Figure 1 shows the time-series of vegetation indices of Lambir Hills National Park (primary tropical rain forest; 4°12’04”N, 114°02’21”E; Nakagawa et al., 2019) and the Lambir oil palm plantation (4°09’07”N, 113°57’58”E) in Sarawak, Malaysia in Borneo observed by the Sentinel-2A/2B–MSI (atmosphere corrected data) and Himawari-8–AHI satellites. Primary tropical rain forests, oil palm plantation forests, and secondary forests are typical landscape features in this area (Ichikawa, 2007). In addition, typical canopy surface images of Lambir Hills National Park are shown in Figure 2. We show only NDVI values for the Himawari-8–AHI satellite because we used the reflectance data at the top of the atmosphere (i.e., atmosphere uncorrected data). For the Sentinel-2A/2B–MSI satellites, we selected observation scenes with cloud cover ≤ 10%, but we plotted all values observed by the Himawari-8–AHI satellites. For the Himawari-8–AHI satellite, we plotted values observed at 3 h around the culmination time because there was no property in the data regarding the cloud contamination. Despite the occurrence of general flowering in Lambir Hills National Park in May 2019, which occurs every 1–4 years, and the color of the canopy surface changing from dark green to whitish green (Sakai et al., 2006; Azmy et al., 2016; Chechina and Hamann, 2019; Ushio et al., 2020), every vegetation index showed no clear seasonal change (low values may be affected by cloud contamination). During the general flowering period, no data were observed by the Sentinel-2A/2B–MSI satellites under clear sky conditions (Figure 1). MSI Green-Red Vegetation Index (GRVI; Motohka et al., 2010) values observed at Lambir Hills National Park were about 0.05 smaller than those at the Lambir oil palm plantation, reflecting the difference in color of the canopy surface between tropical rain forest (dark green) and the oil palm plantation (light green). Nagai et al. (2016a) reported that NDVI and GRVI values observed by the photodiode sensors installed at the top of a crane tower in Lambir Hills National Park were almost constant throughout the year. In contrast, the time-series of the ratio of R, G, and B digital numbers to the total RGB digital numbers extracted from daily canopy surface images taken at Lambir Hills National Park showed characteristics of leaf flush and flowering among individual trees (Nagai et al., 2016a). These results indicate the possibility that long-term continuous observations by future advanced optical sensors on board satellites with a high spatial resolution and high temporal resolution (high frequent) may detect characteristics of phenology for each tree species in a tropical rain forest. In fact, the RGB composite images observed by the PlanetScope constellation of satellites, which consist of approximately 180 microsatellites (as of 16 November 2022; a commercial endeavor),⁷⁸ with an approximately 3-m spatial resolution at a daily interval could detect the characteristics of flowering phenology among tree species in Lambir Hills National Park during the general flowering event in May 2019 (Miura et al., 2023). The swath range that can be observed at one time by each PlanetScope microsatellite is narrow (25 km), but the entire observation system of approximately 180 microsatellites resolved the trade-off between spatial and temporal resolutions. Some previous studies indicated this advantage of PlanetScope constellation of satellites observations in alpine, temperate, and tropical regions (Leach et al., 2019; John et al., 2020; Wang et al., 2020, 2023; Moon et al., 2021; Wu et al., 2021).

How can we detect the characteristics of phenology in tropical rain forests consisting of evergreen broad-leaved trees, with seasonality much less clear than that of deciduous trees, with optical sensors on board satellites? Previous studies reported that the RGB composite images observed by the Sentinel-2A/2B–MSI satellites, with the highest spatial resolution among the optical sensors on board public satellites, detected the color change on the canopy surface of Castanopsis sieboldii, Castanopsis cuspidate, and Lithocarpus edulis (evergreen oak tree species) caused by leaf flush (light green) and successive flowering (cream) in Japan (Nagai et al., 2020b; Shinohara and Nasahara, 2022; Shin et al., 2023). These plants are insect-pollinated flowers of Fagaceae and their flowers bloom across the whole canopy surface. The timing of flowering is different among these tree species (L. edulis is approximately one month later than C. sieboldii and C. cuspidate). In a tropical rainforest in Borneo (Lambir Hills National Park), Miura et al. (2023) reported that the spectral reflectance observed by the PlanetScope constellation satellites detected the characteristic of color change on the canopy surface of Dryobalanops aromatica, Shorea ochracea, Swintonia foxworthyi, and Pentace borneensis during a general flowering in 2019. These results indicate the possibility that satellite-based observations may be used to track the phenological timing and patterns of various tree species in tropical rain forests in tandem with ground-truth information.

Remote-sensing can detect the color of the canopy surface of various tree species, which indicate the characteristics of leaf traits (leaf size, leaf biomass, leaf thickness, amount of pigments in a leaf, and angle of leaves) and structures (tree structure and height) (Sims and Gamon, 2002; Luke et al., 2013; Noda et al., 2014; Asner et al., 2015; Noda et al., 2021; Rahmandhana et al., 2022). The characteristics of leaf longevity, which is explained by leaf-flush and leaf-fall phenology, were correlated with the type of photosynthesis, leaf traits and structures, and climate (Wright et al., 2004; Kikuzawa, 2005; Onoda et al., 2011; Kikuzawa et al., 2013). These facts indicate the importance of discriminating each tree species by referring to the characteristics of phenology and canopy structures and mapping the geographic distributions for each tree species at a broad scale. In Lambir Hills National Park, however, the geographic distribution of each tree species is heterogeneous due to microtopography (Lee et al., 2002). The collection of ground-truth information for various tree species, thus, is both an important and challenging task. Integrating in situ and satellite-based phenological observations should also result in the tree discrimination of forests in the Asian tropics.

As mentioned above, the Sentinel-2A/2B–MSI satellites have the potential to remarkably improve phenology observations in the tropics. The MSI observations occur at 5-day intervals, however, and in 2019, we only obtained seven scenes with cloud cover ≤ 10% (Figure 1). In addition, we confirmed that cloud contamination appeared in Lambir Hills National Park and the Lambir oil palm plantation images on 5 March, 10 March (only in the Lambir oil palm plantation), 25 March, 4 April, and 20 November (only in the Lambir oil palm plantation) by visually checking the RGB composite images. In addition, we confirmed that a cloud shadow appeared in Lambir Hills National Park image on 20 November. Therefore, it appears that the Sentinel-2A/2B–MSI satellites cannot observe the characteristics of phenology for each tree species in the suitable period under clear sky conditions. In contrast, the observation frequency of the Himawari-8–AHI satellite under clear sky conditions is much higher than that of the Sentinel-2A/2B–MSI (Figure 1). In fact, even though the sky image taken at 10:30 (LST) in Lambir Hills National Park, which was the time of the Sentinel-2A/2B–MSI satellites passage, showed cloudy skies, many other images showed clear skies at other times (Nagai et al., 2018).

In previous phenology studies that analyzed time-series of satellite-observed vegetation indices, the researchers used the 8-day, 10-day, and bimonthly composite data in order to eliminate noise and missing data caused by atmospheric noise and cloud contamination and smoothed the composite time-series data by applying some fitting functions (Zhang et al., 2003; Delbart et al., 2006, 2015; Erasmi et al., 2009, 2014; Pennec et al., 2011; Wu et al., 2014; Garonna et al., 2014; Buitenwerf et al., 2015; Kobayashi et al., 2016; Park et al., 2016; Gao et al., 2019; Jin et al., 2019). These analyses were based on the hypothesis that vegetation indices observed under clear sky conditions, which might show the most accurate value, were higher than those under cloudy and rainy conditions. Therefore, the smoothed values are estimated values but not true values. The values smoothed by applying the fitting functions may include two types of systematic errors: (1) an actual value was eventually eliminated, or (2) an actual missing value was eventually misread as a true value. For instance, in the case of an abrupt decrease of vegetation caused by a landslide, vegetation indices after the landslide may be eliminated as noise by applying some fitting functions. In addition, in the case of successive cloudy and rainy conditions in a certain period, smoothed vegetation indices may be misinterpreted as a decrease of vegetation. Thus, from a statistical viewpoint, we should avoid smoothing as much as possible.

The frequency of the observations of vegetation indices in East Asia by the Himawari-8–AHI satellite is 10-min intervals. If we only select the vegetation indices observed during the daytime that were little affected by the solar elevation angle (e.g., 3 h around noon), we can obtain many data observed under clear sky conditions (Miura et al., 2019). Figure 3 shows the relationship between NDVI values observed by the Himawari-8–AHI satellite and state of cloud cover based on in situ observed sky images in Lambir Hills National Park during 8:05 and 16:45 LST (UTC + 8) from 2015 to 2016. The occurrences of higher NDVI values (> 0.8) coincided well with the timings of in situ “very sunny” or “sunny” sky conditions, despite including missing data of sky images. The analysis of Figure 3 indicated that the number of days with clear sky conditions ranged from 57 to 96, which was much more than vegetation index observed by the Terra/Aqua–MODIS satellites (1–5 days per month in the southwest monsoon period [May–October] and 0–2 days per month in the northeast monsoon period [November–April] in Borneo; at a daily interval; Nagai et al., 2014a). In addition, the confidence in the change in vegetation indices, which was detected by the actual change of vegetation or not (i.e., systematic noise caused by atmospheric noise and cloud contamination), may also increase by checking high-frequency continuously observed data under clear sky conditions. So, if we can extract data observed only under clear sky conditions, we scarcely need to smooth by applying fitting functions. Despite a coarse spatial resolution (1,000 m), the vegetation indices observed by the Himawari-8–AHI satellite may indicate fairly accurate values.

At present, however, the Himawari-8–AHI satellite is still not suitable with regard to spatial resolution for phenology observations in tropical rain forests. In fact, despite the high frequency of observations under clear sky conditions, no characteristic change in the time-series of NDVI was shown in May 2019, when the general flowering occurred in Lambir Hills National Park (Figures 1, 2). The time-series in GRVI, which can detect the change of color of the canopy surface (Motohka et al., 2010; Nagai et al., 2014b), might capture some characteristic of temporal change. In Lambir Hills National Park, the time-series in the ratio of RGB digital numbers for each individual tree extracted from daily canopy surface images showed differences among tree species around the general flowering period (Nagai et al., 2016a). In contrast, those for the whole canopy showed almost constant values throughout the year (Nagai et al., 2016a), perhaps because not all individuals and tree species flowered at the same time. In this case, the target region of canopy surface images was within at most 100 m (Nagai et al., 2016a), but in the Lambir Hills National Park, which consists of over 1000 tree species (Lee et al., 2002), vegetation indices observed by the Himawari-8–AHI satellite can detect the average phenology of various tree species within a 1,000-m-by-1,000-m area. Therefore, for accurate phenology observations in tropical rain forests, we require an onboard optical sensor with a high spatial resolution to discriminate each tree individual (≤ 10 m) at high temporal frequencies (on the order of 10-min interval) to eliminate cloud contaminations in satellite data.

Despite the uncertainty caused by the heterogeneity of tree species and microtopography, previous studies indicated that the time-series of vegetation indices could be used to accurately detect the spatiotemporal variation of leaf flush and leaf fall in deciduous forests in Japan by validating the indices against long-term continuous in situ observed data (Miura et al., 2019; Yan et al., 2019). Despite the effect of microtopography (elevation) on phenology and the differences in timing and patterns of leaf flush and leaf fall among tree species (Inoue et al., 2014; Nagai et al., 2014b; Shin et al., 2021a), leaf flush and leaf fall within a narrow region (e.g., 1,000-m square) occur rapidly and nearly simultaneously. The Japanese government is planning to launch a geostationary satellite with an optical sensor with a 3- to 4-m spatial resolution. ⁹ Such future developments in optical sensors on board satellites with high observation frequency and high spatial resolution will remarkably improve the accuracy of phenology observations in tropical rain forests, where the phenological timing and patterns differ among the numerous and highly diverse tree species (Osada, 2018; Reich et al., 2004).

Figure 4 shows that the interannual variation of deforestation from 2001 to 2020 could be detected by analyzing the time-series of the daily GRVI observed by the Terra/Aqua–MODIS satellites [500-m spatial resolution; adapted from Nagai et al. (2014b)] in Indonesia in Sumatra, where marked land-cover and land-use change has occurred due to deforestation and expansion of oil palm plantations (Ichikawa, 2007; Fitzherbert et al., 2008; GEAS, 2011; Koh et al., 2011; Miettinen et al., 2011; Hansen et al., 2013; Carlson et al., 2014; Nagai et al., 2014a; Estoque et al., 2019; Najib et al., 2020). Here, we defined deforestation as having occurred at points where the ratio of number of days observed GRVI < 0 under clear sky conditions to total observed GRVI under clear sky conditions was above 80% (Nagai et al., 2014a). This hypothesis was based on the fact that GRVI < 0 after leaf fall or when there was no vegetation (Motohka et al., 2010; Nagai et al., 2014b). In the deforested area detected by the Terra/Aqua–MODIS satellites (Figure 4), we also identify that vegetation is sparse by visually inspecting the RGB composite images observed by the Sentinel-2A/2B–MSI satellites with a 10-m spatial resolution (shown in brown; Figure 5). In addition, we identify that GRVI observed by the Sentinel-2A/2B–MSI satellites showed under 0 (Figure 6). From 2001 to 2020, the deforested areas continuously expanded in Riau (central Sumatra; Figure 4C) and Lampung (southern Sumatra; Figure 4B). In Indonesia, the loss rate of primary forests has declined since 2016 according to the Secretariat of the Convention on Biological Diversity (2020); however, our analysis suggests that the deforestation is still ongoing.

Interannual variation of deforestation in the tropics has been detected by analyzing the time-series of vegetation indices observed by the Landsat series (30-m spatial resolution at 16-day intervals; Hansen et al., 2013). However, for those satellites capturing data at 16-day intervals, it is possible that no data are observed under clear sky conditions throughout an entire year especially in the Asian tropics, which is one of regions with active atmospheric water circulation. In addition, in the case of the ETM + sensor on board the Landsat-7 satellite, each observation scene always included partial missing data due to a systematic error in the sensor (malfunction of the scan line corrector; Wang et al., 2021). For the Terra/Aqua–MODIS satellite observations with a 500-m spatial resolution, we could detect large-scale land-cover and land-use change in the tropics, including the establishment of oil palm and acacia plantations after deforestation (Nagai et al., 2014a), but we could not accurately detect local-scale changes such as the loss of vegetation caused by a landslide, which typically occurs at an area smaller than a footprint of satellite data (1 pixel size of satellite data) with a coarse spatial resolution (500-m or 1,000-m; Miura and Nagai, 2020). In contrast, high-frequency observations by the Sentinel-2A/2B–MSI satellites with a 10-m spatial resolution at 5-day intervals may improve the accuracy of detecting interannual variation of land cover and land use in the tropics. Using the time-series of the vegetation index observed by the Sentinel-2A/2B–MSI satellites, whose data will be accumulated over a long period, should help us to more accurately detect the interannual variation of the geographic distribution of deforestation by applying the same analysis as was applied to the Terra/Aqua–MODIS satellites data.

To improve and validate the accuracy of satellite-based land-cover and land-use maps, we must collect ground-truth information at multiple points (Tsutsumida et al., 2019). One way to achieve this is by using digital camera images with the geolocation information, time, and date shot at multiple points that have been uploaded on the Mapillary website, ¹⁰ which is a crowdsourcing project. For some reported points, we can use images taken on different dates, thus allowing us to obtain evidence of land-cover and land-use change over a short period. Funada and Tsutsumida (2022) also indicated the usability of the street-level photographs published on the Mapillary to map the geographical distribution of cherry flowering in Fukushima in Japan. The text and images uploaded to the Degree Conference Project (DCP) website¹¹ are also useful ground-truth information regarding land cover and land use. Previous studies reported the suitability of information on the DCP for use in validating satellite-based land-cover and land-use maps (Iwao et al., 2011; Soyama et al., 2017).

Figure 7 shows Mapillary images at four locations that were typical landscapes in Indonesia in Sumatra, according to field notes recorded in 1978 (full details are mentioned later). Mapillary also contains many images taken at intervals along the route of a participant’s trip by using a car-mounted camera. Such uploaded data may have a geographic bias and be concentrated in areas that are strongly affected by anthropogenic activities (e.g., on streets). Compared with Mapillary, there is less systematic bias in the geographic distribution of target points published on the DCP website, which allows users to choose the intersection of latitude and longitude integer values in remote regions that have been little affected by human activities. However, the data volume of Mapillary (more than 1.8 billion street-level images as of 16 November 2022; See text footnote 10) is much larger than that of the DCP (about 1.32 million photographs as of 16 November 2022; See text footnote 11), making the usability of Mapillary superior. By actively uploading captured images with geolocation information on Mapillary, especially in the points and areas where images have not yet been uploaded, field scientists can help to reduce the missing areas of in situ observations.

Another important issue is how to obtain ground-truth information regarding land cover and land use in the past. This solution deepens our understanding of the conversion processes of deforestation and agricultural expansion. We may estimate the historical land cover and land use by examining the present conditions. For instance, we can assume that areas now covered by oil palm and acacia plantations in Sarawak, Malaysia in Borneo were once covered by tropical rain forests (Nagai et al., 2014a). Without ground-truthing, however, those values will always remain as estimates. To solve this issue, landscape descriptions in research field notes may provide useful ground-truth information regarding past land cover and land use. A group of Japanese scientists at Kyoto University launched the “Inheriting field notes” project (Takata et al., 2014; Yamada, 2015)¹² and have published 46,281 sets of digitalized field notes on their website. ¹³ Those field notes include field trips in the Middle East, the Mediterranean, Africa, South Asia, East Asia, Southeast Asia, and Oceania from 1967 to 2016 (mainly the 1980s and 1990s). At the time of the field trips, researchers could not use the tools to capture geolocation information, such as a Global Positioning System (GPS) receiver or a digital camera with this function. However, by thoroughly examining the time-series of objective descriptions in the field notes, we can roughly identify the landscape at a certain point.

For instance, by applying a text mining approach to field notes recorded in Indonesia in Sumatra in 1978 (1802 items) we analyzed the frequency of the words used (Yamamoto et al., 2015). The field notes included typical words regarding landscapes, such as rice paddy (total of 283 cases), banana (164 cases), cassava (136 cases), and rubber (185 cases), which allowed us to identify the landscape at that time. These words are also useful ground-truth information regarding the land cover and land use in Indonesia in Sumatra in 1978. We plotted the locations of rice paddy, banana, cassava, and rubber extracted from the field notes in Figure 5. In addition, by using the Mapillary images (Figure 7), we compared the land cover and land use in 1978 published on field notes with those at the present time. Despite the difficulty in checking precise geolocations, we could validate that there was no land-cover and land-use change at two points, where rice paddy and banana were recorded in the field notes. However, like the Mapillary images, many points recorded on the field notes may be located in areas that were strongly affected by anthropogenic activities. The retirement and decease of owners will accelerate the loss of personal analog data such as field notes. Rescuing and archiving of these analog data and information is an urgent issue (Shin et al., 2020).

In conjunction with the development of optical sensors on board satellites, researchers also need to gather ground-truth information obtained at multiple locations in various periods. The quality of satellite data depends on the accuracy and precision of atmospheric and geometric corrections of those data. Previous studies used locally collected data such as daily phenology images and spectrum data observed from towers (Nagai et al., 2014a,2020a,b; Nakaji et al., 2014; Lopes et al., 2016). Such ground-truth information provides accurate and precise data collected over a long period. However, the number of locations in the tropics where phenology images and spectrum data are being collected is still limited (Nakaji et al., 2014; Nasahara and Nagai, 2015; Lopes et al., 2016; Alberton et al., 2017; Nagai et al., 2018, 2020a).

Another way to obtain detailed ground-truth information from multiple locations for a broad-scale picture of historical changes in land cover and land use is to examine “social sensing data,” a type of big data. These include videos posted to YouTube and old television programs (De Frenne et al., 2018; Shin et al., 2022b), and text and photographs with geotag information posted to social networking services (e.g., Twitter, Instagram, and Flickr; Fernández-Bellon and Kane, 2020; Silva et al., 2018; Song et al., 2020; Yoshimura and Hiura, 2017). The interests and movement of people at various locations can also be tracked by analyzing the access statistics of Google (Google Trends: Takada, 2012; Proulx et al., 2013), ¹⁴ number of visitors at Wikipedia (Fernández-Bellon and Kane, 2020), and geolocation information of mobile phones (Chang et al., 2021; Pintér and Felde, 2021). For instance, the analysis of Twitter posts was useful for evaluating the spatiotemporal variation of the timing of leaf coloring in Japan (Shin et al., 2021b). Kotani et al. (2021) and Shin et al. (2022a) analyzed the time-series of Google Trends and/or Yandex statistics (a major search engine in Russia)¹⁵ to assess the spatiotemporal characteristics of people’s interest in the use of berries in Arctic and the Russian Far East regions, which the authors used as proxy data of ripening phenology. Likewise, people’s interests in oil extracted from illipe nuts (Borneo tallow nut), which are seeds of Dipterocarpaceae species (Blicher-Mathiesen, 1994), may be useful as ground-truth information for ripening phenology in Sarawak, Malaysia Borneo.

As an example of a detailed land-cover and land-use map in Asia with a high spatial resolution, the Japan Aerospace Exploration Agency has published the land-cover and land-use maps of Japan and Vietnam with 10-m or 30-m spatial resolutions by integrative analysis of data from multi-satellites such as the Sentinel-2A/2B–MSI, the ALOS–AVNIR2 (Advanced Visible and Near Infrared Radiometer type 2), the Landsat series satellites, and the ALOS2–PALSAR2 (Hirayama et al., 2022; Hoang et al., 2020). ¹⁶ However, these maps of Japan did not account for the interannual variation in land cover and land use. In addition, PFTs were classified into broad categories, such as deciduous broad-leaved forest and evergreen coniferous forest. To accurately evaluate the spatiotemporal variation of the heat, water, and carbon cycles and biodiversity in the tropics, and to understand the sensitivity of vegetation to environmental change and succession, accurate classification of PFTs and detection of their interannual variation are needed. For instance, traits of photosynthesis and evapotranspiration differ among ecosystems and tree species in the tropics (Ishida et al., 2005; Kenzo et al., 2004, 2006, 2011, 2015). The improved classification of PFTs and discrimination of tree species based on photosynthesis, leaf traits, and leaf and canopy structures are important tasks because these traits help to account for the sensitivity of the flowering, leaf-flush, and leaf-fall phenology and leaf longevity to environmental changes and succession.

The soil in the tropics is oligotrophic (Fujii et al., 2018), and land-cover and land-use change due to deforestation has strongly affected not only the heat, water, and carbon cycles (Carlson et al., 2014; Kumagai et al., 2013; Takahashi et al., 2017), but also coastal ecosystems due to the outflow of nutrients from the soil surface to rivers (Tanaka et al., 2021). To accurately evaluate the spatiotemporal variation of ecosystem functions, ecosystem services, and biodiversity triggered by anthropogenic activities and climate change, we need to improve our understanding of the interactions between terrestrial and marine ecosystems. The SeaWiFS (1.13-km at a daily interval) ¹⁷ and Aqua–MODIS satellites, which were launched around 2000, observe ocean color and allow for estimation of chlorophyll concentration (O’Reilly et al., 1998; Schollaert et al., 2003; Gregg and Casey, 2004; Siswanto and Tanaka, 2014; Groom et al., 2019). In addition, the Second Generation Global Imager (SGLI) on board the Global Change Observation Mission-Climate (GCOM-C) satellite (250-m at 2-day intervals)¹⁸ improved the accuracy and precision of ocean color observations (Murakami, 2016; Matsuoka et al., 2021). Further research should examine, for instance, the relationship between the interannual variation of deforestation and the spatiotemporal variation of the chlorophyll concentrations in the coastal areas of some river basins.

To understand the spatiotemporal variation of ecosystem services and biodiversity, we must evaluate the spatiotemporal variation of the interaction between land-cover and land-use change and anthropogenic activities. For this, the nighttime light data observed by the Visible Infrared Imaging Radiometer Suite (VIIRS) on board the Suomi NPP satellite (Day/Night Band [DNB]; 750-m at a daily interval; Elvidge et al., 2017, 2021)¹⁹ may be useful, as land-cover and land-use change has caused the spatiotemporal variation of nighttime light. In a study of 46 cities with more than 50,000 inhabitants, Ivan et al. (2020) reported the suitability of the Suomi NPP–VIIRS satellite-observed nighttime light to evaluate the geographic variation of income. Studies of tropical regions could use these satellite data to examine the relationship between the interannual variation of deforestation and spatiotemporal expansion of the nighttime light.

Despite the language barrier (Amano et al., 2016), the collection of in situ observed data and ecophysiological information in each country and region (especially non-English data and information; Nagai et al., 2016b; Takeuchi et al., 2021) will accelerate the development of integrative analysis and evaluation of in situ and satellite-observed data. As noted by Farley et al. (2018), researchers should aim to conduct more big data analyses by integrating citizen science, which has superior veracity; real-time sensor networks, which have superior velocity; in situ observed data collected by scientists, which have superior variety; and remote-sensing, which has a superior volume. The support of international scientific networking communities such as the Asia-Pacific Biodiversity Observation Network (Takeuchi et al., 2021)²⁰, the Asia-Oceania GEO, ²¹ and the East Asia and Pacific International Long-Term Ecological Research Network (Kim et al., 2018)²² is indispensable for the development of these integrated studies. Takeuchi et al. (2021) emphasized the necessity of satellite observations that provide the academic perspectives and evidence needed to implement natural ecosystem conservation policies. We further encourage the use of in situ observed data to improve the accuracy and precision of analyses of satellite observations and reinforcing the networking of research communities working with in situ and satellite observations in the Asian tropics (Dronova and Taddeo, 2022; Shin et al., 2023).