Spatiotemporal variability of vegetation response to meteorological drought on the Korean Peninsula

Meteorological drought is an extreme climatic phenomenon characterized by below-normal precipitation and high evapotranspiration over months to years (Won et al. 2021) and is a major natural hazard that can have direct and indirect devastating impacts on ecological (Gampe et al. 2021), agricultural (Lesk et al. 2021), environmental (Chiang et al. 2021), energy (Watson et al. 2022), and economic (Naumann et al. 2021) sectors. In addition, droughts are expected to become increasingly severe as climate change intensifies (Kim et al. 2011a).

The types of drought are broadly categorized into meteorological drought due to lack of precipitation and/or excess evapotranspiration (Mohammadi 2023), vegetation drought or agricultural drought due to lack of soil moisture (Kim et al. 2011b; Ejaz et al. 2023), hydrological drought caused by decreased river flows, reservoir inflows, and groundwater (Lin et al. 2023), and socioeconomic drought caused by the gap between water demand and supply (Lee et al. 2022). In general, a sustained meteorological drought caused by insufficient precipitation and/or excessive evapotranspiration leading to a lack of soil moisture results in a vegetation drought or agricultural drought (Kim et al. 2008), and a hydrological drought occurs when it leads to a decrease in runoff (Jung et al. 2022). In other words, most droughts in hydrological, groundwater, vegetation, agricultural, and socioeconomic environments originate from meteorological droughts (Guo et al. 2019; Han et al. 2019; Jung et al. 2022; Won et al. 2022b), and this process of transmission from one type of drought to another is called drought propagation (Xu et al. 2021).

The types of drought covered in this study are divided into two main types: meteorological drought and vegetation drought. Meteorological drought identification is usually performed using drought indices. However, the same type of drought can be identified differently depending on the drought index applied. In this study, we categorized three aspects of meteorological drought that affect vegetation health. The first is meteorological drought stress due to lack of moisture supply from the atmosphere, and we introduced the standardized precipitation index (SPI) (McKee et al. 1993) as a meteorological drought index to express it quantitatively. The SPI has been widely used as an index for meteorological drought monitoring and analysis (Kim et al. 2011c; Won et al. 2020b). The second is meteorological drought stress due to increased moisture demand from the atmosphere, and evaporative demand drought index (EDDI) (Hobbins et al. 2016) was used as a meteorological drought index to explain this. In recent years, evapotranspiration has received increasing attention in the field of drought monitoring, with studies focusing on the importance of this aspect of atmospheric water demand (Won et al. 2018; Won & Kim 2020; Seo et al. 2022). The third is meteorological drought stress, which considers both the lack of moisture supply and the increased demand for moisture from the atmosphere. Meteorological drought indices such as precipitation-evapotranspiration index (SPEI) (Vicente-Serrano et al. 2010) and copula-based joint drought index (CJDI) (Won et al. 2020a) have been applied to quantitatively represent this, and there are also studies that consider both the moisture supply side and the moisture demand side of the atmosphere (Won et al. 2020c; Ejaz et al. 2023). In addition to the three aspects of meteorological drought indices used in this study, there are also studies using the Pedj drought index (PDI), which considers the effects of precipitation and temperature changes (Lashkari et al. 2021).

The vegetation health index (VHI) (Kogan 2001) is a vegetation index that comprehensively monitors the effects of water stress and temperature stress on vegetation and is a widely used remote sensing-based vegetation index (Won et al. 2021, 2022a; Won & Kim 2023; Zeng et al. 2023). In this study, VHI was applied to quantitatively express the vegetation health of the Korean Peninsula. The VHI is based on two assumptions; (1) the VHI defines poor vegetation health as lower normalized difference vegetation index (NDVI) and higher land surface temperature (LST); and (2) in the absence of prior knowledge of the relative contribution of NDVI and LST to vegetation health, the contribution of the two factors is assumed to be equal. The first assumption of the VHI has been confirmed in several studies, and the contribution of NDVI and LST to vegetation health has been found to vary with location, climatic environment, and vegetation type (Bento et al. 2018; Won et al. 2020c).

The VHI is calculated as the average of the two factors when there is no prior knowledge of the vegetation condition index (VCI) and temperature condition index (TCI) contributions (Ejaz et al. 2023). However, the contribution of VCI and TCI varies by region and period. Additionally, when applying a meteorological drought index to relate a relationship with the VHI, the relationship with the VHI varies depending on not only the meteorological drought index but also the time-scale applied to the calculation of the meteorological drought index. There may also be a time-lag effect between meteorological drought indices and VHI. Recently, many studies have investigated the relationship between meteorological drought indices and VHI (Zeng et al. 2022, 2023; Weng et al. 2023). In line with this research trend, it is necessary to examine the relationship between vegetation and meteorological drought on the Korean Peninsula using various meteorological drought indices and VHIs with various weights applied. In exploring the effects of meteorological drought on vegetation activity on the Korean Peninsula, the objectives of this study can be summarized as seeking answers to two questions: (1) Which meteorological drought indices are most closely related to vegetation activity, and what are their corresponding time-scales and time-lags? (2) What is the relative contribution of NDVI and LST in quantifying the vegetation stress experienced by meteorological drought? For this purpose, a correlation analysis is performed between different meteorological drought indices calculated with different lags and time-scales and VHI calculated with different weights. In other words, we want to determine the meteorological drought index (including its corresponding time-scale and time-lag) that shows the strongest relationship between the meteorological drought index and the VHI, and the weights of the VCI and TCI that comprise the VHI.

The study area was selected from 39.975°N to 33.025°N, 124.475°E to 131.125°E to cover the Korean Peninsula (see Figure 1). The applied data period is from January 2001 to December 2021.

Conditional merging (CM) is a method to estimate the value of weather data in unmeasured areas by using the spatial characteristics of satellite data (Vishnu et al. 2022). It is a technique that synthesizes ground observation data assumed to be true and satellite data containing spatial information. In other words, the error caused by applying the spatial distribution of ground observation climate data can be estimated from satellite data containing spatial information, so the error can be corrected by using the spatial characteristics of satellite data.

Figure S2 in the Supplementary Material shows the results of the bias correction using the CM method on a monthly basis, which shows the spatially averaged values of monthly precipitation and monthly PET for the study area. These findings indicate that the accuracy of satellite-derived meteorological data is limited and that bias correction is necessary to improve the utility of satellite meteorological data.

In this study, four meteorological drought indices (SPI, EDDI, SPEI, and CJDI) were applied to analyze meteorological drought from various aspects. SPI, which is widely used as an index for meteorological drought monitoring and analysis, is calculated from observed precipitation data over a user-defined cumulative period (McKee et al. 1993). In other words, SPI is a drought index developed from the perspective that drought is caused by insufficient moisture supply from the atmosphere, and it is calculated using moving average monthly precipitation over various time-scales. The EDDI is a drought index developed from the perspective that drought is caused by an excessive demand for moisture from the atmosphere, and it uses PET to identify drought (Hobbins et al. 2016). EDDI is calculated using moving average PET time series over different time-scales, but a positive value indicates extreme drought; therefore, EDDI is applied with a negative sign in this study to facilitate comparison with other drought indices. SPEI is a drought index that can assess the amount of available water resources by improving SPI, which is calculated using precipitation data only (Vicente-Serrano et al. 2010). In other words, SPEI is an index that monitors drought by the difference between the moisture supply side and the moisture demand side of the atmosphere, which is calculated using a moving average time series of precipitation-PET time series. CJDI is a drought index that combines SPI, which considers the moisture supply side of the atmosphere, and EDDI, which considers the moisture demand side and is calculated using a copula (Won et al. 2020a). In this study, SPI and EDDI time series corresponding to the optimal time-scale of SPI and EDDI selected for each pixel were used to calculate CJDI. Negative values of the drought index indicate more severe drought conditions. The drought classification for the drought index is shown in Table 1 (Ionita et al. 2016). More detailed explanations of the meteorological drought index are provided in Text S1 in the Supplementary Material.

In the above expression, is the observed NDVI at each pixel, and and are the minimum and maximum observed NDVI at each pixel in the corresponding month over the entire period of data (2001–2021).

In the above expression, is the LST observed at each pixel, and and are the minimum and maximum values of the LST observed at each pixel in the month over the entire period of data.

In the above equation, is the weight that determines the contribution of VCI and TCI depending on the environment of the study area, and it is customarily set to 0.5 (Kogan 1997). Instead, in this study, 21 weights were set in 0.05 intervals from 0 to 1 (⁠⁠: 0, 0.05, 0.10, …, 0.90, 0.95, 1) to explore the differences in the environment of the study area, and the correlation with the meteorological drought index was examined. The VHI calculated by applying the that shows the greatest correlation with the meteorological drought index is defined as VHIopt. The lower the VHI, the more unhealthy the vegetation. The classification of drought severity using VHI is shown in Table 2.

Several studies used the Pearson correlation coefficient (PPC) (Pearson 1895) to analyze the correlation between meteorological drought indices and vegetation indices (Zeng et al. 2022; Shi et al. 2023; Weng et al. 2023). PPC, which measures the linear relationship between two variables, assumes that both variables are normally distributed. However, it is not clear whether VHI follows a normal distribution. Therefore, we used the Spearman rank correlation coefficient (SRCC) (Spearman 1961), which does not assume a normal distribution. The statistical significance of the correlations was determined using the -value, which is shown in Figure S3 in the Supplementary Material. In this study, the correlation analysis between SPI (or EDDI or SPEI) for time-scales from 1 to 12 months and time-lags from 0 to 4 months and VHI (with 21 weights) was performed for each month using SRCC. Correlations between CJDI, calculated as a time series corresponding to the best time-scale of SPI and EDDI selected for each pixel, and VHI were also performed, i.e. 46,620 correlations were calculated for one pixel (12 time-scales × 5 lags × 12 months × 3 meteorological drought indices × 21 VHI weights + 1 time-scale × 5 lags × 12 months × CJDI × 21 VHI weights = 46,620). For illustrative purposes, a description of the results of the analysis using the combination of EDDI and VHI (weight 0.15) is included in the manuscript (Figures 2–4) and in the Supplementary Material (Figure S4). This does not mean that the EDDI-VHI (0.15) combination is the most correlated. In fact, the most correlated combination varied widely from month to month and pixel to pixel.

Figure S4 in the Supplementary Material is a spatial distribution of the correlation between EDDI and VHI (0.15). The results are shown with a time-lag that represents the maximum correlation coefficient value per pixel. The correlation pattern between EDDI and VHI (0.15) shows spatial and seasonal variations, with higher correlations at relatively shorter time-scales. At shorter time-scales, positive correlations were found in most regions regardless of season, with the exception of some southeastern regions. In the North, on the other hand, the positive correlation is consistent across different time-scales, mostly in all seasons. More detailed explanations are provided in Text S2 in the Supplementary Material.

The time-lag between meteorological drought indices and vegetation indices can reveal the time-lag of vegetation response to meteorological drought. However, the optimal time-lag between meteorological drought indices and vegetation indices is not the same in space and time (Zhong et al. 2021). In this study, we examined the frequency distribution of correlations between meteorological drought indices and vegetation indices as a function of time-lag. The time-lag was identified as corresponding to the strongest relationship between the meteorological drought index and the vegetation index at each pixel.

Propagation time refers to the length of time from the onset of meteorological drought to the onset of vegetation drought, which can be simplified as a mathematical link between meteorological drought indices at different time-scales and vegetation indices (Xu et al. 2021). Meteorological drought shows different correlations with VHI depending on the applied time-scale, even when the same meteorological drought index is used, which means that applying meteorological drought indices of different time-scales rather than a single time-scale can better represent the actual environmental conditions.

The relative importance of VCI and TCI to VHI was assessed based on the correlation between meteorological drought indices (SPI, EDDI, SPEI, CJDI) and VHI with 21 weighting parameters. The relative contribution of VCI and TCI was identified as corresponding to the strongest relationship between the meteorological drought indices and VHI per pixel. A higher weighting indicates a greater contribution of VCI to VHI, while a lower weighting indicates a greater contribution of TCI to VHI.

Figure S10 in the Supplementary Material shows the spatial average correlation of EDDI-VHI for different s. The weight of the highest spatiotemporal average correlation for each period is designated as the optimal ⁠, and the VHI calculated by applying the optimal is defined as VHIopt. In other words, the optimal VHI (i.e., VHIopt) of EDDI-VHI is as follows: (1) January–September: VHI (0.15); (2) October: VHI (0.55); (3) November–December: VHI (0.7).

The final combination of indices for vegetation drought monitoring was evaluated on a spatially averaged basis based on correlation. The spatially averaged correlations between the meteorological drought indices (SPI, EDDI, SPEI, CJDI) for different time-scales and time-lags and the VHI for the 21 weights were highly variable. Monthly spatial mean correlations between meteorological drought indices and VHI were analyzed to determine the optimal combination of indices for vegetation drought monitoring.

The selection of the optimal meteorological drought index and vegetation index is important when monitoring vegetation drought or assessing its severity, which is essential to improve the applicability of VHI in vegetation drought detection. Bento et al. (2018, 2020) and Zeng et al. (2022) are similar to our study in terms of using the strongest correlation between vegetation indices and meteorological drought indices, and they identified the optimal weighting as corresponding to the strongest relationship between SPEI or self-calibrated Palmer Drought Severity Index (sc-PDSI) and VHI. However, there is a need for a comparative analysis of correlations between different meteorological drought indices with different time-scales and time-lags. In this study, the monthly MDIopt and VHIopt were determined by correlation analysis between SPI, EDDI, SPEI, and CJDI with multiple time-scales and different time-lags and VHI with different weighting s, and the performance of the final selected index combination was evaluated.

Our results showed the existence of monthly differences in vegetation response to meteorological drought (see Figure S11 in the Supplementary Material). This suggests that vegetation response to meteorological drought events should be analyzed separately by month (Zhong et al. 2021). A study across Europe also reported a strong tendency for vegetation health to deteriorate with increasing meteorological drought severity, especially in March (Bento et al. 2018). In a study of arid regions around the world, the relative frequency of significantly correlated pixel counts in the Northern Hemisphere was generally higher in the spring and summer months, while the relative frequency of significant correlations in the Southern Hemisphere was generally higher in the summer and autumn months (Bento et al. 2020). A study has also been reported on seasonal differences in how vegetation is affected by different meteorological droughts, each caused by a lack of precipitation and an increase in evaporative demand, centered on the Korean Peninsula (Won et al. 2021). What the aforementioned studies have in common is that they all used a commonly used VHI weight of 0.5. However, we were able to provide appropriate VHI weights on a pixel-by-pixel and month-by-month basis to more accurately examine the health of the Korean Peninsula's vegetation. In addition, we were able to investigate the response of vegetation to various causes of meteorological drought through the application of meteorological drought indices that consider the moisture supply and/or moisture demand aspects of the atmosphere. This study provides information on which meteorological drought index and which vegetation index are more appropriate to use on a monthly basis for vegetation management across the Korean Peninsula.

In general, the strength and extent of the correlation vary with time-lag (Zhong et al. 2021). To examine whether the Korean Peninsula vegetation responds to meteorological droughts immediately or with a time-lag, we examined the monthly variation in the time-lag corresponding to the strongest relationship between MDIopt and VHIopt.

Table S1 in the Supplementary Material shows the relative percentage of MDIopt's time-lag corresponding to the maximum correlation for each pixel. We can see that 0-month has the highest percentage of MDIopt's time-lag, and the second highest is 1-month. More detailed explanations are provided in Text S3 in the Supplementary Material. In summary, except for July–September, the time-lag of meteorological drought on vegetation was mostly 1 month or less, indicating that the Korean Peninsula vegetation generally responds quickly to the effects of meteorological drought. However, it can also be recognized that there are relatively more areas of vegetation that respond slowly to the effects of meteorological drought in July–September.

The strength of the correlation varies with time-scale as well as time-lag (Zhong et al. 2021). We examined the monthly variation in propagation time corresponding to the strongest relationship between MDIopt and VHIopt to determine the changing impact of short- and long-term meteorological variability on Korean Peninsula vegetation.

Figure S12 in the Supplementary Material shows the relative proportion of MDIopt's time-scales corresponding to the maximum correlation per pixel. The optimal time-scale for MDIopt is 1 month with the highest percentage and the second highest is 2 months. More detailed explanations are provided in Text S3 in the Supplementary Material. In summary, except for summer, the propagation time of meteorological drought on vegetation was 3 months or less, indicating that the Korean Peninsula vegetation is more sensitive to short-term meteorological drought than to the effects of long-term accumulated meteorological drought. However, it can be recognized that there are relatively more areas of vegetation affected by long-term accumulated meteorological drought in summer.

Finally, we compared two vegetation index time series (VHIori and VHIopt) and two meteorological drought index time series (SPI6 and MDIopt) for a specific time period as an example. VHIori was selected for comparison because it is applied without prior knowledge of the VCI and TCI contributions of vegetation health, and SPI6 (SPI calculated with a time-scale of 6 months) was selected for comparison because it is currently used by the Korea Meteorological Administration for drought forecasting and warning purposes.

Figure S13(a) in the Supplementary Material shows the spatial distribution of SPI6 and VHIori from February to May 2004, and Figure S13(b) shows the spatial distribution of MDIopt and VHIopt from February to May 2004. Since Figure S13 is for February–May, a weighting = 0.1 was applied for VHIopt and a weighting = 0.5 was applied for VHIori. SPI6 and VHIori do not show any relationship between meteorological drought and subsequent decline in vegetation health. While SPI6 shows meteorological drought occurring in March in some areas and dissipating in May, vegetation health as represented by VHIori was already deteriorating in February, and we can see that vegetation stress increased across the country in April when meteorological drought entered the dissipation phase. On the other hand, MDIopt shows a spatial pattern that is relatively consistent with VHIopt. MDIopt and VHIopt provide a good representation of the impact of the meteorological drought of February to April 2004 on vegetation health. In particular, it can be seen that in May 2004, when the meteorological drought had entered the dissipation phase, the remaining meteorological drought on the east coast was still affecting vegetation health in the same region.

Several studies have investigated the relationship between meteorological drought indices and vegetation indices using the strongest correlation (Zhong et al. 2021; Shi et al. 2023; Weng et al. 2023). Figure S13 shows that selecting the meteorological drought index and vegetation index using the strongest correlation between meteorological drought and VHI better represents the vegetation response to meteorological drought. These examples provide evidence that the results of this study can provide useful information for vegetation drought monitoring. Investigating vegetation stress caused by various meteorological drought conditions can be used to establish measures to prepare for climate change impacts on vegetation health and can also help in sustainable forest management. This research can be linked to the following goals among the sustainable development goals proposed by the United Nations (https://sdgs.un.org/goals): (1) Goal 13: Take urgent action to combat climate change and its impacts, and (2) Goal 15: Protect, restore, and promote sustainable use of terrestrial ecosystems, sustainably manage forests, combat desertification, and halt and reverse land degradation and halt biodiversity loss.

To assess vegetation drought, it is important to first understand the relationship between meteorological conditions and vegetation status and to accurately measure the response of vegetation activities to meteorological drought. In this study, we applied SPI, EDDI, SPEI, and CJDI to quantitatively express meteorological drought in various aspects, and introduced VHI, a widely used vegetation index, to monitor vegetation health. As a first step to find out how to best monitor the condition of vegetation affected by meteorological drought on the Korean Peninsula, we wanted to investigate which meteorological drought index to use and how to weigh NDVI and LST when calculating VHI.

In this study, we analyzed the meteorological drought index on a monthly basis to monitor the response of vegetation to meteorological drought. Through this study, we investigated how the effects of atmospheric moisture supply and moisture demand contribute to the deterioration of vegetation health. These results may provide useful information for formulating effective mitigation measures to counteract the deterioration of vegetation health on the Korean Peninsula due to meteorological drought. We investigated how water stress and temperature stress contribute to vegetation health deterioration on a monthly basis through the introduction of various weighted VHIs. In this study, we improved the effectiveness of vegetation drought detection using VHI, thus improving the reliability and applicability of VHI.

The results of this study suggest that when monitoring the condition of vegetation affected by meteorological drought, it is better to selectively use meteorological drought indices with seasonal characteristics rather than using a single meteorological drought index, and if possible, vegetation indices should also consider seasonal characteristics. In this study, we examined the optimal meteorological drought index and the optimal vegetation index for each season as a first step toward finding out how to best monitor the condition of vegetation affected by meteorological drought on the Korean Peninsula. While the selection of a seasonal optimal index is important, the selection of a spatial optimal index will also be an important decision for future vegetation drought monitoring, as shown in part by the results of this study. Topography, land use, and hydro-climatic characteristics will also play an important role in the selection of meteorological drought indices and vegetation indices. This is left for future research.

This work was supported by Korea Environment Industry & Technology Institute (KEITI) through Aquatic Ecosystem Conservation Research Program (or Project), funded by Korea Ministry of Environment (MOE) (2022003050007).

H. J. and S. K. conceptualized the whole article; H. J. and S. K. developed the methodology; H. J. and J. W. developed the software; H. J. and S. K. validated the data; H. J. and S. K. rendered support in formal analysis; S. K. and S. K. investigated the data; S. K. brought the resources; J. W. and S. K. rendered support in data curation; H. J. and S. K. prepared the original draft; J. W. and S. K. wrote the review and edited the article; J. W. visualized the data; S. K. supervised the article; S. K. administered the project; S. K. rendered support in funding acquisition.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.