*4.2. Temporal Trend*

Trend analyses were conducted for the annual index images using the aforementioned OLS regression and F-statistics. On the basis of the calculated slopes and significant levels, the trends of changes in the drought indices were classified into seven categories: (1) highly significant decrease (*SLOPE* < 0, *p* ≤ 0.01), (2) significant decrease (*SLOPE* < 0, 0.01 < *p* ≤ 0.05), (3) moderately significant decrease (*SLOPE* < 0, 0.05 < *p* ≤ 0.1), (4) no significant change (*p* > 0.1), (5) moderately significant increase (*SLOPE* < 0, 0.05 < *p* ≤ 0.1), (6) significant increase (*SLOPE* > 0, 0.01 < *p* ≤ 0.05), and (7) highly significant increase (*SLOPE* > 0, *p* ≤ 0.01) [35], as shown in Figure 4. Figure 4 shows a significant PCI decrease for 8.8% of the study area and a significant increase for 13.6% of the area. The remaining area exhibited no significant change during the last 18 years. The area with a significant increase in the PCI is mainly located near Mudanjiang, Jilin, and Yanji City, while the area with significantly decreasing PCI is mainly distributed near Liaoyang, Anshan, and Dandong City. For TCI, 3.3% of the area experienced a significant increase and 3.3% of the area witnessed a significant decrease. The area with a significant decrease was distributed near the north of Shuangyashan and Qitaihe City and south of Anshan City, suggesting that these areas experienced very severe droughts in the recent 18 years. The positive trend was mainly observed in Mudanjiang and Tonghua City. With regard to the VCI, a significant increase was seen for 5.3% of the study area, whereas a significant decrease was seen for 21.3%, indicating that a larger area tended to experience very severe droughts recently. The areas with a significant decrease in the VCI are mostly located in the southern coastal region, Yanji in Jilin, and the northern part of Shuangyashan. The rest of Shuangyashan and Baishan exhibited an increase in the VCI. VHI trend analysis showed that 4.9% of the GCM area experienced a significant drought alleviation, whereas 14.1% experienced a significant drought aggravation. The spatial pattern of the VHI trend is similar to that of the VCI trend. SDCI trend analysis revealed that about 11% of the central area of the GCM experienced a significant increase (i.e., drought condition is relieving); meanwhile, almost an identical extent in the southwest experienced drought aggravation. TVDI trend analysis showed that a sparse area of 9.0% of the GCM experienced a significant increase, whereas 3.5% of the area experienced a significant decrease. The region with the decreasing TVDI was mostly distributed near Mudanjiang, Liaoyuan, and Tonghua.

Thus, the trend analyses indicate di fferent drought trends for di fferent indices. For the GCM, both PCI and SDCI show similar patterns that the central area is getting wetter and the southwest area is getting drier. VCI and VHI exhibit similar patterns showing sparse areas with decrease (drying) and overwhelming increase (wetting).

**Figure 4.** Slopes of drought indices during 2001–2018. (**a**) PCI, (**b**) TCI, (**c**) VCI, (**d**) VHI, (**e**) SDCI, and (**f**) TVDI.

#### *4.3. Correlations between Drought Indices and Meterological Factors*

The annual average precipitation and temperature were collected from 14 prefecture-level cities in the GCM. Six annual drought indices were plotted for these cities, as shown in Figures 5 and 6. As shown in Figure 5, PCI and SDCI both show similar annual patterns with precipitation. This is because precipitation is an important input for both these indices. The other four indices (VCI, TCI, TVDI, and VHI) do not show any clear relation with annual precipitation. TCI mainly reflects the surface temperature variation, while for the other three indices, NDVI is an important input.

**Figure 5.** Relationship between annual drought indices and precipitation in the GCM. (**<sup>a</sup>**–**<sup>n</sup>**) are the 14 prefecture-level cities. (**a**) Anshan; (**b**) Baishan; (**c**) Benxi; (**d**) Dandong; (**e**) Fushun; (**f**) Jilin; (**g**) Jixi; (**h**) Liaoyang; (**i**) Liaoyuan; (**j**) Mudanjiang; (**k**) Qitaihe; (**l**) Shuangyashan; (**m**) Tonghua; (**n**) Yanji.

**Figure 6.** Relationship between annual drought indices and temperature in the GCM. (**<sup>a</sup>**–**<sup>n</sup>**) are the 14 prefecture-level cities. (**a**) Anshan; (**b**) Baishan; (**c**) Benxi; (**d**) Dandong; (**e**) Fushun; (**f**) Jilin; (**g**) Jixi; (**h**) Liaoyang; (**i**) Liaoyuan; (**j**) Mudanjiang; (**k**) Qitaihe; (**l**) Shuangyashan; (**m**) Tonghua; (**n**) Yanji.

Dandong has a relatively larger precipitation volume with high variations from 2001 to 2018 because of its special geolocation. Its forest coverage is as high as 65%, endowing the region with high water-holding capacity. Precipitation-based drought indices in this area indicate clearly humid characteristics. For example, the precipitation was very high in 2012 and 2013, and the average annual temperature was lower (Figure 6d). Therefore, most of the indices identified the lack of a drought condition in these 2 years. Benxi, which is a city neighboring Dandong, had slightly lower annual average temperature and precipitation. Its drought condition was similar to Dandong according to di fferent indices. Anshan and Liaoyang are located inland and receive less precipitation than Dandong and Benxi. In particular, in 2014, the precipitation of Liaoyang was only 300 mm, and the average temperature was higher than 10 ◦C, which was significantly higher than that in the other years. PCI and SDCI clearly indicate the drought situation in Liaoyang. Anshan received precipitation as low as approximately 400 mm in 2014, and the average annual temperature was higher than 11 ◦C. PCI and SDCI also reflected the drought situation of Anshan correctly. Fushun received an annual precipitation exceeding 1,000 mm in 2010 and 2013, and its average annual temperature was between 5.5 ◦C and 6 ◦C; meanwhile, in 2011 and 2014, it received a very low precipitation of 500 mm, and the PCI and SDCI values clearly reflect these changes. Baishan and Tonghua are located close to the Changbai Mountain Nature Reserve. The annual precipitation and temperature were relatively stable from 2001 to 2018. The drought indices also tended to vary smoothly. Jilin and Liaoyuan experienced similar annual precipitation variations in the recent years. In 2011, their annual precipitation was the lowest, only about 500 mm, which was captured by PCI and SDCI. Yanji, Jixi, Mudanjiang, Qitaihe, and Shuangyashan are located inland north of the Changbai Mountain Nature Reserve and have high latitudes, low annual average temperatures, and annual precipitations less than 800 mm. Their PCI, TCI, VCI, VHI, and SDCI values are lower than those in other regions. In the GCM, the TVDI, VCI, TCI, and VHI values also exhibit distinct annual patterns that do not appear to be related to the annual precipitation and temperature. This may because these indices do not consider precipitation, and indicate droughts based on the LST anomalies and vegetation health. The annual average temperature is usually not sensitive enough to reflect LST anomalies; hence, LST-based indices fail to show consistency with annual average temperature variations. Further, vegetation health may be sometimes a ffected by factors besides drought, and there is usually a time lag before drought can cause deterioration of vegetation health. This may be the reason why vegetation-based indices also fail to show consistency with the precipitation and temperature patterns.

#### *4.4. Correlations between Drought Indices and Land Cover Types*

An examination of the annual land cover data from 2001 to 2018 revealed very little land cover change over these years. Therefore, in this study, we assumed that there was no land cover change, and we used the land cover map of 2018 as the current condition of land cover to investigate the correlations between drought indices and land cover types. According to the 2018 land cover map, in the GCM, deciduous broadleaf forests, grasslands, and savannas accounted for 46.8%, 30.4%, and 18.1%, respectively. The other seven land cover types account for less than 5%.

The distribution of vegetation types has a strong relationship with regional climatic factors, and surface temperature is an important climatic factor that a ffects the zonal distribution of vegetation types. Temperature is an important limiting factor for plant growth in the GCM; forest vegetation types, in particular, have distinct vertical zonal distribution characteristics. NDVI represents vegetation characteristics, and there exists a clear relationship between NDVI and LST. Le Page et al. [36] found that the negative correlation between NDVI and LST in agricultural area is due to drastic evaporation that decreases LST. However, there exists a positive correlation between NDVI and LST in the northeastern part of the study area, and this correlation is explained by the simultaneous forest leaf loss and fall in surface temperature (the coldest months). Figure 7a,b present the mean values of the six drought indices for di fferent land cover types in 2018 and 2001. For the NDVI-based index (i.e., VCI) and the LST-based index (TCI), when the land cover changes from deciduous needleleaf forests to unvegetated

lands, VCI decreases, while TCI increases. These two indices exhibit distinct patterns in di fferent land cover types. The variability in the slope of the inverse LST–NDVI relationship in association with local topographic and environmental conditions has been assessed in previous studies [37]. The validity of VHI as a drought detection tool relies on the assumption that the NDVI and LST at a given pixel vary inversely over time with variations in VCI and TCI driven by local moisture conditions. However, over vast areas and long periods, the LST–NDVI relationship is nonunique and often nonnegative [37–39]. According to [37–39], NDVI and LST are positively related usually in energy-limited ecosystems, which implies that high temperature promotes the growth of vegetation. In this case, VHI and TVDI may not be appropriate for indicating drought conditions. However, for water-limited ecosystems, where high temperature may inhibit vegetation growth, NDVI has a negative relation with LST, conforming with the assumption of TVDI. In this case, TVDI and VHI are applicable for indicating drought conditions. Since the water-holding capacity of deciduous broadleaf forests is stronger than that of grasslands, deciduous broadleaf forests generally are more humid than grasslands. This is reflected by almost all the drought indices. Savannas comprise a mixed forest–grassland type of vegetation and have drought indices intermediate between those of the deciduous forests and grasslands.

**Figure 7.** Drought indices and their changing slopes for different land cover types: (**a**) Drought indices in 2018; (**b**) drought indices in 2001; and (**c**) slopes (×100) of drought indices from 2001 to 2018.

We examined the zonal statistic to the slope value based on land cover types. Figure 7c shows the mean *SLOPE* values (scaled by 100) of drought indices from 2001 to 2018 for di fferent land cover types. The TVDI *SLOPE* values of evergreen needleleaf forests is the highest among all land cover types; the *SLOPE* values of the other five indices are low and negative. This suggests that evergreen needleleaf forests in the GCM clearly have experienced severe drought conditions. Savannas and deciduous needleleaf forests also have similar but weaker changing patterns with positive slope for TVDI but negative slope for the other indices, which is similar to Zribi et al. [40], who analyzed drought affection on vegetation coverage based on time series Vegetation Anomaly Index (VAI). Our analysis of slope and land cover types showed that VCI and VHI have negative values with similar variation patterns across different land cover types. This means that the VCI and VHI decreased from 2001 to 2018, indicating an overall drying trend. The sharpest trend occurs in the case of unvegetated Lands. Since PCI is a precipitation-based index, when the precipitation increases, the slope of PCI is positive, indicating a wetting situation. Broadleaf croplands and deciduous broadleaf forests, which together account for a large proportion of the GCM, have high water-holding capacity. Figure 7c shows that the slope values for these types is positive, indicating a wetting situation.
