3.1. Evaluating the ERA5-Land reanalysis with GTS observation
The 34-year means of annual and seasonal temperature and precipitation averaged over the 27 stations along with their standard deviation were visually compared between the ERA5-Land reanalysis and GTS observations (Fig. 2). Based on the GTS data, the annual average temperature was approximately 8.5 ℃, with the seasonal averages of -5.7 ℃ in winter (December through February, DJF), 8.2 ℃ in spring (March through May, MAM), 21.3 ℃ in summer (JJA), and 10.4 ℃ in autumn (September through November, SON) (Fig. 2a). Compared to the GTS data, the means from the ERA5-Land were lower by approximately 1°C for all the seasonal and annual averages. Additionally, the ranges of one standard deviation from the annual and seasonal means were similar between the two datasets. For precipitation, the annual mean from the GTS data was approximately 2.6 mm/day, with averages of 0.5 mm/day in DJF, 1.5 mm/day in MAM, 6.0 mm/day in JJA, and 2.0 mm/day in SON (Fig. 2b). The annual and seasonal means from the ERA5-Land were higher than those from the GTS by 0.4–0.8 mm/day. The ranges of one standard deviation were generally consistent between the two datasets, except for the autumn season, which had a higher deviation in the GTS than ERA5-Land. The results from the visual comparisons based on means and standard deviations supported the fact that the temperature and precipitation values from the ERA5-Land reanalysis were consistent with those from the GTS observations.
As a quantitative assessment considering the spatial aspect of variance, parametric and non-parametric correlations were estimated along with the scatter plots between the ERA5-Land and GTS data (Fig. 3 and Supplementary Fig. S1). The annual, DJF, and JJA temperatures from the ERA5-Land were linearly associated with those from the GTS over the 27 stations, although with the underestimation of the GTS (lower temperature in the ERA5-Land) (Fig. 3a, 3b, and 3c). The significantly positive associations between the ERA5-Land and the GTS temperatures were quantified as the correlation coefficients with higher than 0.95, 0.9, and 0.75 in Pearson’s, Spearman’s, and Kendall’s correlations, respectively, during all the annual and seasonal periods (all p <0.01) (Table 1a). For precipitation, statistically significant positive correlations between the ERA5-Land and the GTS were shown for all the annual and seasonal means at the 1% significance level, except for the Pearson’s r-value of MAM mean with a significant correlation at the 5% level (Table 1b). The scatter plots of precipitation showed a linear association between the ERA5-Land and the GTS with the overestimation of ERA5-Land precipitation to the GTS (more precipitation in the ERA5-Land) (Fig. 3d, 3e, and 3f). The strength of the linear relationships between the two datasets was relatively weaker during the dry season (DJF) than that in the wet season (JJA). Particularly, the highest correlation coefficients and nearly one-to-one relations between the two datasets were shown for both temperature and precipitation during the summer when vegetation and climate interactions were examined in this study.
As a quantitative assessment considering a temporal aspect of variance, the three correlation analysis methods (Pearson, Spearman, and Kendall) were conducted at each of the 27 stations using the 1982–2015 time-series of annual and seasonal temperature and precipitation. The results are shown in the maps of the Pearson’s correlation coefficients at the 27 stations (Fig. 4 and Supplementary Fig. S2) and further summarized as in the histogram (Fig. 5). The results of Spearman’s and Kendall’s correlations are shown in the supplementary materials (Supplementary Fig. S3 and S4). For temperature, there were significant positive correlations at the 1% level for the annual, DJF, and JJA means in all 27 stations (Fig. 4a, 4b, and 4c). As shown in the histogram, more than 85% of the 27 stations (more than 23 stations) had high correlation coefficients for annual and seasonal mean temperatures with Pearson’s r-values higher than 0.8, 0.8 of Spearman’s ρ, and 0.6 of Kendall’s τ (Fig. 5a). The results indicated that there was a strong association between the ERA5-Land reanalysis and GTS observation for all seasonal and annual means of temperature in nearly all of North Korea. The patterns of Pearson correlation for precipitation also showed significant positive associations at all locations (Fig. 4d, 4e and 4f), except for some stations near the northeastern coastal (i.e., Kimchaek and Sonbong) and the inland mountainous regions (i.e., Yangdok and Pungsan) for annual and JJA precipitation (Fig. 4d and 4f). At approximately 90% of the 27 stations (more than 24 stations), positive associations appeared with moderate correlation coefficients of higher than 0.5 (r), 0.5 (ρ), and 0.4 (τ) in DJF, MAM, and SON, respectively, (Fig. 5b). Compared to the drier seasons, there were relatively lower correlation coefficients (lower than 0.4) in JJA and annual means.
The means and standard deviations of temperature and precipitation between the ERA5-Land reanalysis and GTS observations were comparable (see Fig. 2), and statistically significant positive associations between the two datasets were generally quantified based on both the spatial and temporal correlations (see Table 1 and Fig. 3, 4, and 5). Based on the results of the comparison analysis, the temperature and precipitation from the ERA5-Land reanalysis were consistent with those from the GTS observations, and thus supported ERA5-Land as a suitable climate reanalysis dataset, which can substitute for incomplete observation data in North Korea.
3.2. Spatial distributions and temporal trends of vegetation
We examined the spatial patterns of annually and seasonally averaged NDVI in North Korea during the study period of 1982–2015. As shown in Fig. 6a, the annual mean values of NDVI ranged from 0.3 to 0.8 in North Korea. The higher values (0.6–0.8) were distributed mostly in the inland and northern regions, while the lower values (0.3–0.4) were in the western and near the eastern coastal regions. In the western regions, which includes the capital city of North Korea (Pyongyang), the lower NDVI values were extensively distributed even during the summer growing season, as compared to other regions (Fig. 6b). The seasonal NDVI means of DJF, MAM, and SON are shown in the supplementary materials (Supplementary Fig. S5). To detect the spatio-temporal changes in vegetation cover, the annual and seasonal NDVI trends along with their percentage changes were calculated based on linear regression analysis. Statistically significant NDVI changes primarily occurred in the western region of North Korea. In the western regions, the annual mean NDVI significantly decreased, with a trend of approximately -0.03 to -0.01 per decade (Fig. 7a), which correspond to 10–20% reduction in NDVI over the 34 years of the study period (Fig. 7b). Additionally, the seasonal means of NDVI in the western region consistently showed decreasing trends for the growing seasons throughout the spring (Supplementary Fig. S6), summer (Fig. 7c), and autumn (Supplementary Fig. S6). During JJA, the highest rate of decreasing NDVI, approximately -0.03 per decade occurred in the western region (Fig. 7c), which indicated 10–25% reduction in summer NDVI for the 34 years (Fig. 7d). The results indicated that vegetation in the western region of North Korea was intensively reduced over the thirty-four years of 1982–2015.
3.3. Associations of vegetation with near surface climates
Based on the results presented in the previous section, significantly changed vegetation was detected in the western region of North Korea with a large decrease in NDVI, and sporadically distributed, small, isolated areas of increased or decreased NDVI were detected in the northern and eastern regions. To identify how the NDVI changes can affect the near-surface climate during summer, we conducted a detrended correlation analysis of JJA NDVI with the surface energy variables of SSRD, SSR, H, LH, Bowen ratio, and 2m temperature in JJA at each grid point.
The results showed a positive correlation between NDVI and SSRD in almost all of North Korea, especially higher correlations in the northern and southern mountainous regions with an r-value higher than 0.3, which indicated a significant positive correlation at the 10% level (Fig. 8a). Contrastingly, relatively weaker positive and even negative correlations were observed in the western region (e.g., Nampo and Pyongyang, see Fig. 1 for the locations) and the eastern coastal region (e.g., Hamhung, see Fig. 1 for the location). The detrended correlation pattern of SSRD with NDVI was generally consistent with that of SSR with NDVI (Fig. 8b). The NDVI in the northwestern and southeastern regions, which showed higher r-values with both SSRD and SSR, were positively and significantly correlated with H. However, the NDVI was negatively correlated with H in the western region around Nampo (r = -0.38) (Fig. 8c). The spatial pattern of detrended correlation between LH and NDVI showed a positive association in almost all the grid points, and the statistically significant correlations were in the western and northern regions with an r-value higher than 0.3, which indicated that evapotranspiration in JJA was more active as the amount of green vegetation increased (Fig. 8d). The pattern of detrended correlation between Bowen ratio and NDVI indicated the pairs of two regions with an opposite sign of r-value (Fig. 8e), which was attributed to the results of H and LH (Fig. 8c and 8d). In the northwestern and southeastern regions, where a positive correlation for each H and LH was observed, a positive correlation of Bowen ratio with NDVI was shown. Contrastingly, a negative correlation for the Bowen ratio in the southwestern and northeastern regions, where a negative correlation for H but a positive correlation for LH, was identified. The spatial pattern of the detrended correlation analysis for temperature was generally consistent with that for the Bowen ratio. Negative correlations of temperature with NDVI were found in the western and northeastern coastal regions, especially in the western region including the significant values at the 10% level, and ranged from -0.4 to -0.3 (Fig. 8f). Contrarily, the positive associations were in the northern region, including the significant areas at the 10% level with r-values of 0.3 to 0.4, and also in inland and southeastern regions.
To estimate the change in 2m temperature over NDVI, the slope values of the linear regression between JJA NDVI and 2m temperature at each grid point were estimated (Fig. 9). The spatial pattern of linear regression was generally consistent with that of detrended correlation (see Fig. 8f), which showed negative slope values (decreasing temperature with increasing NDVI) in the western region and positive values (increasing temperature with increasing NDVI) in the northern region (Fig. 9). The significant areas identified by t-test for the slope at the 10% level are shown within the western region of 38.0–40.0 °N and 125.1–126.5 °E and the northern region of 40.5–41.2 °N and 125.9–127.9 °E, which were consistent with the significant regions tested by F-test for the linear regression models (not shown here). Specifically, as the NDVI increased by 0.1, the temperature decreased by approximately 0.4–0.6 ℃ in the western region, while it increased by approximately 0.4–0.6 ℃ in the northern region. Using the area-averaged time-series only over the grid points with significant associations between JJA NDVI and 2m temperature in the western and northern regions, Pearson’s correlation and linear regression analysis were performed, and the results are shown in Fig. 10. In the northern region, JJA temperature increased and JJA NDVI showed almost no change during 1982–2015, but for both, there was an increasing trend before the mid-1990s and no significant trend thereafter (Fig. 10a). The significant positive associations between temperature and NDVI were identified by a positive correlation coefficient (r = 0.35) and a positive slope value (0.68 ℃ per 0.1 NDVI) with significance at the 5% level (Fig. 10b). Contrastingly, in the western region, there were opposite trends between JJA NDVI and 2m temperature with a decreasing trend for NDVI and an increasing trend for 2m temperature, especially after the mid-1990s (Fig. 10c). A significant negative correlation was shown between NDVI and 2m temperature with an r-value of -0.4, and the linear regression results indicated that if NDVI increased by 0.1, then the temperature decreased by approximately 0.77 ℃ (Fig. 10d).