Los datos de información pluviométrica corresponden a registros de lluvia caída en diferentes duraciones. Tenemos datos cada 24 h, 12 h, 1 h, 5 minutos, etc. Antes de completar datos es necesario verificar la calidad la información debido a que hay innumerables fuentes de error: de transcripción, pérdidas de información, etc. Análisis como: estacionariedad, homogeneidad, consistencia, y adecuación.

Una vez hecho esto, podemos proceder a completar la información. Si los datos de la precipitación no di fieren entre ellos más de un 10 %, podemos utilizar para completar el dato que falta de la lluvia diaria de un cierto día, la media aritmética de los datos de los demás observatorios.

Si las diferencias son más importantes, se puede aplicar el procedimiento propuesto por el National Weather Service

También es posible realizar correlaciones estableciendo una función de correlación entre la precipitación de una estación Y y las demás. Le recomiendo que la función de correlación sea de tipo potencial.

Try comparing CRU data with the instrumental data to check the calibration. This will tell you if their data set provides a reasonable approximation for the missing points.

What is the nature or type of your dataset? Give more details that can guide. Meanwhile, most of the times, interpolating with the most accurate dataset is best; using nearest neighbour, kringing or other interpolation types. The use of CRU data is also recommended in-line with Stephan Woodborne.

Dear Kishor

We use the CLIMATOL package, available for R. You can see it at "http://www.climatol.eu/", it is available in English, French and Spanish. We have done very well with this kind of tools. A master student made a routine with support vector machines and it also works very well. We can communicate with you and help you with your data.

Long term regression model through AI and then back interpolation

The best way is by means of autoregressive method. Other option is by using stations close. You have to adjust data to the elevation of the station you are working with.

For reference I suggest you to review my paper about climatic variation on high mountain environment. I used both methods.

Eneche Patrick Samson Udama It's rainfall data (monthly) but data for some months and years are missing. I want to fill in this data. Around 5-10% of data are missing.

Thanks

V. Soto Thank you I will look at it.

Luis Morales-Salinas I appreciate your help. I have used climatol package to climate diagram but not aware of filling missing data. I have data set and we can collaborate till publishing it. I have to fill in monthly precipitation data.

Stephan Woodborne Thanks very much

Hi you ask an interesting question. There are several ways to look at missing data. Pure statistical solutions are to use just in last issue. It is much preferable to work on a single point approach, using closest stations for instance and month to month correlations, with one or several close stations -if there are some. For months with very low or no rainfall it is quite easy. For very rainy months it must be paid close attention to the choice of the closest stations. It is also preferable to work with daily data in this case if there are some stations in the area with daily data available, and to compare daily time series of close stations, to be sure not to miss a very intense event in the area, which would be mainly missed by a usual statistical approach.

Filling missing climatic data could be done in 2 steps. First, a good knowledge of climate pattern over the region of investigation is a prerequisite. This includes the knowledge of, for example, rainfall pattern (one or two rainy seasons, rainy months, etc.). Then, the data quality control (per cent of missing data) to check whether it is worth filling the gaps (per cent of missing data less than a threshold of 25%) or simply cancelling the data and search for other datasets. Once, the decision of filling gaps emerges two approaches are used: the deterministic (example inverse distance weighting) or geostatistics tools. Deterministic approaches are relatively simpler but associated to larger uncertainties. Below the link to a publication where rainfall and streamflow data were filled out: Article Evaluation of recent hydro-climatic changes in four tributar...

Dear Kishor: You have all our help, my email is [email protected]. We have already made a code in R that fills very well monthly average data and we can share it with you. Please contact us. Best regards.

Thanks Dear Luis-Morales-Salinas

I am however very suspicious about automatic processes to fill in gaps, as I said previously it must be paid close attention to the local variability, which can be shadowed by a "blind" procedure. It also depends of course on the size of the dataset to process.

Can be used Interpolation techniques (Many are) which is very close to your nearest values.

I have more than 30 years daily observation data recorded by Automatic weather stations. There is a lot of missing data,, Please help me fill in the missing values,, I will send the data to whom who can help me,

i hv more than 30 years daily observation data recorded by Automatic weather stations. There are alot of missing data,, Plz help me filling the missing falues,, I will send the data to whom who can help me,,

many thanks in advance,,

Kishor Prasad Bhatta Thirumurugan Perumal Gil Mahe Luis Morales-Salinas Djigbo Félicien Badou V. Soto Eneche Patrick Samson Udama

Dear Ali, If you wish we can help you with your problem about correcting and filling in missing data. What is the time interval of the data (Monthly, Daily or hourly). As Gil told us, this method is based on SVM and has its limitations, which you should be clear about.

You may be interested by this paper in which we describe 2 methods to fill in daily missing rainfall data

https://www.researchgate.net/publication/48413622_Study_of_water_stress_and_droughts_with_indicators_using_daily_data_on_the_Bani_river_Niger_basin_Mali

Hello Dr. Bhatta, I suggest the regional vector method Brunot-Moret (1969), available in the Hydraccess software; the normal ratio method (Mohanty et al. 2014) is also effective

Dear Kishor:

Maybe you can check out some R packages, and see which one best meets your needs. I think it is key that you identify the time interval to interpolate. I imagine short periods are no problem, but at the year level, it might be inconsistent. Here some examples that may be useful.

MICE

Amelia

missForest

Best Regards.

https://iwaponline.com/hr/issue/48/4 @

Assessment of different methods for estimation of missing data in precipitation studies

Mohammad-Taghi Sattari;

Ali Rezazadeh-Joudi;

Andrew Kusiak

this paper shall let your problem solved

@

I suggest to estimate the missing data from the satellite or from any other meteorological parameters.

https://gking.harvard.edu/amelia

you can tried with AMELIAII, i used to filling the gaps of rain time series and i obteined good results. Of course depend how is your data base, but you can tried because its easy and fast to use

Here's a comparison of different techniques and a free python routine (VS) to recover missing data, based on data-pattern analysis:

https://doi.org/10.1175/JHM-D-19-0220.s1

VS can be used for different station data, not only rainfall.

I think Missforest may be suitable for hydro-climatological data as well:

https://towardsdatascience.com/missforest-the-best-missing-data-imputation-algorithm-4d01182aed3

https://academic.oup.com/bioinformatics/article/28/1/112/219101?login=false

https://cran.r-project.org/web/packages/missForest/index.html