Ground and satellite based assessment of meteorological droughts: The Coello river basin case study

https://doi.org/10.1016/j.jag.2017.06.005Get rights and content

Highlights

  • Monthly rainfall interpolation is improved applying elevation as ancillary data.

  • Satellite estimates provides inaccurate interpolations without a downscaling implementation.

  • Prediction errors of rainfall interpolation are propagated to drought classification.

Abstract

The spatial distribution of droughts is a key factor for designing water management policies at basin scale in arid and semi-arid regions. Ground hydro-meteorological data in neo-tropical areas are scarce; therefore, the merging of ground and satellite datasets is a promissory approach for improving our understanding of water distribution. This paper compares three monthly rainfall interpolation methods for drought evaluation. The ordinary kriging technique based on ground data, and cokriging with elevation as auxiliary variable were compared against cokriging using the Tropical Rainfall Measuring Mission (TRMM) Multi-Satellite Precipitation Analysis (TMPA). Twenty rain gauge stations and the 3B42V7 version of the TMPA research dataset were considered. Comparisons were made over the Coello river basin (Colombia) at 3″ spatial resolution covering a period of eight years (1998–2005). The best spatial rainfall estimation was found for cokriging using ground data and elevation. The spatial support of TMPA dataset is very coarse for a merged interpolation with ground data, this spatial scales discrepancy highlight the need to consider scaling rules in the interpolation process.

Introduction

Meteorological droughts are mainly determined by a continuous deficit of precipitation, temperature increasing and relative humidity diminishing (Mishra and Singh, 2010, Palmer, 1965). They constitute a significant risk for water security, therefore, drought characterization in terms of intensity, severity and spatial distribution is a relevant tool for water management and planning (Mishra and Singh, 2011, Valdés and Kohler, 2008, Garrote et al., 2006). Accordingly, the spatial pattern estimation of droughts is a key milestone for designing mitigation and response actions.

Accurate rainfall estimate is crucial for developing drought management plans, so effective drought plans must be supported on improved monitoring and prediction techniques (Wilhite et al., 2001). However, meteorological gauge networks are sparse in several regions all over the world and the distribution of rain gauge stations in steep mountain terrains is very often insufficient for areal rainfall estimation.

Thiessen polygons, the isohyetal method, inverse distance weighting and surface interpolators based on splines are useful interpolation techniques of rainfall, but are not always consistent in mountain terrains. The main limitation of these methods is that a high density network is required for the spatial estimation. Several researches have shown that geostatistical techniques provide better estimates of precipitation than traditional methods (Akhtari et al., 2009, Mair and Fares, 2010, Aalto et al., 2013, Delbari et al., 2013).

Because of the use of ancillary geographical information, co-kriging (CK) tends to improve interpolation results in regions with spatially sparse observations of rainfall. Some studies have investigated the utility of elevation as ancillary information (Goovaerts, 2000, Di Piazza et al., 2011, Diodato and Ceccarelli, 2005, Bárdossy and Pegram, 2013, de Amorim Borges and Franke, 2016, Adhikary et al., 2017). The accuracy of the spatial estimation of rainfall is improved in regions with a high correlation between elevation and rainfall (Goovaerts, 2000, Mair and Fares, 2010, Adhikary et al., 2017).

Satellite rainfall estimates constitute a promising source of supplementary information for improving the spatial estimation of rainfall (Serrat-Capdevila et al., 2014, Zhang and Jia, 2013, Zeng et al., 2012). However, the indirect nature of the satellite data generates a systematic bias against the rain gauge observations.

Several attempts have been achieved to merge rain gauge data with satellite estimates and provide robust estimates of rainfall fields. Rozante and Soares Moreira (2010) combined daily TRMM products with surface observations using a deterministic weighting function. Li and Shao (2010) applied a nonparametric kernel smoothing technique to merge TRMM and ground observations of rainfall. An approach based on the matching of the probability density function of rain gauge data and satellite estimates was applied by Shen et al. (2014), they showed the importance of the spatial and temporal scales in the merging process. In the context of near-real-time hydrological monitoring, Zhang and Tang (2015) adjusted the TRMM (3B42RTV7) cumulative distribution function (CDF) of daily rainfall with the CDFs from ground datasets across China.

Geostatistical interpolation algorithms have been broadly employed to merge rainfall data from satellite estimates and ground-based measurements (Gundogdu, 2015, Akhtari et al., 2009, Bajat et al., 2012, Bayat et al., 2015). The conditional merging technique (based on the ordinary kriging method) generated accurate precipitation estimates in Namhan River watershed (Korea) for spatially heterogeneous precipitation events (Jongjin et al., 2016). Verdin et al. (2016) applied ordinary kriging and k-nearest neighbor local polynomials to the residuals between rain gauge observations and satellite-based estimates, they found a substantial improvement in the estimated fields of rainfall for the two blending methods, particularly during extreme seasons.

Bayesian modeling is another approach employed for the spatiotemporal blending of satellite-derived estimates with rain gauge measurements (Lin and Wang, 2011, Jin et al., 2014). Verdin et al. (2015) proposed a Bayesian Kriging approach to blending rain gauge observations with satellite-based estimates, in their method, rainfall is assumed as a realization of a Gaussian process and a linear function of covariates (elevation and satellite-derived estimates), they modeled the residuals from the linear function by applying ordinary kriging and defined the model parameters as random variables via Markov Chain Monte Carlo. Despite the progress shown, the development of merging algorithms for high-resolution precipitation information from multiple independent sources is still a challenge (Nie et al., 2015).

The Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) provide precipitation estimates from multiple satellite-borne precipitation sensors. Recent research in the Andes indicates a poor performance of daily TMPA in comparison with gauge data, and a progressive improvement through temporal aggregation (Mantas et al., 2015, Ochoa et al., 2014). Mantas et al. (2015) found the poorest performance of monthly TMPA (3B42V7) in complex topography areas (mainly in highlands), with RMSE values greater than 75 (Fig. 7 in Mantas et al. work). Ochoa et al. (2014) showed a satisfactory performance of TMPA-V7 against ground rainfall data on coastal sub-catchments. Sahoo et al. (2015) identified major differences between TMPA precipitation estimates and gauge- based products in tropical regions where gauge networks are sparse.

According to the aforementioned, TMPA project datasets are able to provide ancillary information for improving drought characterization. However, TMPA estimates present contrasting performance in Andean regions and more research is needed for enhancing the spatial estimation of rainfall in Neotropical complex topographies for drought estimation. This work aims to evaluate the usefulness of TMPA product 3B43V7 and the Shuttle Radar Topography Mission elevation data as supplementary information for meteorological drought estimation in a typical Andean River Basin (Coello river basin in Colombia). To this end, ordinary kriging fields based on monthly rainfall gauging datasets was compared against cokriging estimates using TMPA 3B43V7, and cokriging estimates applying SRTM elevation data as ancillary information. The performance of the three interpolated monthly rainfall fields was evaluated through independent validation.

Section snippets

Study area

Coello river basin is located in the northwestern side of Tolima territory in Colombia and joins the Magdalena River Basin. It drains an area of 1899.31 km2 with an elevation range of 5300–280 m (Fig. 1). Elevation variation induces a complex land forms system in the basin, which makes difficult the spatial estimation of rainfall and drought characterization. Coello River supplies water for irrigation to about 25,600 ha in the Magdalena valley and the crop production area covers 15.80% of the

Contrasting rainfall observations: ground data and TMPA tiles

TMPA 3B43V7 estimates are highly correlated with the areal average of the rain gauge measurements in the flatlands of Coello River region (Tile 7). This relation decreases for the individual association of 3B43V7 product with each rain gauge station dataset located in the TMPA tile (Table 4). By contrast, the correlation tends to be weaker for both area-averaged and point rainfall observations in complex topographies (Headwaters). This is the case of tiles 5, 6 and 10 (Table 4).

In regard to the

Discussion

The Coello river basin is located in the inner Andes and is rain shadowed by high mountains ranges, this orographic effect is not represented by the TMPA estimates due to its coarse spatial resolution (0.25°). Following our results, the use of satellite estimates as ancillary data without the implementation of a downscaling procedure cannot improve rainfall interpolation. This highlight the need to incorporate downscaling methods in the spatial interpolation of rainfall (Tozer et al., 2012).

Conclusions

Monthly rainfall interpolation for the Coello river region is improved applying elevation as ancillary data. Particularly, for wrinkled terrain zones we found a high reduction of RMSE and MAE against ordinary kriging (OK) and ordinary cokriging with TMPA data as covariate (OCKT). We observed a dramatic reduction of the interpolation performance in locations with a sparse rainfall gauge network.

The development and application of downscaling operators to TMPA datasets are needed for improving its

Acknowledgements

TMPA datasets were provided by the NASA/Goddard Space Flight Center's Mesoscale Atmospheric Processes Laboratory and PPS, which develop and compute the TMPA as a contribution to TRMM. This research was partially funded by Universidad del Tolima (Grant no. 410112). Thanks to The Shuttle Radar Topography Mission NASA and Instituto de Hidrologıa Meteorologıa y Estudios Ambientales of Colombia (IDEAM), for providing digital elevation model and rainfall data of the Coello river basin region.

References (52)

  • J. Aalto et al.

    Spatial interpolation of monthly climate data for Finland: comparing the performance of kriging and generalized additive models?

    Theoret. Appl. Climatol.

    (2013)
  • S.K. Adhikary et al.

    Cokriging for enhanced spatial interpolation of rainfall in two Australian catchments

    Hydrol. Process.

    (2017)
  • R. Akhtari et al.

    Assessment of areal interpolation methods for spatial analysis of SPI and EDI drought indices?

    Int. J. Climatol.

    (2009)
  • B. Bajat et al.

    Mapping average annual precipitation in Serbia (1961–1990) by using regression kriging?

    Theoret. Appl. Climatol.

    (2012)
  • B. Bayat et al.

    Identification of long-term annual pattern of meteorological drought based on spatiotemporal methods: evaluation of different geostatistical approaches?

    Nat. Hazards

    (2015)
  • Y.A. Bayissa et al.

    Spatio-temporal assessment of meteorological drought under the influence of varying record length: the case of Upper Blue Nile Basin, Ethiopia

    Hydrol. Sci. J.

    (2015)
  • A. Bárdossy et al.

    Interpolation of precipitation under topographic influence at different time scales?

    Water Resour. Res.

    (2013)
  • P. de Amorim Borges et al.

    Comparison of spatial interpolation methods for the estimation of precipitation distribution in Distrito Federal, Brazil

    Theoret. Appl. Climatol.

    (2016)
  • M. Delbari et al.

    Spatial interpolation of monthly and annual rainfall in northeast of Iran?

    Meteorol. Atmos. Phys.

    (2013)
  • N. Diodato et al.

    Interpolation processes using multivariate geostatistics for mapping of climatological precipitation mean in the Sannio Mountains (southern Italy)?

    Earth Surf. Process. Landforms

    (2005)
  • L. Garrote et al.

    Linking drought indicators to policy actions in the Tagus basin drought management plan?

    Water Resour. Manag.

    (2006)
  • P. Goovaerts

    Geostatistics for Natural Resources Evaluation

    (1997)
  • I.B. Gundogdu

    Usage of multivariate geostatistics in interpolation processes for meteorological precipitation maps

    Theoret. Appl. Climatol.

    (2015)
  • G. Huffman et al.

    TRMM Version 7 3B42 and 3B43 Data Sets

    (2012)
  • G.J. Huffman et al.

    The TRMM Multi-Satellite Precipitation Analysis (TMPA)

    (2010)
  • A. Jarvis et al.

    Hole-Filled SRTM for the Globe Version 4, Available from the CGIAR-CSI SRTM 90m Database

    (2008)
  • Cited by (0)

    View full text