PPresil-BM

Overal description

Bayesian Harmonic model for time series. For each year the program choose using Bayesian Factors among:

• the model of the year before
• a 3 harmonic model(yearly, semesterly and envery 3 month) with a priori parameter partially informed using the full time series
• a flat trend model.

Using the fitted model the program extract several yearly seasonal statistics and a single interyear standard variation statistics and their standard errors. In alternative is possible to obtain also the fitted value of the model with relative standard error

Software dependency

The code is writte in Python 3.6 and
requires the following packages available on pypi: xarray, pandas, numpy, rasterio, sklearn, patsy, scipy
requires the following package avalable on github: linear_model (https://github.com/asherbender/bayesian-linear-model) here linear_model.py is already included with change to pass python3

optionally for the drawing of graph and using this program as a library it requires: seaborn

The annual statistics

The 3 annual statistics are estimated as default:

• Annual Mean
• Annual Standard Deviation
• Annual Day of Maximum

Other statistics are optional:

• Day of the minimum and its value
• Value of the maximum
• Day of maximum change and its value expressed as difference between the two observed day around moment of maximum change

The Input Format

The program expect as input file a gdal readable geospatial raster dataset in which band name is formatted as time or as a string in which the last token is recognizable as a date, using as token separator the string "_".

PPresil options: optional arguments:

-h, --help            show this help message and exit
--inputfile INPUTFILE
                      path to input file
--suffix SUFFIX       suffix for output
--step STEP           size of chunk to analyze
-n N                  n-element of size stepxstep to analyze
--noffset NOFFSET     offset to start counting n-element of size stepxstep
                      to analyze
--change              allow change of seasonality
--Expected            outout expected value and do not estimate annual
                      phenological statistics
--Event               store the date and value of the maximum change within
                      each year
--PP                  perform Posterior predictive test [experimental]
--dense               assume that time series is dense and each year can be
                      fitted with its own seasonality model
--freqsubsample FREQSUBSAMPLE
                      frequency of sampling from model in days used in order
                      to estimate phemolgical statistics
--reps REPS           replicated of model sampling in order to estimate
                      standard variation of prediction
--dask                replicated of model sampling in order to estimate
                      standard variation of prediction
--netcdf              replicated of model sampling in order to estimate
                      standard variation of prediction
--minmaxFeat          add to summary statistics min and max value and min
                      position
--small               save on single file even if option n was used

The Output Format

The output format is a netcdf with as variables the statistics requested in input.