Fighting flooding in Mozambique

Disaster risk management: “Flood frequency curves” aim to forecast extreme flooding in the Limpopo River basin, which, while vital to the region’s economy, is a threat during the wet season.

By Daniel Maposa, James J. Cochran and ‘Maseka Lesaoana

Aerial view of the Limpopo River as it winds its way through southern Mozambique, where it often crests its banks and sends floodwaters rushing through towns and farmland. Source: U.S. military

Aerial view of the Limpopo River as it winds its way through southern Mozambique, where it often crests its banks and sends floodwaters rushing through towns and farmland. Source: U.S. military

On a worldwide scale, floods and droughts account for the vast majority (approximately 90 percent) of all people affected by natural disasters [13]. This is certainly true in the Limpopo River basin (LRB) in southeast Africa, particularly Mozambique, where several efforts in disaster risk management – including generating accurate short- and long-term forecasts of extreme floods via operations research methodology – are underway in the lower LRB region to mitigate the devastating impact of persistent flooding.

Geographic Background

Mozambique is located on the Indian Ocean coast of Southern Africa. Agricultural and other economic activities in the LRB form the backbone of the Mozambican economy. The LRB catchment area covers more than 412,000 square kilometers and is drained by the Limpopo River and its tributaries, including the Olifants and Changane rivers, which drain the largest areas.

The LRB (Figure 1) is home to about 14 million people in Botswana, Mozambique, South Africa and Zimbabwe. Urban centers throughout the region, where the river serves industries, power stations and municipalities, are major users of the basin’s water resources. In rural areas, the basin’s water resources are primarily used for domestic purposes, irrigation and watering livestock.

The Limpopo River (Figure 1) flows approximately 1,750 kilometers (main length) along the border of South Africa and Botswana, enters Mozambique and empties into the Indian Ocean approximately 60 kilometers downstream of the town of Xai-Xai. The LRB significantly narrows in the coastal area where the river’s course meanders for nearly 70 kilometers through its lower valley, from Xai-Xai to the sea, and forms a circular alluvial valley inland of about 15 kilometers in diameter as it enters the sea.

Among African rivers that drain into the Indian Ocean, the Limpopo is second in length to the great Zambezi River. The floodplains of the Limpopo River are fertile and heavily populated.

The LRB comprises several tributaries and sub-basins. Its longest tributary, the Crocodile River, drains a catchment area of 29,000 square kilometers. The Olifants and its tributaries form the largest sub-basin of the LRB, covering a catchment area of 79,000 square kilometers while bringing the greatest amount of water to the Limpopo River. The other large sub-basin is the Changane River, which is characterized by a very low run-off, exists entirely within Mozambique and covers a catchment area of 43,000 square kilometers.

The LRB is divided into three main geographic sections: the Upper Limpopo, the Middle Limpopo and the Lower Limpopo.

The region suffers from extreme climate conditions, punctuated with frequent droughts and floods. Annual rainfall in the LRB is highly seasonal; more than 95 percent is received between October and April. The Limpopo River is characterized by low to negligible flows during the dry season, resulting in severe droughts, while severe flooding is common during the wet season. [3, 4, 9, 12, 14, 15]

River Benefits and Drawbacks

Figure 1: Limpopo River basin location and riparian states. Source: http://www.limpoporak.com/ (adapted from WMO (2012)

Figure 1: Limpopo River basin location and riparian states. Source: http://www.limpoporak.com/ (adapted from WMO (2012)

Among its many benefits, the Limpopo River supplies water to the largest irrigation system in Mozambique via the Massingir Dam, which helps improve the standard of living for both the rural and urban communities throughout the region. The LRB and the associated irrigation system not only supplies much needed water for communities and industry, it helps alleviate poverty in the region and thus contributes to one of the millennium development goals (MDGs) of poverty reduction in developing countries.

Since the end of the Mozambican Civil War in 1992 (after nearly two decades of conflict), the lower LRB has attracted a wide range of researchers from diversified fields who have studied the region’s climate and river system in an effort to better predict and thus mitigate the damaging impact of the drought-flood cycle. These researchers include non-governmental organizations, universities and independents. In addition, donors such as the World Bank, the British Department for International Development (DFID), United States Agency for International Development (USAID), Germany’s Deutsche Gesellschaft fur Internationale Zusammenarbeit (GIZ) and African Development Bank provide financial and technical assistance in disaster risk management in the basin. These organizations work in collaboration with the Limpopo Watercourse Commission (LIMCOM) and the government of Mozambique.

The low-lying nature of the lower LRB region across the coastal floodplain makes it susceptible to flooding during periods of high river flows. The high incidence of flooding is mainly attributed to tropical cyclones that form in the Indian Ocean, as well as the fact that Mozambique is a downstream country through which all floods from the neighboring countries pass. According to historical records on natural disasters, over the 52-year period from 1956 through 2008, Mozambique experienced 10 droughts, 20 floods, 13 tropical cyclones, 18 epidemics and one earthquake.

Residents on the roof of a house surrounded by floodwaters in Chokwe district in Mozambique in 2013.  Source: AFP/Ussene Mamudo

Residents on the roof of a house surrounded by floodwaters in Chokwe district in Mozambique in 2013. Source: AFP/Ussene Mamudo

In recent years, floods were experienced in the lower LRB in 2010, 2012 and 2013. The most catastrophic and expensive of recent natural disasters were the floods of 2000; in February and March of that year, the Limpopo River reached levels never previously recorded. It swelled from less than 100 meters wide to between 10 kilometers and 20 kilometers wide for more than a 100-kilometer stretch, and it inundated more than 1,400 square kilometers of farmland. This disaster, which was primarily a result of heavy rainfall brought by cyclones, completely flooded the town of Chokwe and parts of Xai-Xai, killed more than 700 people and caused an estimated $500 million in damage in the LRB of Mozambique alone.

The 2014 International Disaster and Risk Conference (IDRC) held in Davos, Switzerland, advocated for collaborative efforts in disaster risk reduction. In line with this view, Irina Bokova, director-general of UNESCO, stated that: “Every year, more than 200 million people are affected by natural hazards, and the risks are increasing – especially in developing countries, where a single major disaster can set back healthy economic growth for years. As a result, approximately one trillion dollars have been lost in the last decade alone.” [1, 3, 4, 5, 6, 12, 15, 16]

Flood Research

As part of the early efforts to reduce disaster risk in the LRB, the World Bank project installed 19 real-time rain gauge stations within the lower LRB in Mozambique. These rain gauge stations collect data and information that are fundamental for establishing, operating, and maintaining an effective and efficient river forecasting system or flood forecasting and early warning system. Some flood forecasting and early warning models were also applied in Mozambique using “Flood Watch” and its upgrades and geo-spatial streamflow forecasting modeling, but less than 14 percent of the installed real-time stations are still working, and the flood forecasting and early warning system is struggling to be operational, mainly due to inadequate project design, technical issues and maintenance problems.

Figure 2: Flood frequency curve of posterior distribution with 95 percent Bayesian credible intervals (dashed lines) at Chokwe hydrometric station.

Figure 2: Flood frequency curve of posterior distribution with 95 percent Bayesian credible intervals (dashed lines) at Chokwe hydrometric station.

Literature on long-term forecasts based on hydrological studies in the basin is scarce. However, long-term forecasts based on estimation of the return period (or expected number of years between occurrences) of an extreme flood event (e.g., flood height, river discharge or precipitation) have played and continue to play a major role in the planning, design and management of hydraulic structures such as bridges, dams, spillways and other engineering structures in the LRB. In one instance, Mondlane et al. (2013) performed a comparative analysis of extreme flood frequency distributions based on 20 years of rainfall data recorded at the Xai-Xai precipitation station. The Gumbel distribution appeared to approximate the multi-modal distribution of the collected data better than other distributions, while the two-parameter Weibull distribution was a better model for the simulated data, which was negatively skewed and unimodal. The work by Mondlane et al. met with limited success, mainly due to its small sample size, to identify the prevailing or suitable distribution for the LRB at Xai-Xai in the lower LRB. Had it been successful, the prevailing distribution, once identified, would have been used to model long-term forecasts for the LRB, which are instrumental in effective management of river flows. In other studies in the basin, Maposa et al. (2014) compared 10 candidate distributions using hydrometric data (flood heights) collected at the Chokwe and Sicacate hydrometric stations and found the generalized extreme value distribution to be consistent at the two stations (sites), with Gumbel, three-parameter gamma and lognormal distributions providing alternative models for the sites.

Figure 3: Flood frequency curve of posterior distribution with 95 percent Bayesian credible intervals (dashed lines) at Sicacate hydrometric station.

Figure 3: Flood frequency curve of posterior distribution with 95 percent Bayesian credible intervals (dashed lines) at Sicacate hydrometric station.

The work by Maposa et al. complements and advances the work by Mondlane et al. (2013) in identifying the prevailing flood frequency distribution for the lower LRB. The prevailing distribution(s) was identified using large sample sizes of hydrometric data from multiple sites as opposed to a small sample size of precipitation data from a single site used in Mondlane et al. The flood frequency distribution identified is used to model the long-term distribution of floods in the basin, which is fundamental in river flow management.

The authors of this article are now attempting to improve on these results by producing flood frequency curves (which show flood levels and their corresponding return periods; see Figures 2 and 3) at two sites, Chokwe and Sicacate, in the lower LRB, based on Monte Carlo Markov chain Bayesian estimates of the generalized extreme value distribution. [7, 8, 10, 15]

Materials and Methodology

The authors obtained data from the Mozambique National Directorate of Water, the water management authority in Mozambique operating under the Ministry of Public Works. The hydrometric data was recorded at the Chokwe (1951-2010) and Sicacate (1952-2010) hydrometric stations. In its original form, the data was daily flood heights recorded in meters; from these data we have selected the highest peak flood height in each hydrological year, resulting in a series of annual maximum daily flood heights. This leads to the at-site approach known as “block maxima,” an approach that generally yields good estimates when the data records are sufficiently large (as is the case in this study). Since the study involves tail estimation based on data sets with little information available, the authors used the Bayes approach to capture and take into account all the available information including additional information through prior elicitation.

The parameters of the generalized extreme value distribution were estimated using Bayesian estimates and the trivariate normal prior (or conjugate prior). Maximum likelihood estimates were used for the initial values of the trivariate normal priors. The posterior predictive distribution was used to plot the flood frequency curve, which is then used to estimate the desired return periods of given return levels.

Descriptive statistics reveal that the distribution of annual daily maximum flood height at Chokwe is right-skewed whereas that of downstream Sicacate is left-skewed. This indicates that the majority of the annual daily maximum flood heights at Chokwe upstream are generally low whereas those of Sicacate are generally high.

Figures 2 and 3 present the flood frequency curves for Chokwe and Sicacate, respectively. The results of flood frequency curves show that the 13-meter flood height in 2000 that occurred at the two sites in the lower LRB was more disastrous at Chokwe, where it had a return period in excess of 200 years, and less disastrous at Sicacate, where it was at approximately the 100-year flood level. [2, 7]

Concluding Remarks & Notes

The broad aim of the authors’ study is to develop flood frequency curves for the lower LRB at Chokwe upstream and Sicacate downstream. The skewness of the annual daily maxima data series in this study reveals that water levels are usually high at Sicacate and usually low at Chokwe, hence the 13-meter flood height of the 2000 flood was a rare event at Chokwe and not very rare at Sicacate. The estimates of the return levels and their corresponding return periods from the flood frequency curves appear to be consistent with previous studies in the upper LRB in South Africa. For instance, Smithers et al. (2001) studied the February 2000 floods in the Sabie River catchment upstream of the South Africa/Mozambique border, which is in the upper LRB, using a regional approach and found the return periods to be in excess of 200 years at most sites and approximately 100 years at a few sites in the Sabie River catchment region, which is part of the large LRB.

The authors’ future work will look into application of Bayesian analysis and MCMC methods to the nonstationary at-site flood frequency analysis of extremes, and application of the Bayesian analysis and MCMC methods to the estimation of a regional trend in annual maxima and statistical modeling of spatial extremes.

Finally, it should be noted that the flood frequency curves project described in this article is part of Ph.D. research work by lead author Daniel Maposa (supervised by the co-authors) and registered at the University of Limpopo, a public university in South Africa. All three authors participated in this project in their capacity as independent researchers, driven by their shared interest in reducing the associated risk and mitigating the deleterious impact of these floods on humans and property. The authors hope that the outcomes and recommendations of this ongoing project will be communicated to the Mozambique National Directorate of Water in the Ministry of Public Works and Housing, the custodian of water resources information such as hydrometric and meteorological data in the country.

Daniel Maposa is a graduate student and lecturer in the Department of Statistics and Operations Research, School of Mathematical and Computer Sciences at the University of Limpopo, South Africa. The project described in this article is part of Maposa’s Ph.D. research under the supervision of the co-authors.

James J. Cochran (jcochran@cba.ua.edu) is a professor of applied statistics and the Rogers-Spivey Faculty Fellow in the Department of Information Systems, Statistics and Management Science of the University of Alabama’s Culverhouse College of Commerce and Business Administration. In addition, Cochran is a Fellow of the American Statistical Society and a co-founder of Statistics Without Borders.

‘Maseka Lesaoana is an associate professor in the Department of Statistics and Operations Research, School of Mathematical and Computer Sciences at the University of Limpopo in Limpopo Province, South Africa.

References

  1. Arya, A.S., Boen, T. and Ishiyama, Y., 2014, “Guidelines for earthquake resistant non-engineered construction,” UNESCO, Paris.
  2. Beirlant, J., Goegebeur, Y., Segers, J., Waal, D.D. and Ferro, C., 2004, “Statistics of Extremes: Theory and Applications,” John Wiley & Sons Ltd, West Sussex.
  3. Chilundo, M., Kelderman, P. and ‘Keeffe, J.H.O, 2008, “Design of a water quality monitoring network for the Limpopo River basin in Mozambique,” Physics and Chemistry of the Earth, Vol. 33(8-13), pp. 655-665.
  4. Hakala, A. and Pekonen, L., 2008, “The impact of the irrigation system and agricultural production on water on water quality in Chokwe Irrigation Scheme,” thesis, Savonia University, Finland, in collaboration with Universidade Eduardo Mondlane, Mozambique.
  5. IDRC Davos, 2014, “Integrative Risk Management-The role of science, technology & practice,” 5th IDRC Davos 2014, Programme & Short Abstracts, August 2014, Davos, Switzerland, available at: The IDRC Davos 2014 Conference Proceedings (http://idrc.info/outcomes/conference-proceedings/) (last accessed Jan. 17, 2015).
  6. Jackson, J., 2013, “Mozambique floods spur roof births, ruin and diarrhoea,”Agence France-Presse (AFP), http://reliefweb.int/report/mozambique/mozambique-floods-spur-roof-births-ruin-and-diarrhoea (last accessed Jan. 19, 2015).
  7. Maposa, D., Cochran, J.J. and Lesaoana, M., 2014, “Investigating the goodness-of-fit of ten candidate distributions and estimating high quantiles of extreme floods in the lower Limpopo River basin, Mozambique,” Journal of Statistics and Management Systems, Vol. 17, No. 3, pp. 265-283, doi:10.1080/09720510.2014.927602.
  8. Maurya, N.V., Arora, K.D., Maurya, K.A. and Gautam, A.R., 2013, “Exact modelling and analytical study of annual maximum rainfall with Gumbel and Frechet distributions using parametric estimation techniques,” World of Sciences Journal, Vol. 1, No. 2, pp. 11-26 (www.engineerspress.com) (last accessed Jan. 18, 2015).
  9. Mohamed, E.A., 2014, “Comparing Africa’s shared river basins – The Limpopo, Orange, Juba and Shabelle basins,” Universal Journal of Geoscience, Vol. 2, No. 7, pp. 200-211.
  10. Mondlane, A.V., Hansson, K. and Popov, O., 2013, “Analysis of mathematical models and their application to extreme events,” World Academy of Science, Engineering and Technology; International Journal of Mathematical, Computational Science and Engineering, Vol. 7, No. 4, pp. 378-387.
  11. Mozambique Floods 2013, consolidated early recovery strategy, Humanitarian Country Team, Maputo, April 25, 2013.
  12. Musiya, T., 2013, “UN: Mozambique floods displace 150k, leave 38 dead,” Associated Press CHOKWE, Mozambique (AP), available at: http://article.wn.com/view/2013/01/29/UN_Mozambique_floods_displace_150k_leave_38_dead_0/ (Last accessed Jan. 18, 2015).
  13. Smakhtin, V., 2014, “Managing floods and droughts through innovative water storage solutions,” 5th IDRC Davos 2014, available at: The IDRC Davos 2014 Conference Proceedings, http://idrc.info/outcomes/conference-proceedings/ (last accessed Jan.18, 2015).
  14. South Africa Limpopo, 2013, The Limpopo River, 2013 Limpopo Tourism and Parks Board, http://www.golimpopo.com/fact-details_limpopo-river_15.html (last accessed Jan. 18, 2015).
  15. WMO, 2012, “Limpopo River Basin – A proposal to improve the flood forecasting and early warning system,” World Meteorological Organization, with the support of Limpopo Water Course Secretariat and the riparian states of Botswana, Mozambique, South Africa and Zimbabwe.
  16. WMO, 2013, “Global Climate 2001-2010: A decade of climate extremes,” summary report, WMO no. 119, World Meteorological Organization, Geneva, Switzerland.