A flow register is non-stationary if some of the underlying properties (mean, variance) change over time. A trending series is an example of a non-stationary series. Non-stationarity occurs in addition to sudden changes, staggered changes in the series of records or if there is a marked fluctuation in the records. Fluctuations are related to natural variability (climate variability), especially in short records.
It is important to highlight that climate variability is not the same thing as climate change. When climate change is considered, the alterations occur in the long term. In terms of climate variability, the climate differs from one period to the next, but on average it maintains a stationary condition, unless there is also evidence of climate change. Climate variability may have the greatest influence on the behavior of maximum flow series with short records.
Causes of non-stationarity
The causes of the non-stationarity of record series can have different origins, namely:
- Problems with data records.
- Changes in the basin: changes in land use, construction of diversion structures, reservoirs and/or works for flood control.
- Variations in climate: climate variability and climate change.
Identifying the most likely cause of non-stationarity requires a detailed investigation of flow records and historical information related to the basin. Some requirements are:
- Verify the quality of the data.
- Visualize the historical behavior of the records and compare it with neighboring stations.
- Evaluate tests to verify if the causes of non-stationarity are due to climate variability.
- Research basin information related to land use changes, urbanization, reservoirs, diversion systems, etc.
Treatment of non-stationarity in maximum flow series
If the records show strong changes (mean and variance), dependencies, and trends and are used for frequency analysis, the results obtained, at best, may represent the average response during the recording period and if not properly addressed the results may be inconsistent. Depending on the causes of non-stationarity, the following actions may be considered:
- Non-stationarity caused by problems with the records: the recommended action is to correct the data, but if this is not possible, it may be necessary to use only a part of it.
- Non-stationarity caused by changes in the basin: in this case it is preferable to use only the part of the records after the change. However, care must be taken with the exchange rate and its variation over time should be evaluated.
- Non-stationarity due to short registrations: in short records, the most likely cause of non-stationarity is climate variability. If climate variability is confirmed as the cause of non-stationarity, it is recommended to perform the analysis considering covariates related to these climatic mechanisms. For the analysis of climate change, it is suggested that we have at least 50 years of records. Another consideration that must be taken into account is to choose study areas that have been little intervened.
- Non-stationarity caused by non-obvious causes: if it is not possible to identify the causes that cause non-stationarity, the records can be used in frequency analysis, but with the respective caveats in the results obtained.
Recommendations
- Before starting a frequency analysis, you must be aware of the length of the records, study if they have any non-stationarity and understand their cause or origin very well. In this way, the estimation of extreme events at the study site can be properly addressed or even identified if a regional frequency analysis needs to be considered.
- Failure to consider non-stationarity when estimating extreme events can lead to an underestimation, or overestimation, of peak events, with the consequent increase in the risk of hydraulic works.
- When covariates are used in frequency analysis, it is very important to understand their compartment and what is the appropriate way to incorporate them into studies. In the case of covariates associated with ENSO, it is important to investigate what effects climate change can have on them, either in their frequency and/or magnitude. In recent years, several studies have been developed that evaluate these aspects.
