I'm studying to study hydrological regime of a watershed for a statistics project and I'm wondering if running a Mann-Kendall test would be useful for the analysis of these data set.
When comparing relationships for a paired watershed study, they typically recommend having at least ten years of data to evaluate the relationships, before proceeding with treatments. If you have other long term rainfall and stream gauging stations in proximity or at least within same physiographic area, you might be able to test whether your ten years are fairly representative, or whether the ten years include some extremes of flood or drought that may affect interpretations or utility for forecasting. When you order and plot annual or partial duration flood data, or low flow data, you may note one or more obvious outliers when plotting over the ten year record. I think the USGS uses about 25-30 years as the minimum to be considered a long term station. Watershed land use and stability changes might also be considered. A severe wildfire or mass clearcutting in a forested watershed, or water withdrawals for irrigation are examples likely to produce some marked changes. If you have a quality gauging site, stable, neither aggrading nor degrading, identify bankfull flow and frequency, Rosgen has suggested that two times bankfull flow depth at thalweg is about the 50 year floodprone depth. In the area I worked, ten times (one order of magnitude) above the mean annual flow approximates the 100 year flood. Comparing your ten years of data with other nearby and available long term rainfall and discharge data will help give a sense of how representative were those ten years in comparison to long term data sets. With ten years data, you should be able to develop a pretty good flow duration curve, with the recognition that the extremes are not well defined and cannot yet be relied upon if projected. This may be a sufficient trend for some to do some basic water resource analysis and planning. For research level work, ten years is about the minimum, and it may not be enough for some types of research. If your intent were to be to evaluate some topics as climate change, ten years is not enough by itself.
That is an interesting question. In general, it is acknowledged that Mann-Kendall trend test, for instance, needs at minimum 10-12 values before being considered 'valuable' or 'reliable'. Some other sources would mention 20, even 30. I guess it depends.
Practically, I would suggest evaluating the existence of a trend in your case using the Mann-Kendall test and a general linear trend test. The underlying assumption here would be: if there is a significant trend, it will be detected if your data reflects such a trend. If both tests are void, I would highly recommend not concluding there is no trend, but rather that no evidence of an existing trend was found based on the 10-years sample at study. Maybe because the sample is small, or maybe because the methods used were not sensitive to an assumed existing trend. Avoid generalizations in such case.
Also, consider seasonality. Often in hydroclimatic series, one would tend to forget testing such feature. You might want to apply M-K trend test with seasonality.
Generally 30 years of data are taken into consideration. In case data are not continuous and less than 30 years, it depends on the sentiveness of the study whether to go with the availabke data set or not.
When comparing relationships for a paired watershed study, they typically recommend having at least ten years of data to evaluate the relationships, before proceeding with treatments. If you have other long term rainfall and stream gauging stations in proximity or at least within same physiographic area, you might be able to test whether your ten years are fairly representative, or whether the ten years include some extremes of flood or drought that may affect interpretations or utility for forecasting. When you order and plot annual or partial duration flood data, or low flow data, you may note one or more obvious outliers when plotting over the ten year record. I think the USGS uses about 25-30 years as the minimum to be considered a long term station. Watershed land use and stability changes might also be considered. A severe wildfire or mass clearcutting in a forested watershed, or water withdrawals for irrigation are examples likely to produce some marked changes. If you have a quality gauging site, stable, neither aggrading nor degrading, identify bankfull flow and frequency, Rosgen has suggested that two times bankfull flow depth at thalweg is about the 50 year floodprone depth. In the area I worked, ten times (one order of magnitude) above the mean annual flow approximates the 100 year flood. Comparing your ten years of data with other nearby and available long term rainfall and discharge data will help give a sense of how representative were those ten years in comparison to long term data sets. With ten years data, you should be able to develop a pretty good flow duration curve, with the recognition that the extremes are not well defined and cannot yet be relied upon if projected. This may be a sufficient trend for some to do some basic water resource analysis and planning. For research level work, ten years is about the minimum, and it may not be enough for some types of research. If your intent were to be to evaluate some topics as climate change, ten years is not enough by itself.
But proper trend analysis you can take 10..10...10 yeas group it is helpful.
within 10 years span we can analyzes the trends but it is having limitation if there is no fluctuation ..or very lees trends better use more than 20 years
Precipitation is temporally and spatially variable and as you are studying precipitation, I think that 10 years is not enough. Although it is very depend on the topography of your study area. For example if you are studying a small island, 10 years could be acceptable. but 30 years is a usual duration for all situation.
In addition to Sayed's suggestion, I think if your data is daily data, then 10 years may be okay. If monthly, then 20 years and above might be nice but if you can access more data, the better for your result.
Well its enought time to study the trends over 10 years ;) Of course the 'trends' you detect or not will always be a function of both the length of record and the features of those 10 years compared to much longer periods. Probably why people are saying you need 30 years of data is because most runoff records 30-50 years long. Of course you can do the statistics but then be aware of the limitations and put your findings in context of longer records from other nearby/similar catchments.
I would recommend 30 years minimum if you are using monthly or longer time step data. It would also depend on the assumptions that you start out with. Knowing that we live in a "non-stationary" world (hydrologically) and the long-term variance may not revolve around a mean value, the longer the record and the better your situational awareness is of the study area, the better off you will be.
Rainfall data, as we all know, is highly variable over time. You could capture trend over any time period. BUT, that trend would represent the time period analyzed. So the real question is: what trend are you trying to capture? If you're trying to capture the trend over a particular short period then you could use that particular period of record for the analysis. But what would it mean?
If you're trying to capture trend for the purpose of identifying dependence on climate change, then obviously a short period wouldn't do. One would want to use a period over which we think climate change is occurring. Current thought is that rapid change began in the early 1970s. Therefore to capture that change, one would want data beginning at that time. We might assume the longer the period of record, the better for assessing climate change. For example, using a period of record from the early 1970's to the present would likely give a better estimate of the trend.
On the other hand, if trying to get an estimate of precipitation trend associated with urban heat islands, one would want a period covering the period of the development of the causes of the heat island. But of course there are difficulties in separating out different causes. And similarly for assessing stream flow dependence on urbanization.
Then there is the question of whether trends are linear or not, and if not, then what?
However your question is about the usefulness of the Mann-Kendall test. If you peruse the literature on climate change, you'll find Mann-Kendall is often used. But is has limitations. It is a non-parametric test to see of a trend is monotonically increasing or decreasing. You might look at https://vsp.pnnl.gov/help/Vsample/Design_Trend_Mann_Kendall.htm
The short answer is an emphatic NO. A reasonable time frame for hydrometric trend analysis must be longer than the underlying mechanism of change. The change in rainfall is generally related to global change, which can be 50 years or more. Streamflow may be change as a result of rainfall and climate change, in which case 50 or more years of data is necessary. Streamflow can also be caused by urbanization or other human interferences, in which case data spanning over the natural regime and urbanization regime must be used in any trend analysis.
Ferdous Ahmed Thankyou so much for the answer. I've received data of over 30 years of the watershed I was looking for. However, the watershed I'm trying to study has only 1 station that has measured discharge. Does studying data of only one station give credits to my research?
A study based on data from a single station will be most representative of that location. You will want to study the watershed closely to convince yourself (and future readers of your report) that the site is representative of the watershed--e.g., .?are there processes that either add or remove water from portions of the watershed that would not appear in your period of record
All previous answers are valid. It just depends on what you want to do with your results.
For sure you can not ncapture the low frequency signals that could appear in a longer time series.
But with ten years of daily data you can already produce a quite consitent statistic on extreme rainfall daily events, by using diverse distribution laws, and adjust the best to your sample distribution.
I strongly recommend you however to search for a nearby station with a longer time series, to assess how is the variability during the 10 years recorede, in regard of the variability on a longer time series, assuming that the nearby station you choose follows a not much different distribution low.
Kishanlal Darji Ferdous Ahmed Geoffrey M. Bonnin Hello, I'm obtained monthly and annual data for a period of 30 years. Now should I run MK test on annual data or the monthly annual data?
Khronostat is a free software to study stationnatity and non stationnatity in tome series. It includes pettitt test, lee and heghinian test, hubert test, buisson test, autocorrelation test.
If you want to perform precipitation and discharge trend analysis, use a minimum of 30 years of data (Preferably from the beginning of a decade). From your responses, I understand you have data.
Now, to answer you next question, I would suggest analyzing Seasonal and Annual trends. You can even go for monthly analysis if you want to investigate monthly trends for some reason!.
Please follow the tutorial given below to do the analysis.
Stream discharge is dependent on both short and long term weather. Many areas are subject to decadal fluctuations of precipitation which could introduce significant uncertainties in an analysis of just a single decade. Ten years is really not long enough to establish a mean annual precipitation. I'd prefer using three decades to attempt to smooth out that uncertainty. And then use three to five decades to establish any long term trend.
A decade of data could probably determine seasonal fluctuations and establish the relationship between precipitation and discharge but not the trend.