2.3. Analysis
The nature of droughts is complex, and any drought index should realistically consider climatological factors such as rainfall, temperature, air humidity, wind, and soil conditions. The selection of a drought index for analysis is dependent on the specific region, available information (data base), and the objective of the analysis [
7]. Hao and Singh [
8] comprehensively reviewed some drought indices, addressing the principles of the methods along with their limitations and strengths. Svoboda et al. [
9] reported that no singular index can portray drought conditions for all of space and time. Droughts are multidimensional in nature, manifested on different temporal scales, and cannot be fully characterized using a single indicator [
10].
The use of SPI [
11] has been popular, primarily due to its less complicated formulation and requiring only rainfall time series, as well as being capable of characterizing both temporal and spatial climatological drought conditions [
12].
Sayari et al. [
13] applied three drought indices: SPI, precipitation index percent of normal (PIPN), and agricultural rainfall index (ARI) using databases from 1990 to 1961 in Northeast Iran. The future drought conditions in the Kashafrood basin, Iran, due to climate change resulting from low and high greenhouse gas emission scenarios (Special Report on Emission Scenarios or SRES B2 and SRES A2, respectively) were predicted using all three indices. All indices indicated higher drought frequency as a result of climate change under both scenarios. The findings support that even the simple SPI can provide equally good results compared with the more detailed indices (PIPN and ARI).
Saada and Abu-Romman [
14] studied the use of contemporaneous autoregressive moving average (CARMA) time series analysis to model the SPI at a time scale of 12 months (SPI-12) in the northwest mountainous region in Jordan. They used a rainfall database recorded from five rainfall stations from 1983 to 2013 (30 years). The results demonstrated that CARMA (1,1) can model the SPI in the region and that the cross-correlation structures between the stations were well preserved.
Setiawan et al. [
15] applied the normalized monthly precipitation and SPI to study the influence of El Niño events using rainfall data based on 1950–2010 in Indonesia. They found that the influence of El Niño events is better represented by use of SPI. The use of temporal and spatial SPI in more regions and seasons affected by El Niño can more accurately reflect drought outlook.
Tshiabukole et al. [
6] applied SPI to analyze the influence of climatic variability to the seasonal rainfall pattern in Southwestern Congo. They found that the frequency of occurrence of dry periods in successive years is relatively low, although 25 years over the last 50 years have experienced droughts.
Recently, Theresia et al. [
5] used SPI to evaluate the correlation between drought area and the location of small reservoir construction in the Bodri-Kuto river basin In Indonesia. They reported the low suitability of the locations of small reservoirs given the drought conditions. However, Theresia et al. [
5] used SPI defined as the standardized deviation from its respective month [
5]. This use of SPI only indicates each month’s deviation, resulting in different dry spell definitions in different months despite having equal SPI values. Secondly, they applied drought severity criteria based on simultaneous drought frequencies of dry and very dry spells in consecutive months. This joint drought severity definition means the probability of drought conditions may not reflect field conditions.
The SPI in this study was calculated both yearly and monthly. For yearly SPI, we used the mean of yearly rainfall data as the reference; for monthly SPI, we used the mean of monthly rainfall data as the reference applicable throughout the months. In principle, the index calculates the standardized rainfall. When the SPI is less than the average, the index is negative. The larger the negative value, the larger the deviation (smaller than) from its average or reference value, indicating a more severe drought index. The SPI formula is as follows [
5,
16,
17]:
where
is rainfall (mm) at time period
I,
is average rainfall (mm), and
is the standard deviation (mm).
Based on the SPI calculated above, the drought condition was classified as shown in
Table 3.
The analysis performed in this study was as follows:
(1) SPI calculation: The SPI was analyzed for 17 years of rainfall records (2000–2016) from 25 rainfall stations in the river basin. Before the SPI analysis, the missing data were filled with inversed square distance values [
18] and the data were checked for consistency using the double mass curve with correction [
19].
(2) The dry months were also crosschecked using Oldeman’s method [
20]. We confirmed that July to September are the driest months.
(3) Spatial interpolation was performed based on the SPI at 25 rainfall stations. The spatial interpolation throughout the river basin was performed using multi-dimension inverse distance weighting (IDW) in ArcMap by Environmental Systems Research Institute (ESRI) [
21] to obtain the spatial distribution of SPI [
22]. Other approaches to define spatial distribution include using principle component analysis for clustering homogenous regions based on SPI [
23].
(4) Drought classification and mapping: Based on the spatial SPI, drought conditions were spatially classified based on Tshiabukole et al. [
4], as shown in
Table 3.
(5) Severity of drought: The severity of drought is based on monthly SPI. The longer the dry condition (as indicated by SPI), the more severe the drought. Severe drought conditions are experienced when the location is continuously dry condition for three months. The criteria for drought severity are shown in
Table 4.
(6) Location suitability: The suitability of reservoir locations based on the SPI was obtained from overlying the spatial mapping of the drought classification with the locations of the small reservoirs (constructed or planned).