1. Identification of structural breaks in hydrological maxima time series in Paraguay River, Pantanal Region, Brazil Marcus Suassuna Santos 1,2 marcus.santos@cprm.gov.br Carlos Henrique Ribeiro Lima 3 chrlima@unb.br (1)PhD Student, University of Brasilia – UnB (2)Brazilian Geological Survey – CPRM (3)Professor, University of Brasilia – UnB Universidade de Brasília

2. Outline 1.Motivation and objectives 2.Methods to identify single and multiple step changes in the mean 3.Results: a.Step changes in annual maximum water levels b.Step changes in annual precipitation c.Step changes and Long Memory 4.Conclusions 5.Future Work

3. Critical Hydrological Events Warning System SACE PANTANAL http://www.cprm.gov.br/sace/

4. Critical Hydrological Events Warning System SACE PANTANAL 1974: Major impact: cattle and agriculture Surprise effect Past decades: long cycle of droughts - adaptation

5. Structural breaks Objective: identify structural breaks (step changes) in the mean of annual maximum series of water levels and annual rainfall in the Pantanal region, Brazil. Definition: abrupt (step) change in time series statistics - mean, variance, serial dependence, etc. Motivation: Search for causes Understand hydrological time series Model hydrological time series Improve extreme events management systems Minimize the impacts resulting from such changes

6. Structural breaks identification framework Time Series Hypothesis test

7. OLS-based CUSUM test Ploberger and Krämer (1992) Single Structural Break – CUSUM Approach

8. Standardized partial sum process Appropriate for one single break Single Structural Break – CUSUM Approach

9. Breaks detection

10. Multiple breaks detection

12. Number of breaks: RSS and BIC

13. Multiple breaks detection: WATER LEVELS Location of Breaks

14. Multiple breaks detection: WATER LEVELS

15. Spatial Analysis of Breakpoints: WATER LEVELS Ladário 2.5 % breakpoints 97.5 % 1951 1960 1970 1974 1975 1978 Porto Murtinho 2.5 % breakpoints 97.5 % 1958 1961 1968 196919731974 Cáceres (DNPVN) 2.5 % breakpoints 97.5 % 197119731976 São Francisco 2.5 % breakpoints 97.5 % 197219731974 Aquidauana 2.5 % breakpoints 97.5 % 196719721977 Coxim 2.5 % breakpoints 97.5 % 197119731974 Porto do Alegre 2.5 % breakpoints 97.5 % 197119731976

16. Multiple breaks detection: RAINFALL

17. Porto Murtinho 2.5 % breakpoints 97.5 % 195819601972 196719701972 Upper Paraguay River 2.5 % breakpoints 97.5 % 1964 1968 1974 Cuiaba River 2.5 % breakpoints 97.5 % 196919711972 Aquidauana 2.5 % breakpoints 97.5 % ? Taquari River 2.5 % breakpoints 97.5 % - Correntes River 2.5 % breakpoints 97.5 % - Spatial Analysis of Breakpoints: RAINFALL

18. Long Term Memory and Structural Changes Yusof et al. (2013): “Observed long memory behaviour can be due to neglected structural breaks” Independence is an assumption of structural breaks tests May induce the choice of a long memory model

19. Summary An efficient way to identify and locate multiple step changes in hydrological time series was used; Along the river basin some gauging stations experienced step changes around 1960 (high / low) and another ones around 1973 (low / high) in water level time series; Similar phenomena occurred in rainfall time series in the northern part of Paraguay river basin, around 1960 (high / low) and around 1968 and 1971 (low / high).

20. Future work Hydrological simulation may confirm linkages between rainfall and water level step changes if it captures low frequency behavior – preliminary results indicate that groundwater may contribute to those low frequency fluctuations; Early 1980’s was a very rainy period in some major Brazilian river basins, but there is no report of extremely dry years in the 1960’s in other catchments – search for climate predictors; Hidden Markov Models (HMM) offer a possibility to model and simulate such series considering step changes.

21. Thank you for your attention! Marcus Suassuna Santos marcus.santos@cprm.gov.br www.cprm.gov.br