SEARCH

SEARCH BY CITATION

Keywords:

  • drought index;
  • PDSI;
  • soil moisture;
  • soil water index (SWI);
  • regional climate models (RCMs);
  • Greece

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Material and methods
  5. 3. Results and discussion
  6. 4. Summary and conclusions
  7. Acknowledgements
  8. References

This paper presents a framework for making use of drought indices in climate change impact assessment studies. To achieve this goal: (1) linear relationships between drought indices and satellite soil moisture information, derived from the ERS scatterometer [Soil Water Index, (SWI)] for the years 1992–2000, are developed by employing [analysis of covariance (ANCOVA)] and (2) the vulnerability of soil water content to climate change is assessed using regional climate model (RCM) projections. Several drought indices are evaluated for their abilities to monitor SWI, on a monthly basis, at nine locations in Greece. The original Palmer Drought Severity Index (Orig-PDSI) and its self-calibrated version (SC-PDSI) correlated best with SWI in three stations each and precipitation in two. The degree of agreement, however, varies substantially among the sites. Seasonality has a significant effect on the relationship between the SWI and the two aforementioned drought indices (Orig-PDSI, SC-PDSI), presenting a bimodal pattern that fluctuates markedly during the year. ANCOVA has proved to be a useful method for measuring the agreement between SWI and the drought indices (r2 ranged from 46.2% to 79.9%), implying that drought indices can be an important information source for detecting and monitoring drought. 11 different RCM runs are compared for their abilities to reproduce present climate mean and variability of temperature and precipitation. Orig-PDSI is not sensitive to the much warmer future climate change scenarios constructed and, therefore, is not suggested for climate change impacts assessment studies. SC-PDSI, on the other hand, has the potential to be used; however, its responses depend on the time period on which the climate characteristics and duration factors are computed from. Copyright © 2009 Royal Meteorological Society


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Material and methods
  5. 3. Results and discussion
  6. 4. Summary and conclusions
  7. Acknowledgements
  8. References

Soil moisture has a long memory in forcing atmospheric processes over land. In the short term, soil wetness controls the partitioning between latent and sensible heat flux on the earth's surface, affecting boundary-layer characteristics, as well as initiation and maintenance of convection. In the long term, soil moisture modulates floods and droughts (Pan et al., 2001). Drought is difficult to detect and to monitor because (1) it develops slowly; (2) its onset and ending are indistinct; (3) it is not precisely and universally defined; and (4) its impact is non-structural and often spreads over very large areas (Wilhite, 2000). Furthermore, area representative precise in situ soil moisture measurements are expensive and tedious to collect and as a result, only a few large scale measurement networks, providing widespread information of the ground dryness and wetness, exist (Robock et al., 2000). In order to quantify droughts and wet spells in terms of intensity, duration and spatial extent, numerous specialized indices (drought indices) have been devised (e.g., Heim, 2002).

In addition to drought indices, satellite sensor information has started playing an increasingly important role as an alternative to in situ soil water observations (Scipal and Wagner, 2004; Wagner et al., 2007). A global multi-annual remotely sensed soil moisture data set, derived from measurements taken by the ERS scatterometer, a coarse-resolution (50 km) active microwave instrument, was introduced lately (Scipal and Wagner, 2004). Evaluation studies employing modelled soil moisture data and field measurements indicated a reasonable overall performance of the satellite soil information (Section 2.2.2).

The threat that droughts pose to future climate sensitive economic sectors, including agriculture, has necessitated the assessment of potential impacts of climate change at various scales on soil moisture content. Research on how climate change will affect various ecosystems has progressed as an international effort on many fronts, the results of which have been summarized in the Intergovernmental Panel on Climate Change (IPCC) reports (IPCC, 2001). Scenario integrations of General Circulation Models (GCM) provide our best means of assessing the magnitude of future climate changes (Kattenberg et al., 1996). While IPCC assessments in 1990, 1996 and 2001 have attested that the climate modelling has improved dramatically over the last decade, GCMs perform less well when local detail at the grid-box level is considered, particularly for some important processes including soil moisture (e.g., Robock et al., 1998; Srinivasan et al., 2000; Seneviratne et al., 2002). However, most assessment studies have directly projected soil wetness in the future from global and regional climate model (RCM) output (e.g., IPCC, 2001; Meehl et al., 2006). This paper presents an alternative approach by initially developing relationships between drought indices and satellite soil information and afterwards assessing the vulnerability of soil moisture to climate change using RCM projections.

A few studies only, to our knowledge, evaluated drought indices for their abilities to assess soil moisture estimates (Dai et al., 2004; Mika et al., 2005; Szép et al., 2005). Furthermore, only in the first study was the assessment based on field measurements, while in the last two, ground wetness was proxied by the difference between precipitation and potential evapotranspiration (P-PET). In addition, the attractiveness of the first global satellite soil moisture data set, as a new source of soil wetness combined with the promising results of its application, prompted us to employ it as a means to develop our approach.

The main objectives of this study are (1) to evaluate several drought indices for their abilities to monitor soil moisture, on a monthly basis, at several locations in Greece. This is achieved by using soil moisture information derived from the ERS scatterometer (Soil Water Index, SWI) (Wagner et al., 2003) for the years 1992–2000. And (2) to assess the vulnerability of soil wetness to climate change using the most appropriate drought index, while high-resolution future scenarios have been provided by 11 RGM runs from the PRUDENCE project (Christensen and Christensen, 2007). The results (1) should improve upon current techniques used to employ satellite data for drought detection and future climatic impact assessment and (2) may allow crop producers to update and project soil water availability at regional scale for planning strategies on cropping, planting and harvest.

2. Material and methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Material and methods
  5. 3. Results and discussion
  6. 4. Summary and conclusions
  7. Acknowledgements
  8. References

2.1. Drought indices

There are several extensive reference lists of indices that the reader may consult about drought (e.g., Heim, 2002; Ntale and Gan, 2003). The original Palmer Drought Severity Index (Orig-PDSI) (Palmer, 1965) and the Standardized Precipitation Index (SPI) (McKee et al., 1993; 1995) are probably the most commonly used indices (Ntale and Gan, 2003). However, most studies that employed PDSI used its original algorithm that has serious weaknesses and limit its usefulness as a drought analysis tool (Heim, 2002). Wells (2004) developed a self-calibrated version of the PDSI algorithm (SC-PDSI) that dynamically calculates Palmer's algorithm empirical constants at any location automatically.

A summary of SPI and the two PDSI variations (Orig-PDSI and SC-PDSI) are described in detail elsewhere (Wells et al., 2004; Lloyd-Hughes and Saunders, 2002) and need not be repeated here. Kolmogorov–Smirnov (K-S) tests found the gamma distribution unsuitable for fitting the monthly precipitation of 1960–2000 for only 4 out of 28 stations over Greece and only for summer months with no precipitation (Loukas et al., 2003). Thus, the gamma distribution is chosen to model monthly precipitation in this study. Both codes used to calculate the monthly SPI and the two PDSI variations were downloaded by the NADSS website (http://nadss.unl.edu/spi/).

SC-PDSI was further revised (Mod-PDSI) by testing two variations of Priestlay–Taylor's approach (Priestley and Taylor, 1972) with measured and modelled (from air temperature data) solar radiation data (PTDC) as alternatives to Thornthwaite's (TH) method (Thornthwaite, 1948) used for computing PET (Mavromatis and Karacostas, 2005). Both models were compared against the estimated PET with Penman–Monteith approach (Allen et al., 1998) in two cropping regions, denoted as x in Figure 1, in Greece. A significant improvement in root mean squared error RMSE by over 48% in the southern location and 31% in the northern was achieved replacing TH with PTDC approach. Thus, the latter formulation was selected in this work, which requires temperature data (maximum and minimum air temperature) only to provide PET estimates for the mod-PDSI drought index.

thumbnail image

Figure 1. The locations of the nine meteorological stations in Greece

Download figure to PowerPoint

2.2. Data

2.2.1. Regional climate models

Monthly temperature and precipitation from 11 RCM runs were used to provide input for the drought indices (Table I). The RCMs are described in detail elsewhere (Jacob et al., 2007) and need not be repeated here. For most model runs, control simulations corresponded to the period 1961–1990 while future scenarios for 2071–2100 corresponded to the A2 IPCC CO2 emission scenarios (SRES A2) (IPCC, 2001). Five RCMs were run using boundary conditions from HadAM3H: CHRM, REMO, RCAO, RegCM and RACMO. HIRHAM was alternatively driven by HadAM3H and ECHAM5. Two RCMs (HIRHAM and RCAO) and HIRHAM performed the standard experiment twice and four times the original resolution, i.e., around 25 km and 12 km, respectively, instead of 50 km (Table I).

Table I. The regional climate models (RCM) driven by global circulation models (GCM). The grid resolution and the control simulation period for each run are also shown
RCMGCMGrid resolution (km)Control simulation
HIRHAM-HHadAM3H501961–1990
HIRHAM-EECHAM5501961–1990
HIRHAM25HadAM3H251961–1990
HIRHAM12HadAM3H121961–1990
HadRM3PHadAM3P501960–1990
CHRMHadAM3H551961–1990
REMOHadAM3H551960–1990
RCAOHadAM3H501961–1990
RCAO25HadAM3H251961–1990
RegCMHadAM3H501961–1990
RACMOHadAM3H501961–1990
2.2.2. Remotely sensed soil moisture

The profile of soil moisture or SWI is a relative measure of soil wetness of the 1-m soil layer ranging between wilting level (0%) and field capacity (100%) (Scipal and Wagner, 2004). This data set derived from measurements taken by the ERS scatterometer (a radar operated in C-band) during the period 1992–2000. The retrieval algorithm of SWI was based on a change detection approach that naturally accounts for surface roughness and heterogeneous land cover (Wagner et al., 1999; Wagner et al., 2003). The temporal resolution was based on the specifications of the infiltration model used to derive SWI and was set for 10 days. Since the scatterometer could not be operated at the same time as the synthetic aperture radar (SAR) mounted on ERS satellite, temporal and spatial sampling rate was often much reduced, mainly over Europe where the SAR was often on (Wagner et al., 2003; Ceballos et al., 2005).

The quality of soil moisture data derived from the scatterometer was favorably compared with ground wetness information from various sources for a variety of climates including Mediterranean (Wagner et al., 2003; Campling and De Belder, 2003; Ceballos et al., 2005; Pellarin et al., 2008; Laguardia et al., 2006; Zhao et al., 2006). In particular, an r2 of 0.75 for the averaged 0–100 cm soil moisture profile and an RMSE of 2.2 vol% were found between SWI and field observations from the REMEDHUS soil moisture network in a semi-arid Mediterranean environment in Spain (Ceballos et al., 2005). In addition, a reasonable agreement under Mediterranean conditions, r ranged from 0.6 to 0.64, was found between the monthly SWI and global modelled ground moisture of the 0–50 cm layer (Wagner et al., 2003).

2.2.3. The study area and station data

The geographic focus of this study includes nine sites (Figure 1), representative mainly of the climate in continental Greece. For each site shown in Table II (which presents a descriptive contrast between the temperature and precipitation regimes of these stations) the nearest scatterometer pixel (grid) was selected. The mean annual rainfall ranged from 346 mm in Athens to 1060 mm in Ioannina. The air temperature varied from 12.5 °C in Kozani to 18.3 °C in Athens. As a result of the operational conflict with the SAR, the scatterometer did not cover most of the southern part of Greece, including Crete island. Due to the coarse resolution (50 km) and the gaps in coverage, no stations from islands were included either, as it was not possible to identify a sufficiently close scatterometer pixel to any of these stations. The monthly precipitation and air temperature data used for the computation of drought indices for each station are shown in the same table. Representative values of the available water capacity (AWC) for the area of each site, a soil characteristic required by Palmer's algorithm, were identified from the Digital Soil Map of the World (DSMW).

Table II. The weather stations used in the study. The coordinates (latitude and longitude), the altitude, the mean monthly temperature, precipitation and the standard deviation (SD) of the data period and, the available water capacity (AWC) for each station are also shown
StationLat. (N°)Long. (E°)Alt. (m)Data periodPrec. (SD)Temp. (SD)AWC (mm)
Thessaloniki40.62°22.95°311950–2001451 (95)15.9 (0.5)175
Athens37.97°23.72°1071950–2002346 (97)18.3 (0.6)150
Kalamata37.07°22.02°71958–2000759 (170)17.1 (0.6)175
Tripoli37.52°22.40°6631958–2000781 (181)13.4 (0.6)175
Agrinio38.62°21.45°461958–2000885 (193)16.6 (0.5)150
Larisa39.63°22.42°731958–2000411 (90)15.1 (0.5)275
Ioannina39.67°20.85°4841958–20001060 (212)13.5 (0.4)200
Kozani40.30°21.78°6251958–2000491 (126)12.5 (0.6)150
Alexandroupoli40.85°25.92°41958–2000536 (125)14.1 (0.5)275

2.3. Methodology

The covariation between the monthly time series of SWI and several drought indices (the SPI, the three PDSI variations (the Orig-PDSI, the SC-PDSI and the modified version of SC-PDSI with the improved PET scheme (Mod-PDSI)), their respective moisture anomaly indices, P and P-PET, during 1992–2000, was clarified by employing Pearson correlation analysis (Figure 2). To investigate whether the relationship between drought indices and SWI varies within a year, we processed each month separately. This association was further tested and measured by conducting analysis of covariance (ANCOVA) using the GLM procedure of SPSS (Figure 2). The differences in intercepts and slopes between the regression functions for each month, caused by the ‘month’ effect on the SWI–drought index relationship, were evaluated by the mean squared error (MSE), which was calculated as the Type III Sum of Squares divided by the associated degrees of freedom (df), and the significance was evaluated by F-tests. A high MSE of one factor compared with others show that this factor contributes significantly to explaining the total variation in SWI. Still, this may not be significant, if the overall variation explained by the statistical model is low. The partial eta squared statistic, which reports the ‘partial’ significance of each term (based upon the ratio of the variation (sum of squares) accounted for by the term, to the sum of the variation accounted by the term and the variation left to error) is also reported. In comparison, simple regression models, thus assuming the intercepts and slopes are identical during the year, were also developed for each location between SWI and the drought indices (Figure 2). The coefficient of determination r2 and RMSE were used to measure goodness-of-fit of the model between SWI and drought indices for each location.

thumbnail image

Figure 2. Overview of steps for using drought indices in assessment of soil moisture response to climate change. See the specific sections and Section 2.3 for details

Download figure to PowerPoint

In order to choose the most suitable RCM run to provide input for the drought indices, the monthly modelled temperature and precipitation for present-day climate (1960 or 1961–1990) from the 11 runs in Table I were compared to the corresponding observed data series by conducting correlation- and regression-based analysis (Figure 2). The grid cell center location closest to the latitude and longitude of each station was assumed to represent conditions at the location. Differences between model and station latitude have been accounted for using a lapse rate of 6.5°K/km. This average lapse rate, however, does not take into consideration other local deviations or specific weather conditions. Depending on the differences between the RCM output and observations, the soil moisture responses to future climate will be assessed through the most appropriate drought index by developing climate change scenarios, either by using the modelled monthly output of 2071–2100 directly or by just applying appropriately the model differences of [(2071–2100) − (1960 or 1961–1990)] to the observed baseline climatology (Figure 2).

The following general categories that indicate a quick way of interpreting the correlation coefficient r was utilized (Hinkle et al., 1994): 0.0 < r < 0.3: very weak (if any) correlation; 0.3 < r < 0.5: weak correlation; 0.5 < r < 0.7: moderate correlation; 0.7 < r < 0.9: high (strong) correlation; 0.9 < r < 1: very high correlation. The reduced major axis method for slope estimates was preferred over the more frequently used ordinary least squares method properties due to the desirable properties it possesses (Ricker, 1984). Regression analysis statistics including the slope and intercept of the regression lines, RMSE and the modified modelling efficiency EF1 (Yang et al., 2000) were computed with the data analysis tool IRENE (http://www.isci.it/tools). EF1 is a model performance statistic based on absolute differences that can get positive or negative values with 1 being the optimal value.

3. Results and discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Material and methods
  5. 3. Results and discussion
  6. 4. Summary and conclusions
  7. Acknowledgements
  8. References

3.1. The relationship of drought indices with SWI

3.1.1. Covariation and seasonal timing of drought indices with SWI

The distributions of monthly correlation coefficients r between satellite soil moisture data and the drought measures (including P and P-PET) for all stations during 1992–2000 are compared in Figure 3. SWI correlated more consistently, although weakly, with Orig-PDSI and SC-PDSI (the median was equal to 0.44 and 0.4, respectively). Although this difference in the median values was not statistically significant according to the Mann–Whitney rank sum (MWRS) test (P = 0.633), the former drought index (Orig-PDSI), compared to the latter (SC-PDSI), presented slightly and substantially higher scores of the 75th and 25th percentiles, respectively. Furthermore, while statistically significant differences (P = 0.033) were identified in the median values between the two moisture anomaly indices (Orig-Z and SC-Z) and Orig-PDSI, according to MWRS test, the respective differences between the same anomaly indices and SC-PDSI were not great enough (P ≥ 0.1). The slightly better performance of the Palmer's original code vs. its self-calibrated version was not expected since Orig-PDSI is tuned to the conditions in the central United States. The relationship between the moisture anomaly indices and precipitation, on the one hand, and SWI, on the other, is generally weaker, as a result of the high-frequency variations they present compared to the PDSI versions (Dai et al., 2004). The distributions of all drought measures in Figure 3 showed a relatively minor variation in the central location and negative skew. The discrepancies between SWI and the drought measures are due to: (1) retrieval errors from the scatterometer data (Wagner et al., 2003); According to Wagner et al. (2007) ‘the retrieval of soil moisture from the satellite data is a challenging problem due to the confounding influences of other surface variables such as vegetation or surface roughness, and the complex structure of the Earth's surface’. The choice of the retrieval algorithms is important, since they should describe the physical measurement process with sufficient detail, yet be simple enough to allow robust inversion of the satellite data; (2) drought index simulations (due to model deficiencies and inaccuracies of the input data); (3) other meteorological factors that become important determinants of soil moisture content, such as wind speed, air humidity and solar irradiance, but are not considered by any of the drought measures used in this study; and (4) the fact that satellites gather information about the topmost soil layer (Wagner et al., 2007), while Palmer's algorithm variations represent a soil layer beneath the surface. In addition, the drought indices are not designed to be, and should not be considered to be, according to Dai et al. (2004), a direct measure of soil moisture content. Therefore, perfect correlations between these two, should not be expected. To assess the contribution of each of these issues to the discrepancies between SWI and the drought measures is a very difficult task. The Orig-PDSI and SC-PDSI outperformed SPI because: (1) PDSI index is more physically-based than SPI and (2) SPI solely relies on precipitation in contrast to Palmer's algorithm that uses both precipitation and air temperature as input. This allows PDSI to take into consideration the basic effect of surface warming which accounts for 10–30% of PDSI's variance coming through PET, on drought and wet spells (Dai et al., 2004).

thumbnail image

Figure 3. The distributions of monthly correlation coefficients r [n = 108 (12 × 9)] between SWI and the drought indices (including P and P-PET) for all stations during 1992–2000. The median (solid line), means (dotted line) and extreme (outliers) values are also shown

Download figure to PowerPoint

Equally high (r > 0.7) values in the median of monthly correlation distributions between two drought indices (Orig-PDSI and SC-PDSI) and SWI were found in Alexandroupoli (Figure 4). In two stations (Athens, and Kozani), Orig-PDSI was the best index presenting moderate and weak agreement, respectively. Orig-PDSI and Mod-PDSI exhibited similar moderate relationships with SWI in Thessaloniki. Precipitation and SC-PDSI were the best choices in two (Ioannina and Kalamata) and three (Agrinio, Larisa and Tripoli) locations, accordingly, showing weak (with the exception of Larisa where a moderate association was found (Figure 4)) agreement with the satellite soil moisture. It should be added at this point that (1) the unremarkable improvement of the index's performance in Athens and Thessaloniki, when improving the representation of PET in the SC-PDSI algorithm and (2) the lack of sufficient meteorological data required for the development of the PTDC approach in the rest sites, incited us not to consider SC-PDSI and SC-Z in the covariation analysis or any further in this study.

thumbnail image

Figure 4. The distributions of monthly correlation coefficients r (n = 12) between SWI and the best drought index for each station during 1992–2000. The median (solid line), means (dotted line) and extreme (outliers) values are also shown

Download figure to PowerPoint

The seasonal timing of the two most consistent drought indices (Orig-PDSI and SC-PDSI) with SWI was further investigated. Correlation analysis on a monthly basis demonstrated that the relationship between them exhibits a similar, for both drought indices, bimodal pattern that varies markedly during the year (not shown). A moderate agreement appears from January to March, which is the rainy and relatively cold part of the year, and a much weaker correlation during the drier and warmer months (from April to September) with a second maximum in October.

3.1.2. Goodness of model fit between drought indices and SWI

Table III presents the results for the comparison of the two regression models (ANCOVA and a simple regression model, see Section 2.3) between SWI and Orig-PDSI in Thessaloniki. The significance value of each term is statistically important (<0.05). In contrast, the respective statistic of the interaction term was much greater (0.936); that accounts for a negligible amount of variation compared to the error term and therefore we can assume homogeneity of the coefficient of Orig-PDSI within a year. Much larger portion of the variation in SWI can be attributed rather to monthly variations than to drought index fluctuations (as evidenced by the substantial difference of the partial eta squared values between these two in Table III). Although the simple regression model is significant, r2 = 0.224 is significantly lower than the respective with ANCOVA analysis (r2 = 0.752).

Table III. Analysis of covariance (ANCOVA) and simple regression analysis between SWI and Orig-PDSI for Thessaloniki (the degrees of freedom are 96 instead of 108 (9 × 12) due to 12 months of missing data)
ANCOVA
Dependent Variable: SWI_Thes
SourceType III sum of squaresdfMean squareFSig.Partial eta squared
Corrected mode30 958.415a122579.868210.0000.752
Intercept104 374.6031104 374.68500.0000.911
MONTH21 749.046111977.186160.0000.681
Orig-PDSI_Thes4067.16914067.169330.0000.285
Error10 187.50583122.741   
Total164 305.11096    
Corrected total41 145.92095    
Simple regressiona
ModelUnstandardized coefficientsStandardized coefficients  
 BStd. ErrorBetatSig.
  • a

    r Squared = 0.752

  • a

    r Squared = 0.224

1 (Constant)43.5342.395 18.20.000
 Orig-PDSI_Thes5.2301.0050.4735.2060.000

ANCOVA has proved to be an effective method for measuring the agreement between the satellite soil moisture data and the drought indices since it was able to explain the majority of SWI variability (r2 ranged from 0.462 in Tripoli to 0.799 in Larisa) (Table IV). In seven stations, the explained fraction of the variation in the dependent variable was accounted for by the individual terms only with the ‘month’ factor being substantially more important than the ‘drought index’ factor. In Kalamata, the explained SWI variability by ANCOVA was due to the interaction term and the ‘month’ factor. The full factorial model was implemented in Tripoli only. The simple regression fails to account for the monthly effect and results in poor estimates of the relationship between SWI and the drought measures (r2 ranged from 0.01 in Tripoli to 0.24 in Kalamata and RMSE were much higher than the respective with ANCOVA (Table IV)). Additional reasons that may have contributed to the differences between SWI and drought indices include: (1) possible nonlinear SWI behaviour, obviously, not accounted for by the linear approach and (2) the fact that indices reflecting soil water balance, such as PDSI, do not properly model excess rainfall (Akinremi et al., 1996).

Table IV. Analysis of covariance (ANCOVA) and simple regression analysis between SWI and the best drought measure(s) for each location. The coefficient of determination r2, the root mean squared error (RMSE), and the size of samples (n) for both regression analyses and the significant factors (‘mo’ and ‘di’ stand for month and drought index factor, respectively) for ANCOVA are also shown
StationDrought IndexANCOVARegression
  r2RMSESignificant Factorsr2RMSEn
ThessalonikiOrig-PDSI0.75210.3‘mo’, ‘di’0.22418.296
AthensOrig-PDSI0.7983.4‘mo’, ‘di’0.1816.887
KalamataP0.5693.2‘mo’, ‘mo × di’0.2404.390
TripoliSC-PDSI0.4623.9‘mo’, ‘di’, ‘mo × di’0.0105.588
AgrinioSC-PDSI0.5343.8‘mo’, ‘di’0.1065.397
LarisaSC-PDSI0.7995.5‘mo’, ‘di’0.11911.596
IoanninaP0.7343.5‘mo’,‘di’0.1526.380
KozaniOrig-PDSI0.58810.6‘mo’, ‘di’0.10015.696
AlexandroupoliOrig-PDSI0.7779.2‘mo’, ‘di’0.21117.2105
 SC-PDSI0.7789.1‘mo’, ‘di’0.19717.4105

In addition, a moderate linear association (r = 0.0014· AWC + 0.402, r = 0.536) was identified between the distribution of r2s in Table IV with the respective AWC of each location (Table II) indicating that there are spatial factors affecting the relationship between the satellite data and the drought indices. This relationship suggests that the association of SWI with the best for each site drought measure is stronger when soil has a higher water storage capacity.

3.2. Regional Climate Models

3.2.1. Model performance in present-day climate

The differences in mean monthly air temperature time series between the RCM output and observations ranged from −8.7% (or 1.3 °C) for REMO to + 9.6% for HIRHAM12 (or 1.4 °C) (Table V). The majority of RCMs was consistently warm with six of them (RCAO, HIRHAM25, RCAO22, HIRHAM50-E, HIRHAM12 and, HadRM3p) overestimating the measured temperatures at six sites or more for at least three seasons. RegCM and REMO, on the other hand, were consistently cold at five stations for three seasons or more. The mean monthly summer (winter) temperature of the nine stations was overestimated by ten (underestimated by eight) RCMs. All model runs were too warm for Kalamata and Larisa and too cold (except for CHRM) for Athens. RCAO25, RCAO, HIRHAM25 and RACMO agreed more closely with the observed data as they produced (1) the least amount of errors (as evidenced by the MAE and RMSE in Table V), (2) regression lines with intercepts and slopes closest to 0 and 1, respectively, and (3) the closest to unity EF1 estimates.

Table V. Comparison of mean monthly observed, and simulated from regional climate models, air temperature for all stations (n = 108, 9 stations × 12). The correlation coefficient r, mean absolute error (MAE), the modified modelling efficiency (EF1), root mean squared error (RMSE) and, intercept (Int.) and slope (Slop.) of the regression line are also given
 Mean ( °C)rMAE ( °C)EF1Int. ( °C)Slop.RMSE ( °C)
Obs 1961–199015.1      
HIRHAM-H15.40.961.70.72−1.131.102.12
HIRHAM-E16.20.952.30.61−2.531.243.10
HIRHAM2515.70.991.20.80−1.601.151.62
HIRHAM1216.60.981.70.71−0.841.152.19
CHRM15.00.971.90.682.601.162.22
RegCM14.50.942.20.62−2.561.132.70
RCAO16.10.981.30.770.051.061.69
RCAO2515.80.991.10.81−1.021.111.50
RACMO15.60.991.20.80−1.261.111.52
 Mean ( °C)rMAE ( °C)EF1Int. ( °C)Slop.RMSE ( °C)
Obs 1960–9015.2      
REMO13.80.962.20.63−3.241.132.65
HadRM3p15.70.981.50.74−2.121.182.05

The temporal comparison of mean monthly precipitation between the RCMs and the historic measurements for the present-day climate, exhibited consistent model behaviour at all locations. All but three runs (RegCM, REMO and HadRM3p) underestimated station rainfall, from 5.6% for RCAO25 to 33.9% for CHRM (Table VI), with eight of them simulating less precipitation at three seasons (HIRHAM25, RCAO25, HIRHAM-E and HIRHAM12) or more (RCAO, HIRHAM-H, CHRM and RACMO). The average monthly rainfall in winter and autumn was underpredicted by ten and nine RCMs, respectively. Ten models were too dry in Alexandroupoli and Kalamata and nine in Larisa, Thessaloniki and Tripoli. RCAO25, RCAO and HIRHAM12 produced, overall, the best agreement with station data (as evidenced by the correlation and regression-based results in Table VI). Furthermore, the estimates of coefficient of variation (CV) (expressed as SD/mean × 100) of the modelled data were well within 20% of the corresponding values of the observations.

Table VI. Comparison of mean monthly observed and simulated from regional climate models precipitation for all stations (n = 108, 9 stations × 12). The correlation coefficient r, mean absolute error (MAE), the modified modelling efficiency (EF1), root mean squared error (RMSE) and, intercept (Int.) and slope (Slop.) of the regression line are also given
 Mean (equation image)rMAE (equation image)EF1Int. (equation image)Slop. (equation image)RMSE (equation image)
Obs 1961–9054.0      
HIRHAM-H41.00.5425.70.144.60.6835.3
HIRHAM-E40.70.7321.90.26−5.50.8630.0
HIRHAM2540.70.6924.40.18−2.90.8131.4
HIRHAM1247.30.7519.40.35−6.71.028.2
CHRM35.70.6524.50.183.10.6034.5
RegCM64.90.6227.00.0912.00.9835.2
RCAO42.30.8317.80.40−6.80.9124.6
RCAO2551.00.8814.30.52−0.80.9619.1
RACMO39.90.7821.40.28−10.30.9328.5
 Mean (equation image)rMAE (equation image)EF1Int. (equation image)Slop. (equation image)RMSE (equation image)
Obs 1960–199054.3      
REMO62.30.6623.80.2111.40.9432.2
HadRM3p60.90.7328.00.07−20.51.5041.0

Jones et al. (2003) defined acceptable differences of around ± 1 °C for monthly temperature, due to ‘elevational’ differences in the way monthly temperature is calculated by the models and observations and for other variables of being within about 30% of the CV of the measured variable. Among the best RCMs identified for simulating present climate, RCAO25 reproduced better, in terms of the summary statistics considered, both mean monthly temperature and precipitation. It is concluded that the output from that model run is close enough to reality and hence, simulations for future time periods are likely to provide meaningful information. However, the discrepancies between monthly Orig-PDSI and SC-PDSI series with observed and modelled input are sufficiently large that using RCAO25 output for 2071–2100 directly with the drought indices would lead to dubious future soil moisture responses. Therefore, the assessment of soil moisture content to future climate was attempted with climate change scenarios derived from applying the mean monthly rainfall and temperature anomalies computed from RCAO25 output to baseline climate. For temperature, the difference in temperature between 2071–2100 and 1961–1990 simulated periods was added to the recorded monthly temperature data for the baseline period. For rainfall, the ratios in modelled precipitation {[(2071–2100) − (1961–1990)]/(1961–1990)}, on a monthly basis, were multiplied by the observed precipitation values.

3.2.2. Future scenarios

The climate scenarios from RCAO25 predicted for 2071–2100 average annual air temperature increases are varied from that of 3.0 °C in Larisa to 5.6 °C in Kalamata, when compared to the respective modelled data of the control period. In Agrinio, the respective statistic was much lower (1.7 °C). On seasonal basis, the corresponding projections indicated consistent, for all sites, temporal courses with mean increases, in the range of 3.4 °C for spring to 5.2 °C for summer, which varied markedly during the year (Figure 5a). The future projections of rainfall presented much more variation on an annual basis (from minor reductions of −1.6% and −6.5% in Ioannina and Tripoli, respectively, to substantial increases of 144.5% in Larisa and 150% in Kozani). On a seasonal basis, the climate change scenarios predicted large reductions in summer rainfall amounts (ranged from −36.4% in Kozani to −93.6% in Alexandroupoli) and much more substantial increases in autumn and winter (except for Kalamata and Agrinio) (Figure 5b).

thumbnail image

Figure 5. Climate change projections of (a) temperature (T2071–2100 − T1961–1990) and (b) precipitation ratio (Prec2071–2100/Prec1961–1990) from RCAO25 for each station for the period 2071–2100. This figure is available in colour online at www.interscience.wiley.com/ijoc

Download figure to PowerPoint

Subsequently, the effects of the temperature and precipitation anomalies projected from RCAO25 on historical monthly climate data were investigated by comparing the respective time series of the selected for each station, drought index (Table VI). In the case of Orig-PDSI, these time series were produced when monthly temperature and precipitation for the observed and perturbed climate were separately used as input. For SC-PDSI, the comparison between the index values, calculated with actual and future climate, was achieved by forcing SC-PDSI to calibrate over the historical data set so that its computed values, with perturbed and actual input, would quantify drought in the future climate in a manner that reflects deviations from the historical climate (Mavromatis, 2007). In other words, the climatic characteristic and duration factors calculated, based upon the historical climate, were afterwards applied to estimate SC-PDSI values for the future climate projected with RCAO25.

The striking difference between the two drought index responses with observed and perturbed climates is demonstrated in Figure 6 for Alexandroupoli, where both drought indices are equally efficient (see Figure 4). A very good agreement (r = 0.887) was found between the Orig-PDSI computed from the observed records and the climate change scenarios (Figure 6a). Both the K-S two-sample test and the frequency histogram of drought severity levels, confirmed the above-mentioned conclusion. In the case of SC-PDSI, however, its computed values with the perturbed climate series, significantly and consistently underestimated the respective values with present climate data (Figure 6b). Regarding Orig-PDSI, the effects of climate change scenarios were ‘absorbed’ by the soil moisture anomaly Z. The monthly time series of Z, for observed and perturbed climate, followed the same pattern (not shown), since the two climates differ only by the temperature/precipitation change suggested by RCAO25. The strong agreement of the two Orig-PDSI time series in Figure 6a is a result of the small contribution of Z

  • equation image(1)

where i is the current month and p = 0.897 and q = 1/3 are the duration factors to the estimated PDSI values. The response of SC-PDSI to the same climate change scenario was dramatically different (Figure 6b), since both the duration factors (p, q) (which determine the drought index sensitivity to deficit and excess moisture conditions and scale the final PDSI value so that approximately 2% of the values are + /−4 (Wells et al.2004)) and the climatic characteristic K (Zi = Ki·di where d is the moisture departure) which is used to calculate drought index value in Eqn (1), were computed based on the baseline climate which, in our case, is much colder and wetter than the future one. As a consequence, the SC-PDSI index stayed highly negative (<−4) for most of the perturbed climate.

thumbnail image

Figure 6. A comparison between (a) the original PDSI (Orig-PDSI) with observed (1958–2000) and future (observed + climate change projections) climate and, (b) the self-calibrated PDSI (SC-PDSI) with observed (1958–2000) and future with full calibration and calibration in Z only climate, in Alexandroupoli

Download figure to PowerPoint

3.3. Soil moisture response to future scenarios

When the soil moisture model, based on Orig-PDSI, was combined with the corresponding drought index estimates (derived from the future scenarios of temperature and precipitation projected from RCAO25) it predicted minor moisture changes that ranged from −1.7% in Thessaloniki to 0.3% in Alexandroupoli (Table VII). The year-to-year soil moisture variability presented minor increases in Alexandroupoli (0.4%) and slightly larger decreases in Thessaloniki and Athens (−2.1% and −3.8%, respectively). These very small changes were due to the strong agreement of Orig-PDSI for observed and perturbed climates. The response of the soil moisture models based on the SC-PDSI to similar climate change scenarios was dramatically different (for the reasons explained in the previous section) with substantial moisture reductions varying from 10.2% in Tripoli to 42.5% in Alexandroupoli. In contrast, the dramatic rainfall increases in Larisa in autumn, winter and spring (214%, 196% and 220%, accordingly), despite the drier summer conditions by 53%, and the relatively smaller temperature rise (by 3.0 °C on annual basis) predicted by RCAO25, caused dramatic soil moisture increases by 133%. The interannual variation of moisture content, however, followed opposite direction with the projections in the mean (except for Agrinio) and ranged from 2.1% in Alexandroupoli to 32.1% in Tripoli (Table VII). Future precipitation projections in Kalamata and Ioannina resulted to very small increases in the average conditions of soil moisture (by 0.4% and 0.9%, respectively) with more substantial changes in year-to-year variability.

Table VII. Mean and interannual (SD, standard deviation) estimations of soil moisture responses to future climate for each drought index/station combination for full calibration and calibration in moisture anomaly Z only
StationDrought IndexFull CalibrationCalibration in Z only
  Mean (%)SD (%)Mean (%)SD (%)
ThessalonikiOrig-PDSI−1.7−2.1
AthensOrig-PDSI−1.0−3.8
KalamataP0.4−7.4
TripoliSC-PDSI−10.232.1−20.795.1
AgrinioSC-PDSI−32.9−2.6−23.33.0
LarisaSC-PDSI132.720.3116.523.4
IoanninaP0.99.6
KozaniOrig-PDSI−0.4−4.1
AlexandroupoliOrig-PDSI0.30.4
 SC-PDSI−42.52.1−39.410.8

A question is raised whether the computation of climate characteristics and duration factors in SC-PDSI should be limited to the calibration interval only (Dr S. Goddard, 2006, personal communication). Therefore, another version of SC-PDSI was developed that computed the climate characteristic and duration factors over the entire climate period (historic + perturbed period), but the wet- and dry-spell calibration coefficients, which affect Z, were estimated based on data from the baseline climate only.

Calibrating for Z only, resulted in a monthly time series of SC-PDSI for the perturbed climate in Tripoli following a similar pattern (r = 0.76) of those of the historical one. The difference between the two climates (observed vs future), which resulted from a significant increase in the frequency of extremely dry spells in future climate conditions (from 15.3% to 97.7%), caused increased soil moisture losses in terms of the mean (−10.2% vs −20.7%) and in the year-to-year variability (32.1% vs 95.1%) when compared to the ‘full calibration’ scenario (Table VII). In the worst case scenario, extremely wet to moderate dry spells will be completely or almost (frequency less than 1%) disappear in the future climate. Opposite impacts were found in Agrinio and Alexandroupoli, when Z was estimated from the baseline climate only, with lower decreases in the average soil moisture by 9.6% (from 32.9% to 23.3%) in the former location and by 3.1% (from 42.5% to 39.4%) in the latter. These responses derived from a significant reduction in the frequency of extremely dry months, from 80% to 58.4% in Agrinio and from 65.4% to 53.6% in Alexandroupoli (Figure 6b). The two calibration approaches had similar effects in both mean and variability of soil moisture in Larisa. Raises by more than 115% and 20% in the respective statistics were found at this site (Table VII) as a result of the increased frequency of extremely wet spells (by at least 90%) due to enhanced rainfall amounts (see the first paragraph of this section) predicted by RCAO25 for the future climate.

4. Summary and conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Material and methods
  5. 3. Results and discussion
  6. 4. Summary and conclusions
  7. Acknowledgements
  8. References

This paper presented a framework for making use of drought indices in climate change impact assessment studies. It was implemented for soil moisture vulnerability response to future climate, once relationships, linear in this work, between drought indices and soil moisture information were developed. In this study, soil moisture information derived from the ERS scatterometer (SWI), during 1992–2000, was used. As the first stage, several drought indices (including P and P-PET) were evaluated for their abilities to monitor satellite soil moisture information, on a monthly basis, at nine locations in Greece. Orig-PDSI and SC-PDSI correlated best with SWI in three stations each and precipitation in two. The first two indices were equally efficient in one more station. The degree of agreement, however, varied substantially among the locations. The relationship between SWI and the drought measures is far more complex, as suggested by the several issues mentioned in Section 3.1.1, than what can be represented by a simple linear correlation between these variables. Seasonality has a significant effect on the relationship between the SWI and Orig-PDSI or SC-PDSI presenting a bimodal pattern that fluctuates markedly during the year. The ANCOVA has proved to be a useful method for measuring the agreement between the remotely sensed soil moisture data and the drought indices, since it was able to explain the the SWI variability of 46.2% to 79.9%. These results imply that drought indices could be effective indicators of moisture condition and an important information source when used for detecting and monitoring drought in the selected regions. Simple regression failed to account for the seasonality effect and resulted in poor estimates of the relationship between satellite sensor data and the drought measures (r2 ranged from 0.010 to 0.224). 11 different RCM runs were compared for their abilities to reproduce present climate mean and variability of temperature and precipitation. RCAO25 was the best performer for the needs of this study.

The different responses of Orig-PDSI and SC-PDSI to future climate and, as a result, the entirely different effects in soil moisture were explained and displayed (Sections 3.2.2 and 3.3). The former index was not sensitive to the much warmer future climate change scenarios projected by the RCM, as our results suggested, and therefore is not suggested for climate change impacts assessment studies. The latter index, on the other hand, has the potential to be used in climate change studies. However, its responses depended on the interval on which the climate characteristics and duration factors were computed. Two different possible scenarios were investigated in this study. In the first one, all calibrations were based on the historical period only. In the second one, the climate characteristics and duration factors were calibrated based on the baseline (historical) climate while the wet- and dry-spell coefficients were computed based on the historical + perturbed interval. In the first case mean future soil moisture changes in the range of −42% to + 133% were identified while in the second from −40% to + 116%. We are not able to say which scenario is the ‘right’ one. One more question that is raised, regarding evaluating climate change impacts with drought indices, is that all indices compute coefficients over the input period used each time. Thus, a drought index would take a different value for the same month for a different calibration interval because drought indices ‘absorb’ the whole baseline and evaluate the severity of a wet or dry month based on the specific period.

Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Material and methods
  5. 3. Results and discussion
  6. 4. Summary and conclusions
  7. Acknowledgements
  8. References

The data from the regional climate model runs were provided through the PRUDENCE data archive, funded by the EU through contract EVK2-CT2001-000132. The author would like to thank Steve Goddard for providing several variations of SC-PDSI and particularly for his expertise, constructive comments and suggestions. Helpful comments from the anonymous reviewers were greatly appreciated. The SWI product was very kindly delivered by the Institute of Photogrammetry and Remote Sensing (IPF), Vienna University of Technology.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Material and methods
  5. 3. Results and discussion
  6. 4. Summary and conclusions
  7. Acknowledgements
  8. References
  • Akinremi OO, McGinn SM, Barr AG. 1996. Evaluation of the Palmer Drought Index on the Canadian prairies. Journal of Climate 9: 897905.
  • Allen RG, Pereira LA, Raes D, Smith M 1998. Crop evapotranspiration: guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56. FAO: Rome, 293 pp.
  • Campling P, De Belder J 2003. CLIMSCAT quantification report, European Space Agency, Data User Programme 2001, http://dup.esrin.esa.it/projects/summary40.asp.
  • Ceballos A, Scipal K, Wagner W, Martínez-Fernández J. 2005. Validation of ERS scatterometer-derived soil moisture data in the central part of the Duero Basin, Spain. Journal of Hydrologic Processes 19: 15491566.
  • Christensen JH, Christensen OB. 2007. A summary of the PRUDENCE model projections of changes in European climate by the end of this century. Climatic Change 81(Suppl. 1): 730.
  • Dai A, Trenberth KE, Qian T. 2004. A global dataset of Palmer Drought Severity Index for 1870–2002: relationship with soil moisture and effects on surface warming. Journal of Hydrometeorology 5: 11171130.
  • Heim RR. 2002. A review of twentieth-century drought indices used in the United States. Bulletin of the American Meteorological Society 83: 11491165.
  • Hinkle D, Wiersma W, Jurs S. 1994. Applied statistics for the behavioral sciences (3rd ed.). Houghton Mifflin Company: Boston.
  • IPCC. 2001. In HoughtonTJ, DingY, GriggsDJ, NoguerM, van der LindenP, DaiX, MaskellK, JohnsonCI (eds.), Climate Change 2001. The Scientific Basis, Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press: Cambridge; 881.
  • Jacob D, Barring L, Christensen OB, Christensen JH, Castro M.de, Deque M, Giorgi F, Hagemann S, Hirschi M, Jones R, Kjellstrom E, Lenderink G, Rockel B, Sanchez E, Schar C, Seneviratne SI, Somot S, Ulden A.van, Hurk B.van den. 2007. An inter-comparison of regional climate models for Europe: Model performance in present day climate. Climatic Change 81(Suppl. 1): 3152.
  • Jones PD, Lister DH, Jaggard KW, Pidgeon JD. 2003. Future climate impact on the productivity of sugar beet (BETA VULGARIS L.) in Europe. Climatic Change 58: 93108.
  • Kattenberg A, Giorgi F, Grassl H, Meehl GA, Mitchell JFB, Stouffer RJ, Tokioka T, Weaver AJ, Wigley TML. 1996. Climate Models Projections of Future Climate. In Climate Change 1995. The Science of Climate Change, HoughtonJT, Meiro FilhoLG, CallanderBA, HarrisN, KattenbergA, MaskellK (eds). Cambridge University Press: Cambridge, pp 285358.
  • Laguardia G, Niemeyer S, Wagner W, Scipal K. 2006. On the comparison between the LISFLOOD modeled and the ERS/SCAT derived soil moisture estimates. Geophysical Research Abstracts 8: 08 385.
  • Lloyd-Hughes B, Saunders MA. 2002. A drought climatology for Europe. International Journal of Climatology 22: 15711592.
  • Loukas A, Vasiliades L, Dalezios NR. 2003. Intercomparison of meteorological drought indices for drought assessment and monitoring in Greece. Proceedings of the 8th International Conference on Environmental Science and Technology, Lemnos Island, Greece, Vol. B 484491.
  • Mavromatis T, Karacostas TS. 2005. An evaluation of drought indices for predicting wheat production in northern Greece. In: Proceedings of the 7th Panhellenic Conference of Meteorology, Climatology and Atmospheric Physics. Nicosia, Cyprus, Sep 28–30th, 2004, pp 981988.
  • Mavromatis T. 2007. Drought index evaluation for assessing future wheat production in Greece. International Journal of Climatology 27: 911924.
  • McKee TB, Doesken NJ, Kleist J. 1993. The relationship of drought frequency and duration to time scales. In: Proceedings of the 8th Conference on Applied Climatology. AMS Boston, MA, pp 179184.
  • McKee TB, Doesken NJ, Kleist J. 1995. Drought monitoring with multiple time scales. Proceedings of the 9th Conference on Applied Climatology. AMS: Boston MA; 233236.
  • Meehl GA, Washington WM, Santer BD, Collins WD, Arblaster JM, Hu A, Lawrence EB, Strand WG. 2006. Climate change projections for the twenty-first century and climate change commitment in the CCSM3. Journal of Climate 19: 25972616.
  • Mika J, Horvàth Sz, Makra L, Dunkel Z. 2005. The Palmer Drought Severity Index (PDSI) as an indicator of soil moisture. Physics and Chemistry of the Earth 30: 223230.
  • Ntale HK, Gan TY. 2003. Drought indices and their application to east Africa. International Journal of Climatology 23: 13351357.
  • Palmer WC. 1965. Meteorological drought. U.S. Weather Bureau Research Paper 45, 58 pp. (Available from NOAA Library and Information Services Division, Washington, DC 20852).
  • Pan Z, Arritt RW, Gutowski Jr WJ, Takle ES. 2001. Soil moisture in a regional climate model: simulation and projection. Geophysical Research Letters 28(15): 29472950.
  • Pellarin T, Calvet JC, Wagner W. 2008. Evaluation of ERS scatterometer soil moisture products over a half-degree region in southwestern France. Geophysical Research Letter 35: 15.
  • Priestley CHB, Taylor RJ. 1972. On the assessment of the surface heat flux and evapotranspiration using large-scale parameters. Monthly Weather Review 100: 8192.
  • Ricker WE. 1984. Computation and uses of central trend lines. Canadian Journal of Zoology 62: 18971905.
  • Robock A, Schlosser CA, Vinnikov KY, Speranskaya NA, Entin JK, Qiu S, Namkhai A. 1998. Evaluation of the AMIP soil moisture simulations. Global Planet Change 19: 181208.
  • Robock A, Vinnikov KY, Srinivasan G, Entin JK, Hollinger SE, Speranskaya NA, Liu S, Namkhai A. 2000. The global soil moisture data bank. Bulletin of American Meteorological Society 81: 12811299.
  • Scipal K, Wagner W. 2004. Global soil moisture data and its potential for climatological and meteorological applications. The 2004 Eumetsat Meteorological Satellite Conference, Prague, Chech Republic, May 31st–June 4th, published on CD-Rom.
  • Seneviratne SI, Pal JS, Eltahir AB, Schär C. 2002. Summer dryness in a warmer climate: A process study with a regional climate model. Climate Dynamics 20: 6985.
  • Srinivasan G, Robock A, Entin JK, Luo L, Vinnikov KY, Viterbo P, and participating AMIP Modeling Groups. 2000. Soil moisture simulations in revised AMIP models. Journal of Geophysical Research 105: 26 63526 644.
  • Szép IJ, Mika J, Dunkel Z. 2005. Palmer drought severity index as soil moisture indicator: physical interpretation, statistical behavior and relation to global climate. Physics and Chemistry of the Earth 30: 231243.
  • Thornthwaite CW. 1948. An approach towards a rational classification of climate. Geographical Review 38: 5594.
  • Wagner W, Lemoine G, Rott H. 1999. A Method for Estimating Soil Moisture from ERS Scatterometer and Soil Data. Remote Sensing of Environment 70: 191207.
  • Wagner W, Scipal K, Pathe C, Gerten D, Lucht Rudolf B. 2003. Evaluation of the agreement between the first global remotely sensed soil moisture data with model and precipitation data. Journal of Geophysical Research 108(D19): 4611.
  • Wagner W, Naeimi V, Scipal K, Richard deJu, Martinez-Fernandez J. 2007. Soil moisture from operational meteorological satellites. Hydrogeology Journal 15: 121131.
  • Wells N, Goddard S, Hayes MJ. 2004. A Self-Calibrating Palmer Drought Severity Index. Journal of Climate 17: 23352351.
  • Wilhite DA. 2000. Drought as a natural hazard: Concepts and definitions. In Drought a global assessment, WilhiteDA (ed.). Routledge Publishers: London; pp 318.
  • Yang J, Greenwood DJ, Rowell DL, Wadsworth GA, Burns IG. 2000. Statistical methods for evaluating a crop nitrogen simulation model, N_ABLE. Agricultural Systems 64: 3753.
  • Zhao D, Su B, Zhao M. 2006. Soil moisture retrieval from satellite images and its application to heavy rainfall simulation in Eastern China. Advances in Atmospheric Sciences 23: 299316.