seoul Heonilleung Alnus Forest, Korea. 2010–, ADFC
Spasskaya Pad, Russia. 1997/08–2000/10, 2013–, ADFC. JAMSTEC
TERC Grass Field, JPN. 2003–, ADFC, HSSR, SP. U Tsukuba
Takayama Conifer Site, JPN. 2007–, ADFC, HSSR. Gifu U
Takayama, JPN. 2003–, ADFC, HSSR, SP. Gifu U & AIST
Tomakomai, JPN. 2004. ADFC, HSSR, SP. Destroyed by a typhoon. NIES
Tomakomai Experimental Forest, JPN. ADFC, HSSR. Hokkaido U
Teshio CC-LaG Exp Site, JPN. 2006–, ADFC. Hokkaido U & NIES
U of Alaska Fairbanks. 2010–, ADFC. JAMSTEC and IARC
Yatsugatake Site, JPN. 2011–, ADFC. JAMSTEC, Chiba U, JAXA
During the decade since the PEN began, the network has expanded and many studies have been conducted. In this article, we describe the design concept and outcomes of the PEN and discuss its future strategy. We do not cover details of each PEN site, instrumentation, or data because these are very large topics in themselves. Interested readers can access such information at the website mentioned above.
Design and establishment of the PEN
In the 1990s, there was a rapid increase in the use of satellite remote sensing in terrestrial ecology. At that time, global satellite sensors such as Landsat/Thematic Mapper (TM) and NOAA/Advanced Very High Resolution Radiometer (AVHRR) had already accumulated long-term records (>10 years), and computers capable of large-scale satellite data analysis became more affordable than ever. As a result, long-term records of satellite data were analyzed to identify the effects of climate change and landcover changes on ecosystems (e.g., Myneni et al. 1997). The long-term satellite data were also used to construct phenology models on continental or global scales (e.g., White et al. 1997). However, these approaches were based on a limited amount of field evidence, and researchers required long-term in situ data to verify the satellite image data.
In 2002, the Japanese Ministry of the Environment began the “Integrated Study for Terrestrial Carbon Management of Asia in the 21st Century Based on Scientific Advancement”, otherwise known as the S-1 project. That project adopted a “systems approach”, which combined field studies, numerical modeling studies, and satellite remote sensing of global ecosystems (Oikawa and Yamamoto 2013). For this study, it was necessary to clarify the “meaning” of satellite data. This led to the design and establishment of the PEN.
In general, it is difficult to attribute the “meaning” of satellite data to any single factor, because satellite signals are affected by various factors such as the amount of leaves, the leaf angle distribution, leaf optical properties, canopy shape, canopy structure, condition of the understory vegetation, sun direction, observation direction, topography, and aerosols in the atmosphere. However, many of these factors change in parallel through the seasons. Therefore, if the structure of the ecosystem is fairly constant from year to year (with little disturbance), seasonal changes in the patterns of satellite data in each pixel should also be fairly constant from year to year, except for small shifts in timing because of variations in climatic conditions. In other words, by associating seasonal patterns in satellite data with seasonal changes in the ecosystem in each pixel, we can use satellite data as a key index of the seasonal stage in that pixel, and make reasonable guesses about many biophysical factors. This is “satellite-phenology synchronism” hypothesis, and it represents the scientific target of the PEN. If it holds true in many situations, it enables the direct and ecological interpretation of time-series satellite data with the help of some in situ data. The main features of the PEN are described in the following sections.
Long-term continuous ground observations
Several ground studies have been conducted to validate satellite data, including First International Satellite Land Surface Climatology Project Field Experiment (FIFE; http://daac.ornl.gov/FIFE/fife.shtml) and BigFoot (e.g., Turner et al. 2006). However, most of them have been short-term or intermittent. In contrast, the PEN is a continuous and long-term approach to determine how satellite sensors capture ecosystem dynamics. In particular, long-term continuous ground observations are required to validate the “satellite-phenology synchronism” hypothesis.
The PEN is not a strategically organized group, but an open community. There are no strict rules that members must follow, and therefore, the arrangement of instruments differs from site to site. Nevertheless, by accumulating data from multiple sites, the members can conduct cross-site research and the outcomes attract new members. The PEN does not have a large or stable funding source; instead, it is expected that those who are interested in the PEN will join and work with their own resources. The members of the PEN work for mutual benefit by sharing field sites, information, training opportunities, skills, data, and sometimes spare instruments. The data policy of the PEN is that all original data and technical records are open to all members. In principle, anyone can become a member by sending an e-mail to apply to join the PEN mailing list. However, if a researcher wants to use PEN data from any PEN site or prepare presentations or publications using PEN data, she/he needs to obtain permission from two stakeholders: the site PEN administrator (who manages PEN sensors and data at that site), and the site PI (who manages scientific research at that site). The PEN does not have a “one-stop service” to provide permission on behalf of all stakeholders, because that is beyond its governance.
Collaboration with other networks
In the long-term observation of terrestrial ecosystems, there can be many monitoring targets and items. It is impossible to cover them all using the PEN alone. Therefore, the PEN needs to join and collaborate with existing ecosystem study networks; namely, FluxNet (http://fluxnet.ornl.gov/) and International Long Term Ecological Research (ILTER; http://www.ilternet.edu/). Many PEN sites have been located at the field sites of these existing networks. The active research community and good infrastructure (towers, electricity, and local area networks; LANs) at those sites greatly support PEN activities. By sharing inter-disciplinary data, the PEN and these other networks can greatly benefit each other.
The PEN sensor system
The ADFC provides time-lapse images of plants, landscapes, and the sky. Time-lapse photography can record various pieces of information easily and simply. Plant phenology events such as bud burst, blooming, leaf coloring, and senescence are recorded for individual plants or shoots in the images. Such images provide information on how each plant emerges, grows, dies, and disappears, and also record some disturbance phenomena, such as damage caused by strong winds or insect herbivores. Snow cover and snow melt are also recorded. Quantitative information such as LAI can be extracted from canopy photographs taken upwards from the forest floor. Sky photographs record the distribution of cloud cover, which is useful to evaluate cloud contamination of the satellite data.
The ADFC consists of a digital camera (CoolPix 4300 and 4500) and an FC-E8 fisheye lens (Nikon, Tokyo, Japan), which are popular devices for recording canopy structure. These instruments are stored in a water-proof container (Hayasaka Ricoh co. Ltd., Sapporo, Japan). The camera is continuously controlled by a personal computer. At field sites with a poor electricity supply, the “MC-EU1” remote code is used to control the camera equipment, instead of a computer. However, this device is often unstable and stops functioning, resulting in missing data. Consequently, it is seldom used now. Dr. Takahisa Maeda and Dr. Minoru Gamo (AIST) constructed a different system with a similar function to that of the ADFC, and started observations of forest phenology at several sites. Their network, “PhenoMon”, works collaboratively with the PEN.
Standard camera parameters such as white balance, ISO speed, and exposure have been determined for the ADFC and suggested. However, these parameters have not always been applied rigorously. In particular, white balance, which greatly affects image quality, was inconsistently configured for long-term monitoring, resulting in low data quality (Nagai et al. 2013a). These parameters can be reset and configured by computers. Hence, we can switch several sets of parameters one after another and take photographs with different configurations in parallel.
The HSSR measures the hyperspectrum of radiation from the sky, canopy reflection, and canopy transmission in visible and near-infrared light (e.g., Motohka et al. 2011). In general, paired observations of incoming radiation from the sky and outgoing radiation from the canopy are required to detect the spectral reflectance of the canopy. That is, the ratio of the outgoing spectrum to the incoming spectrum gives the spectral reflectance of the canopy. Normally, these measurements are conducted using two spectral radiometers: one pointing upwards, and the other pointing downwards from above the canopy. Therefore, this approach requires two appropriately calibrated sensors. The HSSR system of PEN uses a different approach, with only one spectral radiometer (mainly MS-700, Eko Instruments, Tokyo, Japan) mounted on a rotating stage (Hayasaka Ricoh Co. Ltd.), which flips upwards and downwards by computer control. The use of one spectral radiometer instead of two reduces costs, and eliminates the need to calibrate two sensors. Additionally, we have installed a HSSR on the forest floor below the canopy, at two sites [the Takayama site, Gifu University (TKY), and the Fuji Hokuroku Flux site (FHK)] to record the transmittance spectrum of the canopy and the reflectance spectrum of the forest floor.
Hyperspectral information is useful in plant physiology because it reflects certain physiological features of plants. It is also useful for remote sensing studies because it allows for simulations based on intrinsic spectral sensitivities of arbitrary remote sensing sensors in visible and/or near-infrared bands. Therefore, the hyperspectral information helps to interpret which features of plants can or cannot be captured by the satellite sensors. Another network, SpecNet (Gamon et al. 2006), aims to measure the hyperspectrum of plants. Although the PEN shares similar concepts with SpecNet, the PEN has a broader scope, allowing combinations with ADFC and/or SP. The hyperspectrum in the mid-infrared region is also monitored at some PEN sites [the Mase paddy field (MSE) and TKY sites] (e.g., Motohka et al. 2011). This is a unique feature of the PEN.
The SP (POM-02, Prede Co. Ltd, Tokyo, Japan) records the optical conditions of the sky above the field site. It provides information related to the composition of aerosols and their optical thicknesses, which are necessary parameters for atmospheric correction of satellite images. The information collected by the SP can be used to help compare reflectance recorded by the HSSR on the ground with that detected by satellite sensors.
We believe the combination of the three sensors is the easiest way to achieve the goals of the PEN. Focusing on only these sensors makes management cost-effective and simple. However, the PEN occasionally uses other instruments, such as photosynthetically active radiation (PAR) sensors, and can include short-term or intermittent field studies (e.g., on LAI, leaf spectrum, and leaf angles) to supplement information collected by the core sensors.
Although they are described as in situ observations, the core sensors only indirectly evaluate the target objects. In that sense, they are also a type of short-range remote sensing. To correctly interpret and validate the data from the core sensors, we need authentic in situ data, such as LAI, biomass, chemical composition, species composition, leaf optical properties (e.g., Noda et al. 2013, 2014), and physiological parameters, which are measured directly at the target. Such data is called “back data” in PEN. The PEN encourages the development of protocols for collecting back data, particularly LAI (e.g., Nasahara et al. 2008; Ishihara and Hiura 2011) and leaf optical properties (Noda et al. 2013, 2014).
A serious problem for sensor systems is how to retain their sensitivity and quality over a long-term period. Regular calibration is of course mandatory, but it is difficult to retrieve all sensors from multiple sites to conduct adequate maintenance. Furthermore, every sensor has a limited lifetime. A common problem in the PEN is that substitute sensors of the same type are sometimes unavailable because they have been discontinued by the manufacturer. Sometimes they can be purchased in internet auctions. However, this is not an ideal solution, because the old sensors lack some of the benefits of newer technology, such as less power consumption, better accuracy, better resolutions, and high-speed data transmission. Therefore, sooner or later, the PEN must switch sensors from old to new types. To bridge the gap between data collected with old and new sensors and to maintain continuity, a standard protocol for cross calibration of the sensors is required. This is a common problem with satellite sensors as well.
Storage and handling of PEN data
The PEN has developed an information system to collect and archive data from all of the PEN sites. This system mainly consists of a “PEN Site Server” at each PEN site, and one “PEN Mother Server” that stores all PEN data and information. These are generally operated with open-source softwares such as Linux, Apache, OpenSSH, ImageMagick, etc. The softwares, directory trees, and file naming rules are generally common among all of the servers.
Inconsistent naming of files and/or directory trees sometimes causes problems in archiving large amounts of data, such as consuming unnecessary storage volume by redundant files or data loss caused by collisions of file names. Therefore, the PEN has established strict naming rules for data files. For example, the site identification code (abbreviation) of PEN sites (Table 1) follows the AsiaFlux site code convention, describing each site with three alphabetic characters. Another example is that time zone information is included in the file names of ADFC images, because the JPEG-EXIF information table does not contain a field for time zone. Supplementary information such as site, sensor position, and sensor direction are also inserted in the filename. Applying such rules from upstream of the data line prevents unexpected collisions and duplications of files. Such rules are useful for data management, data retrieval, and automated analyses.
At many sites, the computers controlling the sensors are connected to the LAN and the internet. In the past, the PEN Site Server worked as the LAN gateway, and stored PEN data temporarily. However, more recently, most computers are directly connected to the internet and can send data directly to the PEN Mother Server. This has reduced the role of, and the necessity for, the PEN Site Servers.
The PEN Mother Server is at the University of Tsukuba. It receives all data uploaded from the PEN sites regularly, and automatically arranges the data in suitable directory trees. It processes the HSSR data to produce spectral reflectance. It selects and extracts a sample image from ADFC images taken by each camera every day, applies trimming and resizing, and produces a table of thumbnail images. This table is helpful to general users for indexing and referencing. The mother server also operates the PEN website. It stores and shares data and activity records (logs) from each PEN site, and manages information related to the entire PEN community, such as sensor history, engineering information, and training contents.
Data back-up is a serious issue. The PEN operationally stores daily data in two locations, and an entire back-up is made each year on a separate independent disk system. There is also some remote back-up of PEN data, that is, the back-up data are stored at different institutes in different cities so that the information is not lost even if there is a catastrophic event. Some real-time PEN data are backed-up in a grid computing system known as “GEO-Grid” (Yamamoto et al. 2010).
The maintenance log, which is a record of maintenance and problems in each PEN site, is important ancillary data. During long-term observations, many things can happen to sensor systems, such as power-downs, human errors, malfunctions, calibration, and replacement. It is important that these events are recorded for data quality assurance. From the beginning, the PEN has tried to share the maintenance log. At present, the maintenance log is kept at a website with a dynamic editing function (hiki, http://hikiwiki.org/ja/). For example, the maintenance log of the ADFC at the TKY site for 10 years contains approximately 1,000 events and 20,000 words. At the start of the PEN, its members were unaccustomed to keeping logs. Therefore, the quality of the logs in the initial period of PEN is unsatisfactory because there was a lot of trial and error at that stage. In general, keeping an accurate log is critically important for any type of long-term observation. However, in our experience, there is an insufficient emphasis on training to keep accurate logs in the scientific education system.
The PEN website also collects and organizes information on techniques and knowledge that are common to many PEN sites. It serves as a database for trouble-shooting, as well as an educational resource for new members. It contains an inventory of all PEN sensors, and keeps track of all cameras, spectrometers, and sun photometers by their serial number. It records the history (such as deployment, malfunction, repair, calibration, and update) of each sensor. This helps us to adequately respond to announcements from the sensor manufacturers, for example, product recalls.
To save data handling costs, the systems must be stabilized, automated, and monitored, so as to suppress occurrence of troubles. If both the sensor systems and data handling systems operate automatically with few troubles, few things must be fixed and recorded. From this point of view, investment in infrastructure, especially the LAN (internet), has considerable advantages in the long term. In fact, most sites that have accumulated high-quality data have introduced a powerful LAN.
Development of the PEN
From 2003 to 2004, the PEN was introduced to three sites: the Terrestrial Environmental Research Center in University of Tsukuba (TGF), TKY, and the Tomakomai Flux Site of the National Institute for Environmental Studies (TMK). All of these sites belong to AsiaFlux, which is the Asian subnet of FluxNet. In October 2004, the TMK site was destroyed by a typhoon and was relocated to the FHK site. The TGF, TKY, and FHK are the only sites equipped with the complete set of the core sensors (ADFC, HSSR, and SP). The PEN was introduced to the Kiryu Experimental Watershed (KEW) in 2004 and to the MSE site of the National Institute for Agro-Environmental Science in 2005. These sites are equipped with ADFC and HSSR, but not SP. Since then, the PEN has been introduced at other sites, mainly with ADFC alone.
At the earliest stage, the PEN was introduced mainly to Japanese sites. Internationalization of the PEN started in 2009, when ADFC and HSSR were introduced to the Gwangneung Site (GDK) under a collaboration with Kangwon National University (later, the HSSR was removed because of power trouble). In the same year, under a collaboration with the University of Edinburgh and Forest Research in the United Kingdom, two ADFCs were introduced to the Alice Holt Flux Site (AHS). So far, ADFCs have been introduced to sites in Alaska, Russia, Malaysia, China, and the USA.
Also, PEN has been collecting time-lapse photographs taken in the past at various sites, including the EGAT Tower Site (EGT; 1999–2002) in Thailand and the Spasskaya Pad Site (SSP; 1997–2000) in Russia.
The PEN is now a part of the ground validation system for the Global Change Observation Mission-Climate (GCOM-C) satellite, which is under development by the Japan Aerospace Exploration Agency (JAXA). Meanwhile, the PEN has been incorporated into a biodiversity observation framework (Secades et al. 2014).
As mentioned above, the PEN does not have strong governance in controlling site selection. The distribution of PEN sites is quite heterogeneous, and some important plant functional types are missing from its existing coverage. However, at least in Japan, the three dominant plant functional types are covered; that this, paddy field (MSE), cedar forest (TKC), and deciduous broad leaved forest (TKY and TOM).
Because the original instruments of the PEN are now available in the market, other groups can purchase and introduce them at their own sites. In fact, although not “registered” with the PEN, some sites are equipped with the same ADFC and/or HSSR systems as those used by the PEN. The PEN want to establish collaborative relationships with those sites as well.
Scientific outcomes of the PEN
Phenology and disturbance
Various signals related to phenology and disturbances, observed by both by satellites and on the ground, have been investigated using the ADFCs and HSSRs at PEN sites.
Plant phenology can be detected by using time-lapse digital photographs and RGB indices, which characterize the balance of red, green, and blue (e.g., Maeda and Gamo 2004; Richardson et al. 2007). Nagai et al. (2011a) showed that the phenology of multiple species could be detected quantitatively using RGB indices of time-lapse photographs taken in a single frame by an ADFC. These images were taken through the fish-eye lens directed downwards from the top of the tower. However, a high tower extending above the canopy is not available at all sites. Nagai et al. (2013a) tried to use photographs of the canopy taken upwards from forest floor by ADFC, and showed that they were useful to detect the phenology of deciduous broad leaved forest (TKY), but not evergreen coniferous forest (TKC).
Akitsu et al. (2011) investigated seasonal and year-to-year changes in a grassland (TGF) with time-lapse ADFC images, and showed which phenology event corresponded to each characteristic change in the RGB index. Mizunuma et al. (2011) tested several RGB indices to detect phenology of a beech tree at the MTK site, where fogs often result in poor image quality, and found that HUE was the most robust index. Mizunuma et al. (2013) captured damage caused by late frost in spring in ADFC images of a deciduous broad leaved forest (AHS). Choi et al. (2011) used time-lapse ADFC images and RGB indices to detect the phenology of a Korean deciduous forest (GDK).
Phenology can be an indicator for land use classification by satellite images. Zukemura et al. (2011) investigated seasonal changes in the Normalized Difference Vegetation Index (NDVI) in a rice paddy field (MSE) using HSSR data, and developed a method to detect abandoned paddy fields in satellite images (ALOS/AVNIR2) based on anomalies in the seasonal changes in NDVI.
Nagai et al. (2013c) and Inoue et al. (2014) investigated the timing of leaf flush (start of growing season; SGS) and leaf fall (end of growing season; EGS) for 10 years in ADFC images at the TKY site, and related these events to air temperature. As a result, both timings were modeled based on temperature. However, the sensitivity of these events differed among species, and there were larger errors in EGS than in SGS.
Among the spectral vegetation indices and RGB indices, the green–red vegetation index (GRVI) has shown broad utility and several advantages at many PEN sites. Motohka et al. (2010) analyzed HSSR data obtained at the TKY, FHK, TGF, and MSE sites, and showed that the GRVI detected the seasonal changes in each ecosystem more clearly and robustly than did other indices such as the NDVI. In particular, “GRVI = 0” was shown to be a simple and meaningful threshold to detect leaf flush and autumn color. Also, the GRVI can detect mild disturbances (caused by typhoons) to forests that do not result in tree fall. The seasonal pattern of GRVI can be a good indicator of plant functional types.
In 2004, a large typhoon hit the TMK site, felling most of the trees. Ide et al. (2011) showed that this large disturbance was detected more clearly by the GRVI than by the NDVI, the green-to-red ratio (GR), or the Enhanced Vegetation Index (EVI).
Nagai et al. (2012) showed that the GRVI could detect the phenology of evergreen trees, whereas the EVI and NDVI were insensitive. Nagai et al. (2014a) compared litter-fall data with various indices derived from HSSR and ADFC images, and showed the advantages of the GRVI. Nagai et al. (2014b) demonstrated the utility of the GRVI derived from satellite sensor (the Moderate Resolution Imaging Spectroradiometer; MODIS) and ADFC data.
As a phenology indicator, the NDVI has been widely used (e.g., White et al. 1997). However, using HSSR data taken at the TKY site, Nagai et al. (2010a) pointed out problems in NDVI approach caused by snowpack and understory vegetation. Because there is little correlation between the timings of snowmelt and leaf flush, and both cause rapid increases in NDVI, confusion between snowmelt timing and leaf flush timing is a possible and serious error when using the NDVI for phenology studies.
Signals of seasonal changes are less clear in evergreen plants than in deciduous species. Nevertheless, Nagai et al. (2013b) showed that ADFC images clearly detected seasonal changes in evergreen coniferous forests in Alaska (PFA; black spruce forest) and Japan (TKC; cedar forest). In addition, they demonstrated the significance of understory vegetation in phenology signals in a boreal forest.
On the basis of these findings, some studies on phenology have been conducted on regional scales. For example, Hadano et al. (2013) detected SGS over Japan using the GRVI derived from MODIS data, created a climate-driven phenology model, and predicted changes in SGS under a future climate scenario. Nagai et al. (2015) also used the GRVI derived from MODIS data to detect the timings of SGS and EGS in Japan, and revealed their dependence on latitudinal and altitudinal gradients.
Leaf area index, biomass, and canopy structure
The LAI describes amount of leaves on a unit area of vegetation, and is an important index that characterizes terrestrial ecosystems. It is also an important target of ecological remote sensing. At the TKY site, in situ direct measurements of seasonal changes in the LAI of various species have been carried out every year since 2005 (Nasahara et al. 2008). Based on these data, the following studies have been carried out: Kume et al. (2011) used HSSR data collected at the tower top and the forest floor, and found that the PAR/NIR ratio in transmitted light was a good indicator of PAR transmittance and canopy LAI. Based on this fact, LAI can be estimated by a two-band light sensor on the forest floor without deploying any sensors outside of the canopy. Saitoh et al. (2012a) revealed seasonal changes in the canopy extinction coefficient of the Lambert–Beer theory. Potithep et al. (2013) showed that the relationship between vegetation indices and LAI differed between the leaf-emergence period and the leaf-fall period. Mikami et al. (2006) carefully calibrated ADFC and developed an automated algorithm to estimate canopy gaps on the ADFC images taken upwards from below the canopy. Using HSSR data collected at the TKY site, Maki et al. (2008) evaluated the effects of evergreen understory vegetation at TKY on the satellite data, and estimated the distribution of understory vegetation around TKY. Inoue et al. (2015) investigated the usefulness of ADFC images to evaluate seasonal changes in the aboveground green biomass and foliage phenology in a short-grass grassland in Japan.
Validation of satellite data
Optical sensors on satellites cannot detect the land surface below clouds. Even if the surface is partly visible to the satellite, thin clouds or spotted clouds introduce large errors in the radiometric data acquired by the sensor. Therefore, it is important to detect and remove satellite data contaminated by clouds. Some studies have used ADFC and HSSR data to assess the accuracy of cloud detection, and evaluated the effects of data loss caused by cloud screening on phenology observations (Motohka et al. 2009; Nagai et al. 2011b).
Motohka et al. (2011) investigated the cloud-contaminated data remaining after the standard cloud-screening of MODIS data, by comparing the satellite data with ADFC images and HSSR data collected at the TKY site. They found that as much as 40 % of the MODIS data that passed the cloud screening were still contaminated by clouds. The strongest effect of this cloud contamination was on the NDVI, while other indices such as the EVI, the normalized difference water index (NDWI), and the normalized difference infrared index (NDII) were relatively tolerant to cloud contaminations. They also proposed a method to correct errors arising from the tower coming into the field of view in the HSSR data. This error must be inherent in any type of radiometric observation at flux towers.
Nagai et al. (2008) validated the “satellite-phenology synchronism” hypothesis by comparing NDVI time-series data acquired from a satellite sensor (MODIS) to ADFC images of the canopy. Their results revealed that some specific NDVI values corresponded to similar phenological conditions every year.
Ishihara et al. (2014) evaluated atmospheric correction algorithms for several types of vegetation indices derived from satellite data by using HSSR data collected at the MSE site. Murakami et al. (2011) investigated the effects of the spectral sensitivity of optical bands of satellite sensors on several vegetation indices using HSSR data collected at the TGF site. The aim of that study was to bridge the gaps between data collected by different sensors and to generate long-term consistent records of vegetation indices.
Yamashita and Yoshimura (2008, 2010) developed methods to estimate sky conditions (cloud coverage) and incoming PAR using sky images taken by the ADFC at the RHN site. Nasahara (2009) proposed a simple method to estimate incoming PAR using satellite data, and validated it using PAR data collected in situ at the TKY site.
To our regret, the SP has not yet played all of its expected roles. Consequently, we have not achieved our aim to systematically validate satellite remote sensing data using a combination of the three core sensors. To make full use of SPs, collaborations with atmospheric radiation scientists would be extremely useful and desirable.
Estimation of GPP
Modeling and estimating gross primary production (GPP) are important targets in ecological remote sensing. The simplest approach is directly relating vegetation indices to GPP. Especially in the late 2000s, the EVI was thought to be a good indicator of GPP. However, there is much uncertainty and noise in such relationships. Nagai et al. (2010b) investigated the EVI–GPP relationship using satellite data and HSSR data collected at the TKY site. They compared these data with GPP data collected at the flux tower at TKY, and found that the two largest sources of uncertainty were cloud contamination remaining after cloud screening, and changes in the EVI–GPP relationship from the leaf-emergence period to the leaf-fall period. Ide et al. (2011) used satellite data (MODIS) to show that the GRVI was a better indicator of GPP than the EVI at the TMK site.
Because GPP is largely limited by PAR, other studies have estimated not GPP, but the GPP/PAR ratio (known as light use efficiency; LUE), using some vegetation indices. The photochemical reflectance index (PRI; Gamon et al. 1992) is one main indices used for this purpose. Nakanishi et al. (2006) investigated the relationship between the PRI and LUE using HSSR data and flux data collected at the KEW site (Japanese cypress forest), and revealed the effects of phenology and light intensity. Nakaji et al. (2008, 2014) also investigated relationships between several vegetation indices (including PRI) and LUE using HSSR data and flux data collected at the TMK, KEW, and TSE sites, and the Pasoh site in Malaysia. One problem of the PRI is that it cannot be derived from most of the current satellite sensors. Ishihara et al. (2006) validated a substitute index that can be derived from current satellite sensors, using HSSR data collected at the TMK, TKY, and TGF sites.
Other studies have tried to predict GPP using more explicit approaches incorporating physiological mechanisms. Muraoka et al. (2010) used a process-based ecosystem model and showed that seasonal changes in the LAI and leaf physiology largely controlled the GPP of the canopy. These analyses were based on LAI data estimated by the PEN system and in situ leaf physiological data collected at the TKY site. Muraoka et al. (2013b) showed that the Canopy Chlorophyll Index (CCI) is a reasonable index for photosynthetic capacity based on HSSR data collected at the TKY site. Thanyapraneedkul et al. (2012) used a non-linear function to describe the light–photosynthesis curve (which is based on the relationship between canopy photosynthesis and PAR) and modeled it using satellite data as well as HSSR and flux data collected at the TMK, TKY, MSE, and FJY sites. Saitoh et al. (2012b) used ADFC images collected at the TKY and TKC sites and estimated SGS and EGS for inputs into a numerical carbon budget model.
Some studies have used images captured by the PEN camera to identify indicators of canopy photosynthesis. Saitoh et al. (2012c) compared RGB indices obtained from AFDC images collected at TKC and TKY with GPP data derived from tower observations. Mizunuma et al. (2013) showed that HUE corresponded well with the GPP data collected at the AHS site. Ide et al. (2011) showed that the GR and GRVI, both of which were derived from ADFC images, corresponded well with GPP at the TMK site.
Future tasks and prospects
The validation of the “satellite-phenology synchronism” hypothesis is still insufficient. This is mainly because there have not been large anomalies in climate at PEN sites in this decade. Because phenology events normally occur regularly every year in response to regular seasonal changes in meteorological conditions, the hypothesis needs to be tested in anomalous cases. In Japan, there was an unusually warm spring in 2002 and a cloudy summer in 2003 (e.g., Saigusa et al. 2010). Both of these anomalies occurred before most PEN sites started operating. The PEN sites should continue observations so that any future anomalies can be recorded.
In terms of validating satellite data, the PEN still has many problems. One of them is the footprint issue; that is, the coverage of in situ observations is much smaller than the coverage of satellite observations. For example, the HSSR footprint is approximately 10 m, while moderate-resolution satellite sensors, which are popular in phenology studies, have footprints of 100 m or larger. It is necessary to connect these scales by using high-resolution remote sensing, such as photography by unmanned aerial vehicles (UAV).
Another problem is insufficient coverage of the varieties of land cover types and plant functional types. To increase the coverage of PEN, the costs of PEN sensors must decrease, and they must become more stable and easier to handle. For this purpose, a time-lapse camera such as the ADFC plays an important role because it is less expensive than the HSSR and SP, and yet it provides many outputs. In fact, many phenology-oriented observation networks now operate “webcams” or time-lapse cameras connected to the internet (e.g., Internet Nature Information System in Japan http://www.sizenken.biodic.go.jp/; PhenoCam http://www.oeb.harvard.edu/faculty/richardson/phenocam.html; Wingate et al. 2008). The webcam networks have rapidly gained popularity in many services provided by governments and organizations, for example, in highway traffic control, tourism, meteorology, agriculture, landscape design, national park management, and disaster monitoring. These webcam networks can be useful for phenology observations, also. Some studies are now using these networks for phenology observations and snow mapping (e.g. Graham et al. 2010; Ide and Oguma 2010, 2013). We need to establish a technology to effectively collect, archive, and use the “big data” provided by these multiple sensor networks.
Interpretation of ADFC and HSSR data in a more analytical manner requires an integrated optical–ecological model. This model should include radiative transfer models including single leaf and canopy (e.g., Kobayashi and Iwabuchi 2008), a landscape model describing the geometry of forest stand structure, and an ecological process model (e.g. Sasai et al. 2005; Ito 2010). To construct such a model, an integrated set of back data, such as geometry data collected by unmanned aerial vehicles and laser profilers, are required. Such models will play important roles in simulations involving satellite remote sensing of ecosystems.
The sustainable operation of the PEN relies on its members. The PEN faces various technical problems on a daily basis at the PEN sites and the data archives. The PEN needs members who are proficient at using both old and new technologies, and who can keep updating the PEN systems technically over a long period. The PEN is a small and open community that cannot hire and support such people for decades. Even if the PEN can rely on some commercial services, it must avoid making the core system a “black-box”, because this would stifle innovation in a bottom-up style.
However, most field scientists have been trained insufficiently in such skills as sensors, information/communication technology, electric engineering, electronic engineering, quality control, data bases, and archiving. Therefore, they must learn these skills by on-the-job-training and by trial and error. Consequently, it is inevitable that there will be many mistakes and failures, resulting in data gaps and/or deteriorating quality of the long-term record. This is a problem faced by many field networks, including the PEN.
The PEN started with the simple desire to record daily changes in ecosystems and compare them with satellite data. The long-term daily record of ecosystems, regardless of whether it is collected in situ or by satellites, is an indispensable and fundamental source of information for society, because the science is based on real data and interpretation of such data by using unbiased logic. We believe that this concept is shared by most ecologists, and that the PEN outputs can be easily shared with other scientists and citizens via the internet. If satellite remote sensing becomes more popular in ecosystem studies, and many ecological field sites become equipped with sensors like those used in the PEN, the PEN will be absorbed into routine research activities and complete its objectives.
This research was supported by the Global Change Observation Mission (GCOM RA4 PI#102) of the Japan Aerospace Exploration Agency (JAXA).