Introduction

Climate change has recently been recognized as a critical global issue. Notably, the world experienced abnormally high temperatures in 2023 and the spring of 2024. Japan was one of the most affected countries1: its average temperature in 2023 was the highest since records began in 1891, surpassing the previous record set in 20202. The exceptionally high temperatures in Japan were also observed in the spring of 20243,4, illustrating the pronounced effects of climate change5. Such high-temperature anomalies underscore the urgent need to study the impacts of warming on ecosystems in Japan.

We focus on vegetation phenology — the seasonal events in plants — such as spring leaf flush (SOS: Start of Season), flowering, autumn coloration (End of Season: EOS), and leaf fall, all of which are strongly influenced by temperature6,7,8,9,10,11,12,13. Alterations in phenology are crucial indicators of climate change14,15,16,17,18,19,20. Furthermore, shifts in plant phenology can lead to phenological mismatches, where the timing of plant events becomes unsynchronized with those of pollinators or herbivores, ultimately affecting entire ecosystems21,22,23,24. Therefore, understanding how the elevated temperatures of 2023 and the spring of 2024 have influenced phenology is pivotal in addressing the challenges posed by climate change25,26.

Over the past few decades, vegetation phenology monitoring has heavily relied on satellite imagery, especially from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) sensors27,28,29,30,31. MODIS sensors have been in operation for over 20 years, far exceeding their designed lifespan (6 years), and their platform satellites (Terra and Aqua) are experiencing diminished orbital control, leading to unstable observation conditions for MODIS and compromised data quality32. Consequently, there is an urgent need to employ alternative satellite sensors to ensure the acquisition of high-quality observational data for recent years, including 2023 and 2024.

A promising alternative is the Second-generation Global Imager (SGLI) on the GCOM-C satellite (Global Change Observation Mission-Climate), operated by the Japan Aerospace Exploration Agency (JAXA). GCOM-C/SGLI began observations in 2018. Although it has exceeded its projected five-year design lifespan, it continues to deliver high-quality data reliably with precise orbital control. Several validation studies have highlighted its advantages over MODIS, particularly its higher spatial resolution (250 m) across multiple spectral channels33,34.

This study aims to assess the impact of the high temperatures in 2023 and 2024 on vegetation phenology by estimating SOS from 2018 to 2024 using GCOM-C/SGLI data. A seven-year dataset from 2018 may be too short to assess the long-term impacts of climate change on vegetation phenology. However, long-term variations in phenology are also influenced by non-climatic factors, such as land cover changes and vegetation transitions, which are not the focus of this study. Therefore, for the sake of detecting the impacts of the specific high-temperature events in 2023 and 2024, it is appropriate to use this dataset of the recent past. The SOS estimates were validated by the ground-based time-lapse camera data from the Phenological Eyes Network (PEN)35,36,37,38.

Materials and methods

Area of interest

Our area of interest is six regions in northern and eastern Japan: Hokkaido, Northern Tohoku (N. Tohoku), Southern Tohoku (S. Tohoku), Hokuriku, Kanto, and Chubu (Fig. 1; Table 1). According to Köppen’s climate classification, most areas in these regions, except for Hokkaido and Kanto, are classified as Dfa category, with only high-altitude areas falling under Dfb. In Kanto, most areas are categorized as Cfa, with high-altitude areas falling under Dfa. In Hokkaido, most areas are classified as Dfb, while high-altitude regions are categorized as Dfc39. These regions (except for agricultural and urban areas) are predominantly covered by deciduous forests, which exhibit distinct spring phenology signals. Outside of these six regions, primarily in southern and western Japan, the areas are mainly covered with evergreen forests. These forests do not exhibit leaf phenology as distinctly as the deciduous forests.

Fig. 1
figure 1

Map of the regional classification of northern and eastern Japan used in this study.

Full size image
Table 1 The regional classification of northern and eastern Japan used in this study and their Köppen-Geiger climate classification39. We sometimes refer to Northern Tohoku (N. Tohoku) and Southern Tohoku (S. Tohoku) together as the Tohoku region.
Full size table

Air temperature data

We obtained monthly temperature anomalies (Fig. 2) for the study period using a 1-km resolution daily average temperature dataset in the Agro-Meteorological Grid Square Data (AMGSD) provided by the National Agriculture and Food Research Organization (NARO). AMGSD is generated by spatial interpolation of observation points based on elevation and other geographical features, using in situ data from the Japan Meteorological Agency (JMA) mainly40. Due to the lack of data, we could not conduct the temperature analysis for the northern islands of the Hokkaido region. We calculated the anomalies using the 2018–2022 average as the baseline, considering the exceptionally high temperatures in 2023 and 2024. This study focuses on February to April (spring), as these months are in significant relation to SOS7. Figure 2; Table 2 show that the high-temperature anomalies (0.5 K) were recorded across five regions (N. Tohoku, S. Tohoku, Hokuriku, Kanto, and Chubu) in the spring of 2023, and three regions (N. Tohoku, S. Tohoku and Hokuriku) in the spring of 2024. In contrast, 2018, 2019, and 2022 consistently exhibited lower-than-average temperatures across all regions.

Fig. 2
figure 2

Anomalies of air temperature during the study period in the main islands of Japan each month. These data were made of Agro-Meteorological Grid Square Data (AMGSD) estimated from in situ data by the National Agriculture and Food Research Organization (NARO)40. The anomaly baseline is the average during 2018–2022 of each month.

Full size image
Table 2 Spatial average values of anomalies (in K) of air temperature from February to April (spring) in each region and year.
Full size table

Satellite data

We acquired the GCOM-C/SGLI daily atmospherically corrected surface reflectance product (version 3) from descending orbits, encompassing the main islands of Japan (tiles T0427, T0428, T0528, and T0529), covering the period from January 1, 2018, to June 30, 2024, through JAXA’s G-Portal website. We then performed reprojection of these data: Using JAXA’s “Map Projection & GeoTIFF Conversion Tool for GCOM-C/SGLI HDF Products” (SGLI_geo_map_linux.exe: version 1.2)41, we transformed the SGLI data from a sinusoidal projection to a geodetic coordinate system (latitude-longitude grid). We excluded pixels of low quality or those contaminated by cloud cover or snow/ice by referencing the quality flag (QA flag). Furthermore, we identified and extracted pixels representing specific non-agricultural vegetation types—Grassland, Deciduous Broad-leaved Forest (DBF), Deciduous Needle-leaved Forest (DNF), Evergreen Broad-leaved Forest (EBF), and Evergreen Needle-leaved Forest (ENF)—utilizing the land-use and land-cover map produced by JAXA (JAXA-HRLULC, version 23.1242,43; aggregated at a 250 m resolution).

Estimation of SOS (Start of Season)

Studies using satellites to monitor the SOS have utilized various vegetation indices, such as the Normalized Difference Vegetation Index (NDVI), the Enhanced Vegetation Index (EVI), and the Normalized Difference Greenness Index (NDGI)44,45. While NDVI and EVI are widely used, their reliability is compromised by uncertainties due to snowmelt46,47,48, making them unsuitable for spring phenology monitoring in regions with heavy snowfall like Japan48,49. Furthermore, although NDGI was originally developed for monitoring tundra vegetation45,50, its effectiveness has not yet been validated for forested areas in heavy snowfall regions, including Japan.

In this study we adopt the Chlorophyll Carotenoid Index (CCI) derived from GCOM-C/SGLI data. CCI performs similarly to the Green Red Vegetation Index (GRVI)51, which effectively represents both SOS and EOS with a GRVI threshold of 0 based on in situ data in Japan52. The primary distinction between CCI and GRVI is that CCI uses spectral channels (red and green) with narrower wavelength range than GRVI52,53. It has been shown that CCI derived from GCOM-C/SGLI closely aligns with ground-based observations, demonstrating high accuracy54.

For each GCOM-C/SGLI data, we computed CCI by the following equation54: 

$$\:\begin{array}{c}CCI=\frac{VN5-VN8}{VN5+VN8}\end{array}$$
(1)

where VN5 and VN8 are the daily atmospherically corrected surface reflectance observed by the VN5 channel (530 nm) and VN8 channel (672 nm), respectively. Because these two channels provide a spatial resolution of 250 m33, the CCI is also derived at a 250 m resolution. This represents a significant advantage of the GCOM-C/SGLI over MODIS, whose corresponding green bands have a coarser resolution of either 500 m or 1 km34.

In this study, by adapting the phenology detection method from GRVI52, SOS was defined as the last day CCI exceeded zero between March and June. In cases where CCI exceeded zero during data gaps, the midpoint of the missing period (rounded up to the nearest day) was assigned as SOS. We calculated SOS and its annual anomalies to analyze inter-annual variations. For the baseline of annual SOS anomalies, we decided to use the average SOS over the period from 2018 to 2022, following the policy we described in the previous section for air temperature anomalies.

Field validation of SOS (Start of Season)

Since optical satellite data are affected by aerosols, clouds, and other factors55,56, field validation is essential. We conducted field validation using the “Phenological Eyes Network (PEN)”, a long-term in situ observation network for seasonal and long-term changes in terrestrial vegetation35,36,37,38. In the target regions, there are three PEN sites that have enough spatial and temporal coverage necessary for the validation of this study located in Japan: Takayama Flux Observation Site (TKY), Fuji Hokuroku Flux Observation Site (FHK), and Tomakomai Satellite Validation Site (TOS). TKY and FHK are in the Chubu region, whereas TOS is in Hokkaido (Fig. 3; Table 3). TKY and FHK also belong to AsiaFlux57 and the Japan Long Term Ecological Research Network (JaLTER)58.

Fig. 3
figure 3

Location of Phenological Eyes Network (PEN) sites used for validation. TKY refers to the Takayama Flux Observation Site in the Chubu region, FHK is the Fuji Hokuroku Flux Observation Site in the Chubu region, and TOS is the Tomakomai Satellite Validation Site in Hokkaido.

Full size image
Table 3 Specifications of phenological eyes network (PEN) sites used for validation.
Full size table
Fig. 4
figure 4

Examples of digital camera images (camera ID: y_18d, TKY site, 2023. The bottom numbers represent day of year (DOY).

Full size image

TKY is a DBF located on the northwestern slope of Mt. Norikura in the Hida Mountains. Its main canopy species include Mongolian oak (Quercus crispula) and Erman’s birch (Betula ermanii). The understory is dominated by the dwarf bamboo (Sasa senanensis). The area is covered by snow from December to March. FHK is a DNF located on the northern side of Mt.Fuji, where the canopy is mainly composed of mature Japanese larch (Larix kaempferi), with a little mixture of other species, such as the red pine (Pinus densiflora), the dogwood (Cornus controversa), and Mongolian oak. The understory is dominated by shrub trees as well as ferns (Dryopteris crassirhizoma or D. expansa) and the dwarf bamboo (Sasamorpha borealis). It is intermittently covered by snow from December to February. TOS is a DBF located near the southwestern coast of Hokkaido, where the canopy species include the Mongolian oak, Japanese maple (Acer mono), and Japanese ash (Fraxinus mandshurica). From December to March, the area is mostly covered by snow.

We collected daily images taken by digital cameras (COOLPIX4300 or 4500, Nikon Corp.) installed at each site (e.g. Figure 4). The ID of cameras that we used were y_18d in TKY, t32_d in FHK, and t21_s in TOS. From these images, we calculated the Green Chromatic Coordinate (GCC), also known as the normalized Green ratio (rG) or Green Ratio (GR). GCC has been widely used in field validation and phenology monitoring studies59,60,61,62,63 and is defined by Eq. 2:

$$\:\begin{array}{c}GCC=\frac{G}{R+G+B}\end{array}$$
(2)

where R, G, and B are pixel values (digital number) of each channel (Red, Green, and Blue)59,63. GCC represents vegetation status63, typically taking a value around 0.33 during the leaf-off period and increasing to approximately 0.42 with leaf unfolding. To illustrate the time-series changes more clearly, a 5-day moving average of GCC was calculated, and the timing when GCC exceeded the value of 0.375 during the spring was defined as SOS based on ground-based observations. The threshold of 0.375 was chosen because it corresponds to the mid-stage of spring leaf flush phenology at each site. After this, we compared in situ SOS with the SOS derived from satellite data.

Results

Estimation of SOS (Start of Season)

The estimated SOS, derived from GCOM-C/SGLI data, generally displayed a reasonable geographic pattern, occurring later in regions of higher latitude and inland mountainous areas (see Fig. 5). The western part of the study area (outside the six regions depicted in Fig. 1) is predominantly covered by evergreen forests, which necessitates a different interpretation of the estimated SOS from our target regions.

Figure 6; Table 4 revealed that the SOS in the Kanto region for 2023 was the earliest recorded in recent years, while the SOS in 2024 occurred 3 to 5 days earlier on average in all regions except for Kanto. Notably, in northern Japan (Hokkaido, Tohoku, and Hokuriku), SOS in 2024 was the earliest observed in recent years. These results suggest that SOS advanced significantly in the Kanto and Chubu regions in 2023 and in the Hokkaido, Tohoku, and Hokuriku regions in 2024. In contrast, in 2019, SOS was delayed by an average of 6 to 10 days in all regions, excluding Hokkaido.

Fig. 5
figure 5

Average day of year (DOY) of spring leaf flush (SOS: Start of Season) from 2018 to 2022 derived from satellite data (GCOM-C/SGLI). The four circles represent key areas (World Heritage sites) in the later discussion.

Full size image
Fig. 6
figure 6

Anomalies of day of year (DOY) of spring leaf flush (SOS: Start of Season) derived from satellite data (GCOM-C/SGLI).

Full size image
Table 4 Spatial average values of anomalies of spring leaf flush (SOS: start of season; in days) for each region and year.
Full size table
Table 5 Spatial average values of anomalies of spring leaf flush (SOS: start of season; in days) for each key area and year.
Full size table

The unusually high temperatures in 2023 and 2024 may have significantly impacted some of the most valuable native forests in Japan. To assess this, Table 5 summarizes the spatial averages of SOS anomalies for key areas, including the area around Shiretoko in Hokkaido and the Shirakami Mountains in northern Tohoku (both registered as UNESCO World Natural Heritage sites)64, as well as the area around Nikko in Kanto and Mt. Fuji in Chubu (both recognized as UNESCO World Cultural Heritage sites and renowned for their natural landscapes)64. The locations of these areas are illustrated in Fig. 5. The results indicated that SOS shifted significantly earlier in 2023 in Mt. Fuji and 2024 in Nikko, the Shirakami Mountains, and Shiretoko. In particular, in the Shirakami Mountains, where valuable native beech (Fagus crenata) forests thrive, SOS in 2024 was estimated to have shifted approximately 9 days earlier than the average.

Relationship between phenological shifts and exceptionally high air temperature

We created a scatter plot to explore the relationship between spring temperature anomalies (Table 2) and SOS anomalies (Table 4). Figure 7 demonstrates that in years with higher-than-average spring temperatures, SOS tended to occur earlier. Table 6 suggests exceptionally high temperatures have a tangible impact on SOS, as reflected by the P-value of 9.3 × 10−7, obtained from a one-tailed test of the null hypothesis of r = 0.

Fig. 7
figure 7

Relationship between spring leaf flush day (SOS: Start of Season) anomalies and the average air temperature anomalies from February to April in each region. This figure is based on Tables 2 and 4.

Full size image
Table 6 Statistical summaries for the relationship between spring leaf flush day (SOS: start of Season) anomalies and the average air temperature anomalies from February to April (spring) in each region. r is the correlation coefficient. P is the P-value of the one-tailed test of the null hypothesis of r = 0.
Full size table

Validation of SOS (Start of Season) from in situ data (PEN)

Figure 8 showed that the SOS derived from GCOM-C/SGLI and in situ data differed by up to approximately 8 days. However, the SOS rankings from both sources were generally consistent. The SOS obtained from GCOM-C/SGLI and in situ SOS at each site are as follows.

  • At TKY, SOS derived from GCOM-C/SGLI ranked in ascending order was 2024 (DOY 124), 2018 (124), 2022 (124), 2023 (130), 2021 (133), 2020 (136), and 2019 (141). The GCC ranking was 2024 (129), 2018 (130), 2022 (132), 2019 (139), 2021 (141), and 2020 (142).

  • At FHK, SOS derived from GCOM-C/SGLI ranked in ascending order was 2023 (DOY 110), 2018 (113), 2021 (115), 2024 (116), 2022 (117), 2019 (124), and 2020 (126). The GCC ranking was 2023 (110), 2018 (112), 2021 (114), 2022 (116), 2024 (117), 2019 (123), and 2020 (124).

  • At TOS, SOS derived from GCOM-C/SGLI ranked in ascending order was 2024 (DOY 132), 2023 (134), 2022 (136), 2019 (139), 2018 (139), 2021 (145), and 2020 (145). The GCC ranking was 2024 (137), 2023 (141), 2022 (141), 2019 (142), 2018 (143), 2021 (143), and 2020 (146).

Fig. 8
figure 8

Time series of both the Chlorophyll Carotenoid Index (CCI) calculated from satellite data (GCOM-C/SGLI) and the Green Chromatic Coordinate (GCC) calculated from digital camera images (in situ data) and its images at each Phenological Eyes Network (PEN) site (Left: TKY (Takayama), Center: FHK (Fuji Hokuroku), Right: TOS (Tomakomai)) during the study period (2018–2024). GCC plots represent a 5-day moving average. In situ data for certain periods are missing.

Full size image

Discussion

Climate change is considered to transform various environmental factors, such as precipitation patterns and rising temperatures1,20. Among these factors, temperature is generally regarded as the primary control of plant phenology9,10,11, and we also found a clear correlation between spring temperature and SOS. Table 6indicated that a 1 K rise in spring temperature may shift SOS by an average of 4.4 days earlier nationwide. If this also applies to the long trend of warming, extrapolating to the late 21st century (average of 2081–2100), SOS is projected to advance by approximately 7 days under the RCP2.6 scenario, which predicts a 2 K temperature rise by the late 21st century (assuming the goals of the Paris Agreement are achieved)5, and by about 21 days under the RCP8.5 scenario, which predicts a 4 K temperature rise (assuming no additional mitigation measures beyond current efforts)5, compared to the late 20th century (average of 1986–2005). These trends align with the predictions made in previous studies8,65. They indicated a relationship between climate change and SOS as follows:

  • Fujimoto (2008) suggested that a 1 K rise in temperature shifts budburst by an average of 3.4 days earlier8.

  • Hadano et al. (2013) predicted that compared to the 2000s, SOS in the 2030s will advance by 7 to 12 days, and in the 2090s by 15 to 26 days65.

However, our observed relationship reflects an anomalous event that may not be a part of the long-term trends, so we may not necessarily apply this tendency directly to future projections. Additionally, the reliability of future SOS predictions depends on several factors66,67,68,69. One key factor is the accuracy of phenology models in simulating annual variations in SOS69. Given the complex, nonlinear relationships between SOS and various climatic and ecological factors, using a simple linear trend to predict future SOS may be inappropriate66,67,68,69. This issue should be carefully considered and further discussed in future studies.

Some regions, such as the Shirakami Mountains, experienced significant changes in SOS, including an advancement of 9.3 days in 2024 (Fig. 6; Table 5). This shift in SOS exceeds the projections from previous studies65, suggesting that such a large change could give stress to plants and lead to the decline of native forests which provide valuable ecological functions26,70,71,72,73. These findings emphasize the need for long-term and localized monitoring13. To meet this need, continuous monitoring and analysis using a combination of satellite sensors with long-term records (e.g., MODIS, Landsat), high-temporal resolution (e.g., Himawari), and high-spatial resolution (e.g., PlanetScope) will be essential next steps.

All-season phenology monitoring—including the EOS, effective cumulative temperature, and flowering— is also crucial74,75. Such phenology varies by plant species, making it essential to have a more detailed understanding of vegetation distribution. Notably, the JAXA-HRLULC dataset we used had inaccuracies, especially in subalpine fir forests, which were frequently misclassified as other types of forests or grassland. These forests are already experiencing habitat shifts due to climate change73. To fully understand these impacts, creating more vegetation-focused LULC maps is also crucial.

SOS estimation by GCOM-C/SGLI showed inter-annual variation which was consistent with in situ data. However, they had a discrepancy (bias) of up to 8 days. Such discrepancies have also been observed in previous studies45, and the following causes can be considered:

  • The large discrepancy is primarily due to missing GCOM-C/SGLI data caused by clouds and atmospheric conditions. Therefore, when conducting more localized analyses, it is necessary to take these influences into account55,56.

  • The small discrepancy is likely caused by differences in the spectral response function between GCOM-C/SGLI and the digital camera, as well as the differences in the spatial scale they observe34,76,77 (10–100 m for digital camera; while 250 m for GCOM-C/SGLI).

Discussing the robustness of CCI and the validation methods used in this study is also important. While many indices for phenology detection and methods for field validation have been developed, a clear and definitive approach to phenology monitoring has yet to be established34,77. Therefore, comparing the effectiveness of existing detection and validation methods for different vegetation types, along with developing new methods, will be crucial for future research.

Data availability

The GCOM-C/SGLI map projection & geotiff conversion tool can be downloaded via https://gportal.jaxa.jp/gpr/ ?lang = en. The GCOM-C/SGLI data can be found at the following link: https://repo.gportal.jaxa.jp/. The 250 m HRLULC v23.12 for Japan is available via https://www.eorc.jaxa.jp/ALOS/jp/dataset/lulc_j.htm. The PEN data are available via https://pen.envr.tsukuba.ac.jp/. AMGSD data are available via https://amu.rd.naro.go.jp/wiki_open/doku.php?id=start.