ERA-40 reanalysis hydrological applications in the characterization of regional drought

Authors


Abstract

[1] The use of a land surface model (TESSEL) for drought characterization is proposed. The method is tested with the 1958–2001 ECMWF ERA-40 reanalysis and its results are successfully compared with those obtained with other indices (SPI, PDSI) computed from observations in Iberia, a region highly vulnerable to drought. The proposed index, Normalized Soil Moisture (NSM), requires no local tuning, and, in principle, is readily computable from point observations, reanalysis, forecasts or climate scenarios. NSM from offline integrations has an enhanced consistency for drought analysis when compared with NSM computed directly from ERA-40 soil moisture. Analysis of the NSM fields indicates that while the spatial heterogeneity of precipitation is the main driver of the spatial variability of drought, the spatial distribution of soil characteristics modulates its temporal variability. This suggests the need for better soil maps at the global scale.

1. Introduction

[2] Drought is a recurring phenomenon that has affected civilization throughout history. It affects natural habitats, ecosystems, and many economic and social sectors. Drought characteristics and the wide range of sectors on which it impacts make its effects difficult to quantify. Metrics are needed for comparison of drought severity among different regions, as well as for comparing past and present drought events in the same place. Because of the complexity of its causes, no single index has been able to adequately capture the intensity and severity of drought and its potential impacts [Heim, 2002].

[3] Recent studies used soil moisture as an indicator of weather extremes, namely droughts [Andreadis et al., 2005; Lakshmi et al., 2004; Sheffield and Wood, 2007]. In this paper, soil moisture is derived using ERA-40 [Uppala et al., 2005] surface fluxes and near-surface meteorology to force the land surface model TESSEL [van den Hurk et al., 2000; Viterbo and Beljaars, 1995] for the period 1958–2001. Soil moisture output is used for regional drought characterization and to create a drought index – normalized soil moisture (NSM). A proof of concept is performed over the Iberian Peninsula, where the derived drought index is compared to other indices in use for agricultural and hydrological purposes, such as the standardized precipitation index (SPI) [McKee et al., 1993] and the Palmer drought severity index (PDSI) [Palmer, 1965]. Soil moisture output is then used to study spatial and temporal patterns of droughts and their sensitivity to soil physical characteristics.

2. Experiment Design and Methods

2.1. Model Setup

[4] Direct use of ERA-40 soil moisture analysis for regional drought characterization is limited by the systematic attenuation of the seasonal cycle and damping of inter-annual variability [Ferranti and Viterbo, 2006; Viterbo and Betts, 1999]. This can be primarily attributed to the soil moisture assimilation scheme used in ERA-40, which corrects soil moisture in the first three layers based on the 6-h atmospheric analysis increments of specific humidity and temperature at the lowest model level [Douville et al., 2000]. The use of the ERA-40 land surface model in offline mode prevents the soil moisture problems described above and opens the possibility to incorporate other diagnostics, such as surface water and energy fluxes.

[5] The ERA-40 land surface model TESSEL (Tiled ECMWF Scheme for Surface Exchanges over Land) represents vertical transfers of water and energy using four vertical layers (0–7, 7–28, 28–100 and 100–289 cm) to represent soil temperature and water. The model evaluates the soil response to the atmospheric forcing, estimating the surface water and energy fluxes and the temporal evolution of soil temperature and moisture content. At the interface between the surface and the atmosphere, each grid box is divided into fractions (tiles), with up to 6 fractions over land (bare ground, low and high vegetation, intercepted water, shaded and exposed snow). Each fraction has its own properties defining separate heat and water fluxes used in the energy balance equation solved for the tile skin temperature.

[6] ERA-40 surface fluxes and near surface meteorology were used to force TESSEL. Near surface meteorology includes air temperature, specific humidity and wind speed from the lowest model level, and surface pressure, provided 3-hourly with a combination of the analysis every 6 h and the corresponding 3 h forecasts. Surface fluxes include precipitation, snowfall, and downwelling solar and thermal radiation. To avoid the initial spin-up in ERA-40 precipitation [Uppala et al., 2005], the 3-hourly forcing surface fluxes correspond to the 12–24 h forecast interval from initial conditions at 00 and 12 UTC. The N80 reduced Gaussian grid and the land surface characteristics (including a globally uniform soil texture and soil depth) were the same as used during the reanalysis.

[7] Total depth soil moisture was aggregated in monthly values and normalized by its monthly mean and standard deviation, defining the NSM. Considering θi,j total depth soil moisture, in a specific grid point, for month i and year j, with i = 1, …, 12 and j = 1, …, N. NSM in month i and year j is defined as

equation image

where equation imagei and si are the mean monthly values and standard deviation, respectively. In much the same way as the SPI index, NSM is a normalized index, using the standard deviation as a measure of interannual variability. A representative climatology must be used for computing the standard deviations of soil moisture used in the normalization. The period to calculate standard deviations should be long enough to have an adequate sample of low and high precipitation years; In this paper we used the entire ERA-40 record (N = 44) to compute the standard deviations. In the NSM computations, the offline soil moisture could be normalized by the ERA-40 soil moisture data. Such approach would introduce some biases due to the systematic attenuation of seasonal cycle and reduced inter-annual variability of ERA-40 soil moisture [Ferranti and Viterbo, 2006]. The spatial distribution of soil characteristics (texture and total depth) is an important issue in representing surface processes [Betts et al., 2001]. Nevertheless, it can be difficult to prove unambiguously the benefits of a sophisticated model (with geographically varying soil texture and depth) when compared with a simple model (uniform soil texture and depth). The impact on results of such a simplified approach was checked by performing an additional simulation with variable total soil depth (hereafter new depth – ND). Fields of total soil depth and texture were taken from Food and Agriculture Organization data. However, it was found that, at the ERA-40 resolution, only the depth maps were significantly affected; hence, soil texture was kept unchanged for the ND simulations. Soil depths decreased from a constant 2.89 m depth (TESSEL value) to depths ranging from 0.805 m to 1.525 m (see Figure 3c).

2.2. Drought Indices and Time Scales

[8] The SPI is one of the simplest drought indicators and is not affected by geographical or topographical differences. SPI is computed by fitting a probability density function to a frequency distribution of precipitation summed over the time scale of interest. This is performed separately for each month and for each location in space. The PDSI is based upon a set of empirical relationships derived by Palmer [1965] to express regional moisture supply standardized by local climatological norms. The method begins with a simplified water balance (usually on a monthly basis) using historic records of precipitation and temperature. Soil moisture storage is handled by dividing the soil into two layers and potential evapotranspiration generally is computed using Thornthwaite's method. PDSI is the most prominent index of meteorological drought used in the United States [Heim, 2002], despite its shortcomings in spatial and temporal comparability [Guttman, 1998].

[9] The drought index proposed here (NSM) has a clear physical basis, since it is based on the energy and soil water balances computed by the TESSEL model. NSM is based on a physically casted water balance, unlike the highly parameterized formulation of PDSI. The characteristic time scales of soil moisture, dependent on the atmospheric forcing and soil hydraulic properties, overcome the empirical accumulation time scales used in SPI. The major drawback of NSM is related to the need of continuous forcing, while SPI and PDSI only need monthly data.

[10] The SPI, for time scales ranging from 1 to 12 months, was calculated from observed precipitation over Iberia. The observations included 27 stations over Portugal and data on a regular grid 0.5° × 0.5° over Spain. Data were provided by the national Portuguese and Spanish weather services. The observations were spatially aggregated to the N80 grid, for comparison purposes and for SPI calculations. PDSI data from Dai et al. [2004] were used in this study.

3. Applications Over Iberia

3.1. Validation of ERA-40 Precipitation and Offline Methodology

[11] Since precipitation is the primary factor controlling the establishment and persistence of drought conditions [Lloyd-Hughes and Saunders, 2002], we assessed the quality of ERA-40 precipitation against observations. The general evaluation of ERA-40 precipitation is described by Uppala et al. [2005] and references therein, but the performance over the Iberian Peninsula is not documented.

[12] Precipitation over Iberia has a marked spatial and temporal distribution, with a strong seasonal cycle throughout all regions (Figure 1a) and abundant precipitation in the Northwest. The comparison between ERA-40 and observations shows a systematic underestimation, which is more evident near the coast, and a reduced mean annual cycle. Despite a dry bias in the rainy season, explained variance of ERA-40 monthly precipitation is of the order of 90% in Western Iberia and reduces to about 60% in Eastern Iberia. The correlations between SPI calculated from observations and ERA-40 for different time scales (not shown) range from 0.4 to 0.9 (all the correlations shown in this paper are significantly different from zero at the 95% confidence level), with the lower values in Eastern Iberia for longer time scales. These values result from a good representation of the winter circulations over the North Atlantic in ERA-40, combined with some problems in representing small scale events over Eastern Iberia.

Figure 1.

(a) Mean annual cycle of observed (solid line) and ERA-40 (solid-star line) precipitation over Iberia (spatially integrated). (b) Monthly averaged spatial correlation between observed SPI (at the time scale that maximizes the correlation with NSM) and NSM calculated from the control simulation (solid line); ND simulation (solid-cross line), and ERA-40 soil moisture (solid-star line).

[13] In order to compare ERA-40 and offline soil moisture, NSM calculated from ERA-40 and offline simulations (both original and ND) was correlated with observed SPI at several time scales. Only the SPI time scale with maximum correlation with NSM was retained, since soil moisture acts as an integrator of precipitation with different time scales for each point, determined by the forcing and soil physical characteristics. Since soil moisture increments in ERA-40 have a seasonal cycle, correlations were taken for each month. Figure 1b shows the monthly maximum averaged spatial correlation between SPI and NSM calculated from both offline simulations and ERA-40 soil moisture. Correlations were calculated for SPI time scales ranging from 3 to 24 months retaining only the SPI time scale with maximum correlation with NSM. In all three cases, there is a seasonal cycle that can be associated with the strong seasonal cycle of precipitation. Soil moisture from both offline simulations (original and ND) have substantially higher correlation with observed SPI than the ERA-40 soil moisture, especially during the dry-down months. This is a strong validation of NSM and NSM ND since both are coherent with the observed precipitation. ERA-40 soil moisture variability is not fully representative of the atmospheric circulation in the preceding months, especially during summer, reducing its applicability to drought characterization. The proposed offline methodology overcomes part of known problems in ERA-40 soil moisture, and makes it suitable for drought applications.

[14] The maximum correlation of the observed SPI with NSM from either offline simulations (Figure 1b) are very similar and do not clearly show an advantage of one of the formulations over the other. While the ND simulation is useful to evaluate the robustness of the computed NSM to changes in a soil model, the impact results presented might be affected by several factors: (a) The spatial scales resolved at the ERA-40 dataset; (b) the realism of FAO dataset on soil characteristics; and (c) the realism of TESSEL hydrology.

3.2. Regional Drought Characterization

[15] NSM was compared with SPI from observations, at several time scales, and PDSI for the ERA-40 period over the Iberian Peninsula. Correlations between NSM and SPI increase with SPI time scale and range from 0.4 to 0.8 for a 12 month SPI, with higher correlations in Western Iberia (figure not shown). This dispersion can be primarily attributed to ERA-40 precipitation deficiencies described before. The correlations between NSM and PDSI range from 0.6 to 0.8.

[16] Temporal series of the three indices for South Portugal are represented in Figure 2. The indices are in agreement for the 44-year period, and the 5 major drought periods (1976, 1981, 1983, 1992 and 1995) agree with the historical record for South Portugal. The high degree of coherency between the three indices for the 44-year period studied is an indication that NSM can be used as a robust drought indicator. NSM calculated directly from ERA-40 soil moisture is also represented in Figure 2 (NSM ERA). NSM ERA has a limited coherency with the other indexes, identifying the major drought periods, but with some discrepancies especially in the beginning of the period (1961, 1971, and 1972). NSM from the offline integrations is affected by the quality of ERA-40 precipitation, while NSM calculated directly from ERA-40 soil moisture is further affected by the soil moisture increments that vary both in time and space.

Figure 2.

Time series of drought indices for southern Portugal. NSM calculated from offline integration and directly from ERA-40 soil moisture; SPI for 12 months time scale (calculated from observed precipitation) and PDSI [from Dai et al., 2004]. Enhanced zones indicate values below −1 for NSM and SPI-12 and below −4 for PDSI.

[17] Principal component (PC) analysis of NSM was used to analyze drought regional patterns. In this analysis the empirical orthogonal functions (EOFs) were rotated using the Varimax method [Kaiser, 1958], an approach common in climatological studies [Vicente-Serrano, 2006]. After rotation, the first four EOFs explained 86.2% of the total NSM variance. The spatial classification was based on the general patterns returned by the EOFs, aggregating the grid points through a maximum factor loadings criterion. Figure 3a represents the four zones over Iberia where the first four EOFs have maximum loadings. The analysis enables the consideration of each zone separately using the corresponding PCs, and has the advantage that each PC is uncorrelated with the others. The high correlation (0.96) between PC1 (figure not shown) and NSM for South Portugal (see Figure 2, top) suggests a strong physical basis of the PCs obtained.

Figure 3.

(a) Division of Iberia in four zones according to the rotated EOFs of NSM. (b) Distribution of soil depths used in ND simulation. Soil moisture decay time scales calculated from (c) the offline reference simulation and (d) the offline ND simulation. Note the different color scales.

[18] The temporal scale of autocorrelation for NSM represents the decay time scale of an anomaly [Vinnikov et al., 1996] and was used to identify zones susceptible to long periods of drought. Decay time scales of NSM are represented in Figure 3b. Values vary between one and eighteen months, with a clear spatial distribution. In Northwest Iberia, where precipitation is abundant throughout the year, decay times are reduced; corresponding to soil moisture anomalies quickly attenuated, and shorter drought periods. In contrast, Southeast Iberia is characterized by longer decay time scales. These results emphasize the importance of precipitation in drought modeling, since scattered and reduced precipitation leads to slower recovery from a drought event. The present analysis allows the regionalization of drought events, and gives an indicator of the vulnerability to long drought spells.

3.3. Sensitivity to Soil Depth

[19] The sensitivity of the results shown in the previous section to the soil characteristics was evaluated by analyzing the ND simulation. Figure 3c shows the spatial distribution of soil depths over Iberia and Figure 3d shows the decay time scales of NSM in the ND simulation. Decay time scales are reduced by approximately a factor of three when compared with those of NSM in the control simulation. These results are explained by a reduction of the mean soil depth, which lead to the reduction of the available water and consequently of soil memory. Nevertheless, little change is seen in the spatial patterns, despite an increase of small scale details added by the spatial distribution of the soil depths imposed. Spatial patterns returned by the principal component analysis (not shown) remain unchanged, suggesting that, over the Iberian Peninsula, precipitation is the main driver of the spatial variability of drought. These results indicate that realistic soil depths, when modeling soil moisture, have a limited impact in regional drought patterns, but are important in the temporal characterization of droughts.

4. Conclusions

[20] Since precipitation is the major atmospheric forcing for drought, a validation of ERA-40 precipitation fields against observations was conducted over Iberia. Despite a dry bias in the rainy season, the explained variance of ERA-40 monthly precipitation is of the order of 90% in Western Iberia and reduces to about 60% in Eastern Iberia.

[21] The proposed drought index, based on offline modeled soil moisture, was used as a drought index and compared with the SPI, at different time scales, and with the PDSI. The high degree of coherency between the three indices for the 44-year period studied is an indication that NSM can be used to characterize regional drought events. It is worth noticing that, unlike other drought studies presented recently [e.g., Sheffield and Wood, 2007], the NSM values presented here are obtained without correcting the ERA-40 precipitation for known biases. Besides the identification of major drought periods and their intensity, soil moisture was applied to the study of spatial and temporal patterns of droughts over Iberia. Soil moisture decay time scales were used to identify areas susceptible to long drought spells (due to scarcer annual precipitation) where NSM is related to SPI at longer time scales. Spatial patterns of drought over Iberia, established from principal component analysis, are mainly determined by the spatial distribution of precipitation.

[22] The simulation with variable soil depth showed that, for drought characterization purposes, a correct representation of the soil characteristics is important in the characterization of the temporal variability, but has a reduced impact in drought regional patterns. Soil moisture from offline simulations of TESSEL represents an improvement on the ERA-40 soil moisture values, since the latter have a strong signature of the analysis increments.

[23] The approach to regional drought characterization presented here can be applied to other areas with realistic time variability of ERA-40 precipitation, without the need of specific tuning. Forcing from a future near-real time continuation of ERA-40 (or the new ECMWF reanalysis product ERA-INTERIM) can be used as a tool in the monitoring of drought situations. The proposed methodologies (NSM, decay time scales and regionalization from principal component analysis) can be also applied directly to the output of climate models for drought analysis in climate studies. Early identification of an impending drought could be used to support drought mitigation actions. A better understanding and characterization of past droughts is a key factor in assessing future impacts of climate change.

Acknowledgments

[24] It is a pleasure to acknowledge Bartolomé Orfila for making available the Spanish precipitation dataset, Vanda Cabrinha for the Portuguese precipitation data and useful discussion on PDSI versions and Fátima Espíritu Santo for discussion. We acknowledge two anonymous reviewers for their valuable suggestions. This research was funded by FCT under Grant VAST(POCTI/CTA/46573/2002), co-financed by the EU under Program FEDER.

Ancillary