1.1-11: Metadata changes requested by Uwe Ligges 1.1-10: Revises the method of linking to Rcpp package 1.1-9: Register native routines and disable symbol search 1.1-8: Remove dependency on caTools 1.1-7: Bug fix in climdexInput.csv() 1.1-6: Bug fixes and documentation improvements - Fixed bug where nday.consec.prec.max() failed for even values of parameter "ndays". - Fixed bug in get.temp.var.quantiles where "qtiles" parameter was not being passed through to delegates. - Fixed bugs in documentation where multiparagraph descriptions were not being included. - Thanks to Nick Herold (UNSW) for his valuable feedback that was incorporated into this release. 1.1-5: Updated package to use Rcpp's new 'const correct' Vector::const\_iterator 1.1-4: Metadata changes required by Uwe Ligges 1.1-3: Bug fix; check.basic.argument.validity() used an incorrect test for checking the presence of the input time series - Thanks to Rebekka Posselt from MeteoSwiss for this bug fix. 1.1-2: Updates to maintainer and package documentation 1.1-1: Bug fixes for temperature-based threshold indices; CHANGES RESULTS. MAJOR CHANGES: - Changes code to reflect improved definition of day selection for first and last two (for a five-day window) days. Specifically, removes padding values from ends of series, and wraps the day around for each year at the edges instead of wrapping around into the next year. This changes results for both in-base and out-of-base quantiles for these days by a small amount. API CHANGES: - Removes pad.data.with.first.last.values option, as it is no longer needed. MINOR CHANGES: - Improve test framework to consume less memory and properly test bootstrap quantiles vs reference implementation. - Fix a bug where quantiles couldn't be specified to climdexInput.raw . - Change namespace usage and declarations to not be broken. 1.0-3: Eliminate compiler warnings; fix tests 32-bit platforms. - NOTE: Results for TN/TX 10/90p and W/CSDI may differ between ia32 and x86-64 platforms. This is due to SSE math being used by default on x86-64, and 387 math being used by default on ia32. Identical results to the x86-64 results can be achieved by adding -mfpmath=sse to CXXFLAGS. 1.0-2: Eliminate long lines and spurious directories/files from documentation. 1.0-1: Allow computation of indices without supplying data for all variables; improve NA (missing value) handling. SUMMARY: - Code now allows you to only supply the data you have; formerly forced you to supply tmin, tmax, prec. - NA (missing value) handling has changed; output will only change if NA values were present, and will be limited to presence/absence of NA values. - Data structures have changed; however, If you are not supplying thresholds to be used when calculating TN/TX 10/90p AND you are not poking at the internals of the climdexInput structure, you will not need to change your code. MAJOR CHANGES: - NA (mising value) handling in the code has been revamped. - Fixed bug in TXx/TNx/TXn/TNn/DTR/RX1day where a single NA value would change the result for a month/year to NA. - Changed data presence criteria to mask data with NA for annual if any month has more than 3 NA values or the year has more than 15. - Previous behaviour: The year had to have more than 15 NA values. - Add a configurable threshold for the minimum amount of data that must be present for a particular 5-day window in order for quantile-based indices to be calculated for that day. - Previous behaviour: If any data was present, a quantile was calculated. - The internal data structures, as well as the format of supplied and generated out-of-base quantile data, have changed. - Documentation has been updated to reflect new formats (climdexInput.raw) and data structures (climdexInput-class) - The climdexInput.raw, climdexInput.csv, and get.outofbase.quantiles functions no longer require you to supply tmin, tmax, and prec data; supply any one of those, any two, or all of them. - Availability of indices will be dependent upon data availability. MINOR CHANGES: - Allow lazy evaluation of quantile thresholds. - Improves performance if not using TN/TX 10/90p or W/CSDI. - Change quantile routine for the single-value case to return the value instead of NA. - May cause some results to change from NA to a value; will only affect very sparse input data. - Fix bug with GSL_first, GSL_max, and GSL_sum modes of gsl routine. - Improper selection of periods resulted in occasionally selecting too long a period. - Fix minor bug where tx/tn 10/90p and (likely) w/csdi will not work with a 1- or 2-year base period. - Transition documentation to Roxygen2 - Added a test suite to test the code. - Added function to return list of available indices for a climdexInput object. 0.7-4: - Fixed a bug where tavg.dates was being incorrectly checked. 0.7-3: - Improve performance of creating climdexInput structure and GSL in the southern hemisphere. 0.7-2: - Clean up line length problems with documentation. 0.7-1: - Implement much faster bootstrapping algorithm. 0.6-2: - Update documentation to reflect all changes. - Update and clean up comments. - Update climdexInput.csv to accept climdexInput.raw's new arguments. - Change order of arguments to climdexInput.raw to match historical order. 0.6-1: MAJOR BUG FIXES, in order of importance (changes results): - Changed mistaken code which resulted in quantile type for tn/tx 10/90p being type=6 not type=8. In aggregate, this caused a slight negative bias; individual results may, under some circumstances, vary by very large amounts. - Changed quantile method for precipitation (r99ptot, r95ptot) from type=7 to type=8. This caused a small positive bias in aggregate, with large differences under some circumstances. - Truncate dates so that last day of data isn't omitted under some conditions. This can cause differences when data is centered on the middle of the day in the final year or month of data, and can cause the last value to be NA. - Modified GSL code to not report a GSL 1 day longer than it should have under many circumstances. - Modified GSL code to always use July 1st (if NH) or January 1st (if SH); may affect results under certain circumstances with Gregorian calendars. Minor bug fixes: - Mask data with NA for CDD/CWD where the entire year is wet/dry. Feature enhancements: - Exposed interface to quantile method used in C++ code. - Added multiple options to allow for compatibility with fclimdex. 0.5-2: - Export a few more functions 0.5-1: Feature additions: - Add option to produce output for rx5day that matches fclimdex - Add output of annual variables for TXx, TNx, TXn, TNn, TN/TX 10/90p, DTR, Rx1day, Rx5day. 0.4-1: MAJOR BUG FIXES (changes results): - r99ptot and r95ptot were using incorrect quantiles (all data, not just the base period). AFFECTS ALL PREVIOUS VERSIONS. - Growing season length in Southern Hemisphere was off by 1 year. Does not affect publicly released versions. Minor bug fix: - Fix bug which caused extra year to be included in data when data starts at base period start. 0.3.1-1: Fix introduced issue which broke calculation of GSL for Gregorian calendars; in process, fixed GSL to use July 1st as center. 0.3-2: Fix bug when days are centered at 12:00 (change round to floor) 0.3-1: Add support for GSL in southern hemisphere. 0.2-1: Add more input validation, clean up documentation, add example data, improve performance. No change in results. 0.1-3: Code changes to fix compilation on Solaris. No change in results. 0.1-2: Initial version