3.1. Correlations of Drought Indices for Different Timesteps
The DI values were first standardized with integers based on the drought classifications (
Table 3) for PN, Z-Score and SPI for 18 timesteps and EDI. Then Pearson correlation analysis was employed to detect the relationship among the calculated indices.
The average correlation coefficients among the DIs in different timesteps were calculated for the basin (
Figure 2). EDI was timestep independent, so the correlation coefficients between EDI and other DIs in different timesteps were listed in
Table 4. Generally, DIs calculated using intermediate timesteps varied from 7 to 13 months correlated better (>0.5) with others. While, the correlations of both shorter (1 to 6 months) and longer (18–24 months) timesteps, especially one-month, two-month, and 48-month, with others being relatively low. Additionally, DIs in intermediate timesteps showed better correlation with both shorter and longer timesteps, while DIs in shorter or longer timesteps only showed better correlation with nearby timesteps. This indicates DIs using intermediate timesteps are the best suited to evaluate drought in WRB.
When further calculating the average coefficient for each DI with all DIs in a certain timestep, DIs were found similar timestep-average correlations for each timestep. Specifically, PN showed the least correlation with other DIs with almost all timesteps. Z-Score with shorter timesteps (two-month to six-month) had a better correlation with DIs, while SPI with intermediate and longer timesteps (seven-month to 48-month) correlated better with other DIs. While, EDI is exhibiting better relationship with others in most timesteps. More than 70% of the coefficients of EDI with other DIs larger than 0.50, and about 30% of which were larger than 0.70, which making EDI to be the best correlated DI.
Therefore, Z-Score4, PN10, and SPI24 were selected as the shorter, intermediate, and longer timestep DIs, respectively, and a comparison was made between these DIs with EDI to investigate the proper timesteps of the indices.
Average values of correlation coefficients of each DI (Z-Score4, PN10, SPI24, and EDI) against other DIs were calculated for each timestep (
Figure 3). For example, the blue bar in the first timestep shows the average of three coefficient values (i.e., the coefficient between EDI and PN1, the coefficient between EDI and Z-Score1, and the coefficient between EDI and SPI1), and the first green bar represents the average of three coefficient values (i.e., the coefficient between Z-Score4 and Z-Score1, the coefficient between Z-Score4 and PN1, and the coefficient between Z-Score4 and SPI1), and so on.
As illustrated in
Figure 3, EDI showed higher correlation with shorter timesteps than other DIs, except shorter timesteps DI (i.e., Z-Score4), which had significantly higher correlations than other indices only in timesteps shorter than five months. For intermediate timesteps DI (i.e., PN10), the correlation was higher than EDI only for 10-, 11-, 36-, and 48-month timesteps, and even for these timesteps the correlations were very similar and close to EDI. As for longer timestep DIs (i.e., SPI24), it performed well under the longer timesteps (longer than 18 months) and possessed higher correlations than EDI. Therefore, EDI indicated a better performance than other indices for different timesteps, in general. The best correlation of EDI was observed with the timesteps between 8 to 12 months, with the average coefficients of 0.71–0.76.
3.3. Effectiveness of DIs in Detecting Historical Droughts
It is known that there were four significant droughts in Keshan, i.e., from autumn 1999 to spring 2000, autumn 2000 to spring 2001, autumn 2002 to spring 2003, and autumn 2004 to spring 2005 [
39]. Therefore, we enrolled the period from September 1999 to May 2001 as the representative drought period in Keshan, and the DIs were calculated based on the data of Keshan station during this period. Based on the prior analysis among DIs, only EDI, and Z-Score and SPI in intermediate timesteps (9, 12, and 15 months) were selected to identify how DIs coordinated with the historical deficiency conditions of monthly and multi-monthly precipitation.
Actual monthly precipitation (precipitation received in a particular month), average monthly precipitation (average value for a particular month during the studying period), multi-monthly average precipitation (average value of precipitation in particular preceding months), and moving cumulative precipitation was employed to evaluate a variety of precipitation.
When comparing the indices and precipitation variables in 9-month timestep, recurring deficits between monthly precipitation and average monthly precipitation were observed (
Figure 5). The largest deficit happened from September to November 1999 and in June 2000, with monthly precipitation not even reaching a half amount of the long-term average value for each month. The deficit of the moving cumulative precipitation compared with multi-monthly average precipitation declined gradually from September 1999 to March 2000, implying a mitigation of drought conditions, but then increased from April to June in 2000, as drought conditions took hold again. However, the deficit stabilized from July 2000 to May 2001, and then decreased from March to May 2001, after which time drought conditions ceased.
Compared with EDI, temporal changes of SPI9 and Z-Score9 generally performed better in tracking the deficits shown with moving cumulative precipitation and multi-monthly average precipitation. There were two critical differences between EDI and the other two indices. The first one was observed from February to June 2000, when SPI9 and Z-Score9 showed an increase-decrease-increase change while EDI was initially stable, but then declined. This shows that EDI did not capture the deficit signaled by the moving cumulative precipitation and multi-monthly average precipitation from February to March 2000, and also responded inversely to the decline in deficit in June 2000. The other difference was found during February to March 2001, when SPI9 and Z-Score9 showed an increase in deficit conditions but EDI was decreasing, which also implicated an inverse response to the deficit decline during this period.
The scatter plots of drought indices and precipitation deficit in the nine-month timestep were shown in
Figure 6. DIs ((i.e., Z-Score9, SPI9, and EDI)) did not reflect the deficit between monthly precipitation and average monthly precipitation very well, and r
2 was only 0.037, 0.10, and 0.12 respectively. However, DIs had a statistically close relationship with the deficit between moving cumulative precipitation and multi-monthly average precipitation for the nine-month timestep with 0.997, 0.55, and 0.32 values of r
2.
For the 12-month timestep, multi-monthly average precipitation was relatively high in every month compared to the moving cumulative precipitation at Keshan station (
Figure 7). The gap between these two variables was relatively stable from September 1999 to June 2000, and then became larger in July and August, which led to more severe drought conditions. From September 2000, it began to decrease again and reached a stable level until May 2001.
Changes of EDI, SPI12, and Z-Score12 were consistent with the water deficits indicated by moving cumulative and multi-monthly average precipitation totals. To be specific, SPI12 and Z-Score12 exhibited a similar pattern, with changing lines almost coinciding with each other. EDI deviated from these two indices in two ways. From May to June 2000, EDI decreased but the other two remained mostly stable. In the meantime, the gap between monthly precipitation and its long-term average increased from 14 mm to 56 mm, while the gap between cumulative precipitation and multi-monthly average precipitation remained mostly unchanged. EDI also deviated from the other indices in June and July of 2000. EDI increased but the other two decreased. Monthly precipitation deficits were improving and reached 87% of its long-term average, but the deficit in the moving cumulative precipitation and multi-monthly average precipitation was increasing from 62 mm to 122 mm.
Based on the correlation analysis shown
Figure 8, it was further found that Z-Score12 and SPI12 did have a weak relationship with monthly precipitation variability (r
2 = 0.005). While they showed a significant correlation with deficit between moving cumulative precipitation and multi-monthly average precipitation, and the r
2 was 0.998 and 0.992, respectively. Correlation between EDI and water deficit indicated by moving cumulative and multi-monthly average precipitation in 12-month timestep was also better than its relationship with monthly precipitation change, but the r
2 was 0.47, which is not as strong as the other two indices were.
As shown in
Figure 9, moving cumulative precipitation was larger than multi-monthly average precipitation in 15-month timestep during September and October in 1999, and then went shorter and below the multi-monthly average precipitation. The deficit was 31 mm to 80 mm from November 1999 to May 2000 and became larger from June 2000 until getting a peak value of 186 mm in November 2000. Subsequently, it declined again to 125 mm to 142 mm during December 2000 to May 2001.
Z-Score15 and SPI15 tracked cumulative precipitation changes, and generally experienced decrease-increase-decrease-increase fluctuations during the entire drought period. The indices detected a normal-wet condition in September 1999, and then decreased indicating a mild drought during November 1999 to May 2000. From June 2000, their values dropped below −1.0 indicating a moderate drought. However, EDI performed differently at times. The index detected a mild drought during September 1999 to May 2000 and then decreased further, indicating a moderate drought except in July, October to December in 2000, and in May in 2001, when EDI only performed mild drought conditions.
Scatter plots were also produced for DIs with respectively monthly and multi-monthly precipitation deficits in 15-month timestep
Figure 10. Similar with the other timesteps, Z-Score15 and SPI15 showed a weak relationship with deficit between monthly precipitation and average monthly precipitation, the r
2 was 0.0084 and 0.0089, respectively. While they correlated quite well with the deficit between moving cumulative precipitation and multi-monthly average precipitation in 15-month timestep, with r
2 of both of them higher than 0.95. The r
2 of EDI and water deficit indicated by moving cumulative and multi-monthly average precipitation in 15-month timestep was higher than that in the nine-month timestep but lower than the 12-month timestep.
Based on the above results using different timesteps (nine-month, 12-month, and 15-month), SPI, Z-Score, and EDI picked up the precipitation deficit and can be used to demonstrate actual drought conditions. SPI and Z-Scores were more sensitive to the changes of the deficit between moving cumulative precipitation and multi-monthly average precipitation, while EDI tracked the variation in monthly precipitation deficit more closely. Compared with other timesteps, SPI and Z-Scores best responded to deficits in cumulative precipitation and more closely tracked EDI when a 12-month timestep was used.
In terms of drought severity, SPI indicated more severe drought conditions than the Z-Score and EDI from September 1999 to May 2001, except for the 12-month time step, which was almost the same as Z-Score12. The frequency of drought severity conditions detected by DIs in Keshan is demonstrated in
Table 5. The extreme drought and severe drought detected by SPI9 occurred in May and June 2000, respectively. The two severe droughts detected by Z-Score12 and SPI12 occurred in August 2000, while the severe drought was detected in November 2000 using SPI15. The cumulative precipitation respectively achieved 49.19%, 53.18%, 60.81%, and 67.55% of its corresponding multi-monthly average in each particular month. Overall, Z-Score12 and SPI12 was more consistent with drought severity according to actual conditions.
Several researchers had addressed droughts conditions in northeast China [
40,
41]. Song et al. [
42] indicated that precipitation is a key factor affecting drought in this area, and proposed an index particularly suitable for detecting, monitoring, and exploring spring droughts in the Songnen Plain. Therefore, we selected several precipitation-based indices to evaluate the drought condition in the WRB and assessed their performances of DIs in drought monitoring. Our results proved that it was necessary to investigate various indices to develop a reliable drought monitoring system. This was consistent with previous studies [
17,
43,
44], which indicated that considering more than one DI for drought studies can help investigate how well they cohere with each other and examine the sensitivity and accuracy of DIs. Montaseri and Amirataee [
45] used seven meteorological drought indices to monitor drought characteristics in 12 diverse parts of the world endowed with various climatic conditions and found that the application of SPI can result in higher relative advantages to undertake a comprehensive and accurate analysis. Our study agrees with this study and also found SPI was the most reliable drought indicator among the DIs. The performance of SPI may vary when compared with different indices. However, this conclusion will be true for a meteorological drought monitoring which is based solely on rainfall, and also where rainfall can define drought conditions by its own [
17,
45,
46]. In this study, performance of DIs in different timesteps varied a great deal, and DIs overall exhibited the best coherence in intermediate timesteps. This result showed that timestep selection is as important as a particular DI in drought assessment. These conclusions were supported by previous studies [
28,
29,
47]. Therefore, we recommended that various indices should be investigated in drought monitoring considering intermediate timesteps to develop a reliable drought monitoring system.
All types of drought originate from the deficiency of precipitation and thus many drought indices incorporate the precipitation component in various forms (e.g., SPI, precipitation percentile, decile), especially for a comprehensive drought assessment [
14,
17,
19]. However, the relationship between precipitation and drought needs to be further investigated as other studies indicated that high temperature can also contribute to the quick onset of drought [
48,
49]. Furthermore, according to the Fifth Assessment Report of the IPCC [
50], most ecosystems will be impacted to a greater extent by the climatic extremes in future because most of the global climate models predicted more extremes in the climates such as multi-year droughts. Droughts monitoring and evaluation for ecosystems are becoming more urgent [
24,
26,
27]. Therefore, future studies will need to focus on ecosystem droughts based on multiple drought indices, including remote-sensing indices.