Figures 6a and 6b show the calibrated monthly hydrograph of the Pantabangan Dam and the Arayat gauge, respectively, with the Nash-Sutcliffe (NS) model efficiency coefficient [Nash and Sutcliffe, 1970] and relative error (RE). The NS is defined as
where Qoi is observed discharge, Qsi is simulated discharge, n is total number of time series for comparison, and is mean discharge value observed over the simulation period. The NS equal to 1 corresponds to perfect matching between the modeled discharge and the observed data. The relative error was used to validate the water budget and is defined as
 The drought index was used to determine the beginning and end of moisture deficit periods and the degree of deficit severity, and to identify spatially drought-prone areas for the different drought types. The SA was used on all hydrological parameters, using similar categorization to SI, to identify if the droughts of 1983, 1987, 1991, and 1998 were clearly represented, and how drought occurred during these years for the different drought types. Table 2 shows the best-fit distribution pattern per month using the maximum likelihood estimates (MLEs) for each of the hydrological parameters using the corrected Akaike information criterion (AICc) (equation (5)) and the Bayesian information criterion (BIC) (equation (6)) with the best-fit model having the smallest AICc and BIC values [Akaike, 1974]:
where k is the number of estimated parameters including intercept and error terms in the model and n is the number of observations in the data set.
where k is the number of estimated parameters and n is the sample size.
 It has been found that the Weibull and the threshold Weibull (modified Weibull) distribution functions were the predominant distribution functions for rainfall. For the discharge, Weibull, threshold Weibull, lognormal, and Frechet distributions were identified. Soil moistures had a more variable distribution pattern. For the surface, LEV (largest extreme value), logistic, log logistic, Weibull, generalized gamma, SEV (smallest extreme value), and lognormal were the best-fit distributions, while for root-zone soil moisture, threshold Weibull, LEV, SEV, log generalized gamma, Weibull, normal, and lognormal distribution functions were identified. For the groundwater level, logistic, lognormal, normal, lognormal, LEV, SEV, and threshold Weibull were the best-fit distribution functions. Details of the different types of distribution functions can be found in the work of Meeker et al., . A similar method for selecting best-fit distribution for flood frequency was done by Hadded and Rahman  for Tasmania in Australia with lognormal distribution to be the best selection.
4.1. Temporal Analysis of Drought Parameters
 Monthly temporal distribution of SA for meteorological drought (rainfall), hydrological drought (streamflow and groundwater level), and agricultural drought (surface and root-zone soil moisture) are shown in Figure 8. SA values lower than −1.0 were observed during the drought years of 1983, 1987, 1990–1992, and 1998 reported by the local communities (as well as drought years 1993, 1994, and 1995). Three month and yearly running averages were also derived for the different parameters. SA was significantly below −1 during drought years. Longer time scales may not be useful in this basin because the dry period decreased frequency and intensity (e.g., drought for 1990–1992 disappeared for the 12 month running average) similar to the Vicente-Serrano and Lopez-Moreno  study on the Aragon River basin using SI.
Figure 8. Standardized anomaly index for monthly (a) rainfall, (b) discharge, (c) surface soil moisture, (d) root zone soil moisture, and (e) groundwater level for 1982 to 2000 at the outlet flowing to Manila Bay showing monthly values (blue), 3 month running average (green), and 12 month running average (orange). (Drought occurrence: 1983, 1987, 1990–1992, and 1998).
Download figure to PowerPoint
 For within-year droughts (Figure 9) (drought years selected were based on the recorded droughts during the study period), drought did not occur in all months. Individual droughts had different starting and end dates because of the time lag in which precipitation deficit reduced discharge infiltrating to affect soil moisture at the surface, further percolating down to affect soil moisture at the root zone and the groundwater level. For the 1983 drought, rainfall, discharge, and surface soil moisture had severe deficits from May, while root-zone soil moisture and groundwater deficits started in June and July. For 1987, drought conditions occurred in May for rainfall and discharge, while surface and root-zone soil moisture and groundwater deficits were severe in August. For 1991, there were no drought conditions for rainfall and discharge (lowest SA in June), and mild drought for surface soil moisture in July, while no drought conditions were observed for soil moisture and groundwater (lowest SA in July). For 1998, no drought conditions occurred for rainfall (consecutively low from January to August), while very mild drought conditions occurred for discharge (January, July, and August) and surface soil moisture (April and August). For both root-zone soil moisture and groundwater, drought conditions began in January until September. A time lag of 1 to 2 months was observed before drought in the soil surface reached the root zone and the groundwater, delaying agricultural drought by at least 1 month.
Figure 9. Standardized anomaly index for the basin average (a) rainfall, (b) discharge, (c) surface soil moisture, (d) root zone soil moisture, and (e) groundwater level for drought years 1983, 1987, 1991, and 1998 at the outlet to Manila Bay.
Download figure to PowerPoint
4.2. ENSO and Drought
 The general impacts of El Niño on climate in the Philippines by PAGASA included abnormalities such as delayed rainy season onset, early termination of the rainy season, weak monsoon and tropical cyclone activity, below normal rainfall, and above normal temperature.
 During the drought years 1983, 1987, 1990–1992, and 1998, Nino 3.4 indices were all greater than 0.5, indicating El Niño during these periods and supporting previous studies relating droughts and warm ENSO events. Table 3 shows the ENSO on the basis of Nino 3.4 SSTs, calculated using PAGASA's method of identifying ENSO events. Although El Niño occurred during drought years, it did not persist for the entire year in all cases. Table 4 shows 2 year composites of warm and cold events considered based on the Niño 3.4 index. These are consistent with the El Niño years (1982/1983, 1986/1987, 1991/1992, 1992/1993, 1994/1995, and 1997/1998) and La Niña years (1984/1985, 1988/1989, and 1995/1996) considered by Berri  and Jaranilla-Sanchez et al. . It should be noted that all four years of recorded drought events in the basin fell in the second year of the ENSO composites, further verifying that drought effects are significantly higher in the second year of ENSO.
Table 3. Classification of ENSO Events on the Basis of El Niño 3.4 SSTs From the Hadley Center Sea Surface Temperaturea
|1985||L-||L-|| || |
|1986||L-|| || ||E|
|1989||L||L-|| || |
|1990|| || || || |
|1992||E+||E|| || |
|1993|| ||E-|| || |
|1994|| || || ||E|
|1995||E-|| || ||L-|
|1996||L-|| || || |
Table 4. List of Warm (El Niño) and Cold (La Niña) ENSO Events Considered in the 2 Year Composites for the Years 1982 to 2000a
| Warm ENSO Events (El Niño) (Six Cases)||1982/1983, 1986/1987, 1991/1992, 1992/1993, 1994/1995, 1997/1998|
|Cold ENSO Events (La Niña) (Four Cases)||1984/1985, 1988/1989, 1995/1996, 1999/2000|
 The ENSO 2 year composites (Figure 10) of the five parameters showed similar trends during extreme events. The El Niño and La Niña reversals, which were in agreement with the findings of Lyon et al. , were also observed in the behavior of the hydrological parameters in the Pantabangan-Carranglan watershed in the northern portion of the Pampanga River basin [Jaranilla-Sanchez et al., 2009]. Results from the t tests further determined when the average values differed significantly for SA during El Niño and La Niña years (Table 5).
Figure 10. Standardized anomaly index categorized in the average 2 year composite (six cases of El Niño and four cases of La Niña years) SA for ENSO composites for (a) rainfall, (b) discharge, (c) surface layer soil moisture, (d) root zone soil moisture, and (e) groundwater level.
Download figure to PowerPoint
Figure 11. Spatial analysis showing mild to severe drought in July 1983, July 1987, June 1991, and August 1998 for rainfall, discharge, surface soil moisture, root-zone soil moisture, and groundwater level.
Download figure to PowerPoint
Figure 12. Hot spots: Spatial representation of the effects for July 1983, July 1987, June 1991, and August 1998 for meteorological drought (using rainfall parameter), hydrological drought (using the combined effects of discharge and groundwater level), agricultural drought (using soil moisture), and the combined effects of the different drought types.
Download figure to PowerPoint
Table 5. P Values for the Student t Test Comparing El Niño and La Niña 2 Year Composites for Rainfall, Discharge, Soil Moisture at the Surface, Soil Moisture at the Root Zone, and Groundwater Using SA (α = 0.05)
|El Niño versus La Niña: Using SA for Year 1|
|Soil Moisture (Surface)||0.30||0.67||0.54||0.86||0.65||0.07||0.99||0.59||0.59||<0.01||0.01||0.41|
|Soil Moisture (Root Zone)||0.63||0.52||0.54||0.70||0.90||0.40||0.94||0.50||0.36||0.01||0.01||0.13|
|El Niño versus La Niña: Using SA for Year 2|
|Soil Moisture (Surface)||0.10||0.03||0.01||<0.01||<0.01||<0.01||0.01||0.03||0.37||0.37||0.39||0.90|
|Soil Moisture (Root Zone)||0.11||0.06||0.02||<0.01||<0.01||<0.01||0.01||0.01||0.10||0.20||0.27||0.34|
 During the first year, rainfall, discharge, and soil moisture were significantly different in October and November (onset of the dry season), while no significant difference was found for the groundwater level.
 During the second year, significant differences for rainfall occurred in March and May, in April and May for discharge, in February to August for surface soil moisture, in March to August for soil moisture at the root zone, and in May to August for the groundwater level. This shows that the more severe effects of El Niño and La Niña appeared in the second year. Similar to the within-year drought analysis described in section 4.1, a time lag of about 1 month was observed in the second year after discharge was affected by rainfall deficit. Surface soil moisture has a slow recovery time from rainfall deficits in year 1 (February is toward the end of dry season), soil moisture at the root zone was delayed for at least 1 month and the groundwater level was further delayed 2 months after. During the early onset of ENSO, it is possible to utilize groundwater as an alternative water source during most of the dry season (November to March) because the effects of ENSO on groundwater start in May. May to August is the beginning of the wet season, and if rainfall is not sufficient to supply water during these months for crops during the planting season (May to June), entire crops will be delayed or abandoned.
 Similar studies on soil-moisture simulations and drought analysis [Sheffield et al., 2004; Wang. et al., 2009] in relation to surface warming [Dai et al., 2004] have been done. However, only studies in groundwater verified the time lag during drought events. Anderson and Emanuel  and Frappart et al.  showed that groundwater was significantly affected by ENSO. Precipitation is the mechanism by which the ENSO signal is transmitted to groundwater with a delay of about 1 to 3 months [Anderson and Emanuel, 2008]. In this study, a time delay of approximately 7 months was also found before precipitation deficit affected groundwater. The time lag here was longer because the study area is in the humid tropics, where rainfall is much higher than that in previous studies. It is possible that rainfall and discharge droughts occurred, but soil moisture and groundwater droughts did not. It is important to identify not only temporal variation in drought, but also the combined effects of the different drought types at certain time periods, and the exact location of where these occur in the basin.
4.3. Spatial Analysis of Droughts
 In the Philippines, the strongest effects of El Niño on production are during the dry season [Dawe, 2007]. From temporal and statistical analysis of the different hydrological parameters considered, different types of drought do not occur simultaneously and usually have a time lag of around 2–7 months when moisture deficit occurs from rainfall, trickling down to soil moisture and groundwater deficit. Spatial maps (Figures 11 and 12) on the most severe recorded droughts were constructed to show which hydrological parameters are currently affected and/or more drought-prone “hot spots” during selected time periods. The monthly spatial distribution of the different drought types calculated in SA was overlaid to identify hot spots in the basin most susceptible to combined drought effects during the most severe month of the particular drought year. From Figure 9, the months of July 1983, July 1987, June 1991, and August 1998 were selected in the four drought years, since these are the months where the lowest drought indices were observed. Rainfall was used to determine meteorological drought hot spots, discharge and groundwater level were combined to identify hydrological drought hot spots, and surface and root-zone soil moisture were combined to identify agricultural drought hot spots. Here, the coarse-resolution soil-moisture patterns (both at surface and root zone) are because of the use of the 5 min FAO soil properties as the model inputs.
 For July 1983, all five parameters (rainfall, discharge, surface soil moisture, root-zone soil moisture, and groundwater level) had mild to severe drought conditions, with severe conditions occurring in soil moisture and groundwater causing severe agricultural and hydrological droughts at the central plains of the basin. During this period, the upland areas in the north was affected mainly by drought conditions at the soil surface.
 Similar to the July 1983 results, July 1987 showed mild to severe drought conditions for all five parameters, with agricultural and hydrological drought occurring mainly in the central plains and severe meteorological drought occurring in the northern portion of the basin. During this time, the spatial distribution showed that the central plains of the basin were again affected severely by the combined drought effects.
 June 1991 showed that all five parameters were affected mildly by agricultural, hydrological, and meteorological droughts mostly in the central region. This third drought, which began from the end of 1990 up to the beginning of 1992, was the longest drought period recorded (and longest but only weak El Niño), but its overall effects at the basin scale caused only mild drought affecting portions of the central plain but not the northwestern uplands.
 The 1998 drought was one of the more severe droughts affecting the basin, and its effects on different drought types had some time lag wherein droughts caused by rainfall and discharge deficits mostly occurred in the central and southern plains of the basin, while droughts from soil moisture are in some portions of the central plains and drought from groundwater deficit was mild to moderate at the uplands and southwestern portion of the basin. The combined effects of the drought in this month were dominated mostly by meteorological and hydrological droughts having the most severe effects southwest of the basin. Soil moisture during this period had already recovered from the severe conditions at the earlier part of the year, resulting from the onset of the rainy season (commencing usually in May or June). Although meteorological and hydrological drought occurred in the area, agricultural drought occurred only in some portions of the basin and was not the major cause of drought.
 The selected months showed that different drought types occurred at different time scales and locations. However, some common areas usually affected by severe drought conditions for all four drought years include the upland areas (Pantabangan-Carranglan watershed (rain-fed area)) and the central plains of Pampanga. The uplands usually produce rain-fed crops, while the central portion of the watershed is the major rice production area in the basin (both lowland rain fed and some irrigated areas). If these areas are very susceptible to the combined effects of droughts, then adaptation strategies should be implemented to minimize their impacts, especially for rice-growing periods during El Niño years.
4.4. Impacts of Agricultural Drought on Rice Production
 Asian agriculture and water-resource sectors will most likely be affected by enhanced climate variability. Droughts associated with the 1997 to 1998 ENSO years in Myanmar, Laos, the Philippines, and Vietnam caused massive crop failures and water shortages, and forest fires occurred in various parts of the Philippines, Laos, and Cambodia [Duong, 2000; Kelly and Adger, 2000; Glantz, 2001; Philippine Atmospheric, Geophysical and Astronomical Services, 2001; Cruz et al., 2007]. Rice will be the agricultural crop most influenced in the short term [Brunner, 2002; Dawe, 2009; Roberts et al., 2009]. In the Philippines, agriculture contributed around 14.9% (CIA, available at https://www.cia.gov/library/publications/the world-factbook/geos/rp.html, 2010) of the country's gross domestic product (GDP) in 2009, with rice the major agricultural product.
 The Pampanga River basin is one of the major producers of rice in the country (around 2.09% based on palay (unmilled rice) volume production from 1994 to 2008 for the province of Pampanga, and 16.7% for the entire Region III from the Bureau of Agricultural Statistics ). Rice uses more water [Bouman et al., 2007] than other major crops (e.g., corn and sugarcane) grown in the area, so it is potentially more vulnerable to drought [Roberts et al., 2009]. The total agricultural area in the Pampanga River basin is 355,679 hectares (around 31.2% of basin surface area), and there are two types of rice agro-ecosystems [National Census and Statistics Office, 1971]: irrigated (183,253 hectares, around 51.53%) and rain fed (121,906 hectares, around 34.27%). Previous studies have shown that although rain-fed rice production is lower than irrigated systems because of its high susceptibility to droughts, it remains a vital source of income in the Philippines, especially for many poor farmers [Hossain et al., 2000; Roberts et al., 2009].
 El Niño can affect rice yields, especially in areas without irrigation [Dawe et al., 2009]. For this study, rice was the major crop used to identify El Niño effects on agricultural production. Detrending of constant price agricultural production was conducted for palay (unmilled rice), using the following linear equation:
 This was subtracted from the constant annual price. Here y is the annual price in millions of dollars, and x is the year of rice production (Figure 13a). Detrended constant price agricultural production of rice from 1967 to 2009 (Figure 13b) showed that losses (Table 6) occurred during drought years 1983 (−$33.79M ), 1987 (−$28.35M), 1990–1992 (−$32.98M, −$27.19M, and −$78.13M), and 1998 (−$217.11M), totaling to a loss of $417.55M. Since 16.7% comes from the region ($5.64M) in 1983; $4.73M in 1987; $5.51M, $4.54M, and $13.05M from 1990–1992 (similar to regional estimates from Jose et al. ); and $36.26M in 1998, and around 2.09% of this annual average comes from the Pampanga province ($0.71M in 1983; $0.59M in 1987; $0.69M, $0.57M, and $1.63M from 1990–1992; and $5.54M in 1998), this translates to a total of $8.73M in Pampanga, and $69.73M worth of losses for the entire region on rice alone for both rain-fed and irrigated areas.
Figure 13. Agricultural production of palay (unmilled rice) from 1967 to 2009 detrended using (a) simple linear correlation to calculate (b) annual production gains/losses.
Download figure to PowerPoint
Table 6. Estimated Agricultural Rice Production Losses Occurring During Years With Drought on the Basis of the 1967 to 2009 Production
|Drought Year||Estimated Nationwide Losses (in million US$)||Estimated losses in Region IIIa (in million US$)||Estimated Losses in Pampanga Province (in million US$)|
 To identify the relationship between ENSO, drought, and crop loss, a preliminary analysis on the values for the 3 month drought analysis for surface and root-zone soil moisture was compared with ENSO and loss in agricultural crop volume production. Figure 14 shows that, during El Niño years 1994 to 1995 (mild to moderate) and 1997 to 1998 (moderate to severe), a significant drop in rice volume stock (as shown by the negative crop volume anomaly) occurred. This shows how ENSO affected agricultural drought, resulting in agricultural losses. The crop volume anomalies followed closely the drought indices for both surface and root-zone soil moistures in the two El Niño events. Negative crop volume anomalies were observed during extreme events (moderate to severe El Niño and La Niña). However, because of some data limitations on crop volume, only two El Niño events can be verified in this study.
 For 1982 to 2000, the drought years 1983, 1987, 1990–1992, and 1998 showed that rice was very susceptible to drought. However, economic indicators such as annual crop prices and monthly stock volume are insufficient to characterize precisely how much agricultural loss was incurred during a given drought period. This is because of extraneous factors such as market health, increase/decrease in demand, environmental factors such as volcanic eruptions, and the availability of economic data (usually in terms of provincial or countrywide averages). Additional data are necessary for comprehensive quantification of crop yield losses resulting from drought.
 The UPRIIS identifies at least two crop-growing seasons for rice in the basin: wet cropping (June 16 to November 16, also called the first cropping, usually begins with the rainy season) and dry cropping (December 16 to May 16, also called the second cropping). On the average, crop volume for the first quarter is highest at 97,853.63 t; the second and third quarter averages follow at 72,388.88 and 71,658.25 t, and the lowest crop volume (the most critical months) is during the third quarter at only 15,596.75 t. In the 2 year ENSO composites, it can be seen that, during El Niño, drought is more significant in the second year (from February to August), indicating that the wet cropping is expected to be severely affected with drought.
 On the basis of the above analyses, other strategies can be devised for water-resources management. However, upland management and rain-fed agriculture practices should be verified to determine their suitability, especially in the identified hot-spot areas. Since water sources at the surface are limited, groundwater utilization, as suggested by local communities, could temporarily (within the 2–7 month time lag) be a viable option, especially since the hot-spot areas for the other parameters did not show synchronous severe water deficits in groundwater level during the dry season in El Niño years. Protection of this alternative water resource from contamination and over extraction is deemed necessary for future use. Another important strategy is rescheduling of agricultural activities and proper crop selection during drought years during the second year of El Niño. In worst cases, when drought is extreme, abandoning of the cropping season during this second El Niño year can be done as a last resort. Planning of alternative livelihoods for the community is necessary. Some time delays of around 1–3 months (during the dry cropping season) have been found before significant moisture deficit occurs in the soil that may lead to agricultural drought, so proper planning is needed to prepare for the inevitable effects to agriculture preceding the effects of meteorological and hydrological droughts.
 The Pampanga River basin is primarily agricultural and is one of the major producers of rice in the country. Significant agricultural losses were accrued during drought years, so appropriate adaptation strategies are needed to enable communities to cope with extreme drought conditions, especially in the “hot spots.” Drought is exacerbated by damage caused previously by the 1994 Mt. Pinatubo eruption, which altered the Pampanga River basin network and regional soils through lahars and tephra deposition. This means that some previously irrigated areas are now rain fed, and soil degradation has worsened. Historical data used as model inputs, and for agricultural ecosystems, show best-case scenarios for the outputs of the model on discharge, soil moisture, and groundwater level before and after the eruption of Mt. Pinatubo, as archived in global data sets. Despite the best-case scenarios, moderate to severe drought conditions have been observed, especially in the rain-fed and upland areas of the basin. Therefore, it is important to devise appropriate water-management strategies suited to the current state of the basin.