Monday, September 12, 2016

Homogenisation of temperature and precipitation time series with ACMANT3: method description and efficiency tests

Paper featuring John Coll of ICARUS. Post written by John Coll
International Journal of Climatology July 2016 doi: 10.1002/joc.4822
Time series homogenisation and the multiple breaks problem
During the long period of climatic observations, station location, instrumentation and several other conditions of the observation may change, resulting in non-climatic temporal variation in the observed data.  Such non-climatic changes “inhomogeneities” affect the usability of observed data to the detection of climate change and climate variability.  One important present task of climate science is to provide accurate regional and global mean temperature trend estimates (Rohde et al., 2013; Rennie et al., 2014; Venema et al., 2015), and homogenisation significantly contributes to that. The most frequent way of time series homogenisation is the use of statistical procedures.  The direct aim is to identify and correct statistically significant shifts in the section means (they are the estimated timings of technical changes, often referred as breaks or change points).  To separate the inhomogeneities from the true climatic variation (the latter never should be removed from the data), homogeneity tests are usually applied to the differences between a candidate series and other series of the same climatic area (“relative homogenisation”), rather than directly to the candidate series (“absolute homogenisation”).

We have the general experience that climatic time series contain about 5 breaks per 100 years on average (Venema et al. 2012).  Although statistical homogenisation has a century long history, the theory and development of multiple break homogenisation offering mathematically higher level solutions appeared only in the 1990s coincident with the more widespread use of personal computers.  One early representative of multiple break methods was PRODIGE (Caussinus and Mestre 2004).

During the European project COST ES0601 (known as ‘HOME’, 2007–2011)  two new multiple break methods were created based on PRODIGE: one is the fully automatic ACMANT (Adapted Caussinus–Mestre Algorithm for homogenising Networks of Temperature series, Domonkos, 2011) and the other is Homogenisation software in R (HOMER, Mestre et al., 2013), the interactive homogenisation method officially recommended by HOME.  Both HOMER and ACMANT include the optimal step function fitting with dynamic programming for break detection and the network wide minimization of residual variance  for correcting inhomogeneities (ANOVA, Caussinus and Mestre, 2004; Domonkos, 2015).  Both HOMER and ACMANT provide additional functionality relative to the parent method PRODIGE, and they are assumed to be the most efficient homogenization methods nowadays.

ACMANT3 development and discussion

This paper describes the theoretical background of ACMANT and the recent developments, which extend the capabilities, and hence, the application of the method.  The most important novelties in ACMANT3 are: the ensemble pre-homogenisation with the exclusion of one potential reference composite in each ensemble member; the use of ordinary kriging for weighting reference composites; the assessment of seasonal cycle of temperature biases in case of irregular-shaped seasonal cycles. ACMANT3 also allows for homogenisation on the daily scale including for break timing assessment, gap filling and analysis of ANOVA application on the daily time scale.

ACMANT3 is a complex software package incorporating six programmes, these are: temperature homogenisation with a sinusoid annual cycle of biases; temperature homogenisation with an irregular annual cycle of biases; precipitation homogenisation.  Each of the preceding three has monthly and daily homogenisation versions (; and in total the six programmes incorporate 174 sub-routines.  The software package also includes auxiliary files to support network construction. However, despite its complicated structure, ACMANT provides the fastest method implementation among all the available automatic homogenisation methods.

Considering the similarities of the theoretical background of HOMER and ACMANT, the choice between HOMER and ACMANT for particular homogenisation tasks should be based on the dataset characteristics.  The use of ACMANT is particularly recommended for (1) datasets with little or no metadata; (2) datasets from dense networks with large numbers of time series and where there are high spatial correlations; (3) very large datasets (>200 time series) for which the use of automatic methods is the most feasible and easily managed solution.   

Figure 1: Errors of raw data and residual errors of ACMANT homogenised data in a test dataset of simulated air surface temperatures. AC1, AC2, AC3 mean the first, second and third generation of ACMANT. Upper left: root mean squared error (RMSE) of monthly values, upper right: RMSE of annual values, bottom left: trend bias for individual series, bottom right: network mean trend bias. Smean means systematic trend bias. 

The efficiency tests presented in this paper provide firm indications that ACMANT3 can considerably reduce initial regional trend biases at any spatial scale, although the efficiency achieved depends both on the spatial density and the extent of the intact record of the observational data.  Further research is needed in this important and emerging area, for both the development and testing of statistical methods (Domonkos and Guijarro, 2015) and alongside an analysis of the causes of possible systematic biases in temperature records, with parallel measurements (http://www.surface

The authors also have another ongoing collaboration as part of the Irish Environmental Protection Agency funded “HOMERUN” project (e.g. Coll et al., 2015a,b) which aims to homogenise the large and dense Irish precipitation dataset with ACMANT and HOMER and explore more details about the practical application of these methods.  More details are available from

The ACMANT3 software package together with its manual is freely accessible from data.html.

Caussinus H, Mestre O. 2004. Detection and correction of artificial shifts in climate series. J. R. Stat. Soc. C 53: 405–425, doi: 10.1111/j.1467-9876.2004.05155.x.
Coll J, Curley C, Domonkos P, Aguilar, E, Walsh S, Sweeney J, 2015a.  An application of HOMER and ACMANT for homogenising monthly precipitation records in Ireland.  Geophysical Research Abstracts 17: EGU2015-15502.
Coll J,  Domonkos P, Curley, M, Aguilar E, Walsh S, Sweeney J, 2015b.  IENet: A homogenised precipitation network for Ireland – preliminary results.  10th EUMETNET Data Management Workshop. St Gallen, Switzerland.
Domonkos P. 2011. Adapted Caussinus-Mestre Algorithm for Networks of Temperature series (ACMANT). Int. J. Geosci. 2: 293–309, doi: 10.4236/ijg.2011.23032.
Domonkos P. 2015 Homogenization of precipitation time series with ACMANT. Theor. Appl. Climatol. 122: 303–314, doi: 10.1007/s00704-014-1298-5.
Domonkos P, Guijarro JA. 2015. Efficiency tests for automatic homogenization methods of monthly temperature and precipitation series. 10th EUMETNET Data Management Workshop Oct 28-30, St. Gallen, Switzerland.
Mestre O, Domonkos P,  Picard F,  Auer I, Robin S, Lebarbier E, Böhm R, Aguilar E,  Guijarro J, Vertacnik G,  Klancar M, Dubuisson B, Štepánek P. 2013. HOMER: homogenization software in R – methods and applications. Idojaras Q. J. Hung. Meteorol. Serv. 117: 47–67.
Rennie JJ, Lawrimore JH, Gleason BE, Thorne PW and others. 2014. The international surface temperature initiative global land surface databank: monthly temperature data release description and methods. Geosci. Data J. 1/2: 75–102, doi: 10.1002/gdj3.8.
Rohde R, Muller R, Jacobsen R, Perlmutter S, Rosenfeld A, Wurtele J, Curry J, Wickham C, Mosher S. 2013. Berkeley Earth temperature averaging process. Geoinform. Geostat. 1: 2, doi: 10.4172/gigs.1000103.
Venema V, Jones P, Lindau R, Osborn T. 2015. Is the global mean land surface temperature trend too low? 15th Annual Meeting of the European Meteorological Society Sofia (Bulgaria), EMS2015-557.
Venema V, Mestre, O, Aguilar E, Auer I, Guijarro JA, Domonkos P, Vertacnik G, Szentimrey T, Stepanek P, Zahradnicek P, Viarre J, Müller-Westermeier G, Lakatos M, Williams CN, Menne M, Lindau R, Rasol D, Rustemeier E, Kolokytha, K, Marinova T, Andresen L, Acquaotta F, Fratianni S, Cheval  S, Klancar M, Brunetti M, Gruber C, Duran MP, Likso T, Esteban P and Brandsma T. 2012: Benchmarking monthly homogenization algorithms. Climate of the Past, 8, 89-115, doi:10.5194/cp-8-89-2012.

Thursday, September 1, 2016

New paper by John Coll et al.: Projected climate change impacts on upland heaths in Ireland

Projected climate change impacts on upland heaths in Ireland
Climate Research July 2016 doi: 10.3354/cr01408

Ireland has a high proportion of the northern Atlantic wet and alpine and boreal heaths of high conservation value within Europe.  These upland habitats of and their associated oceanic species and vegetation are of high conservation value, but are also considered vulnerable to climate change.  For example, is anticipated that an amplification of the elevation-dependant warming already detected will accelerate the rate of change in mountain ecosystems, with the potential to exacerbate both the pace and the amplitude of extinctions of vulnerable upland species uniquely adapted to these habitats.

However, projections from different climate models vary markedly and local processes for upland regions are poorly captured, hence more localised modelling studies are required to inform management decisions.  Various modelling approaches have been used to convert species distributions into predictive maps, and bioclimatic envelope models (BEMs) are widely used.   However, confidence in the predictive power of BEMs is compromised by conceptual, biotic and algorithm flaws. Arising from this, the use of consensus methods is popular on the basis that they decrease the predictive uncertainty of single-models to give a probability distribution per pixel as opposed to a single value. 

The use of BEMs for habitats is novel, and only a limited number of studies have applied these methods to landforms and habitats. In this work seven bioclimatic envelope modelling techniques implemented in the BIOMOD modelling framework were used to model Wet and Alpine and Boreal heath distributions in Ireland.  An ensemble prediction from all the models was used to project changes based on a climate change scenario for 2031 to 2060 dynamically downscaled from the Hadley Centre HadCM3-Q16 global climate model. The climate change projections for the individual models change markedly from the consistent baseline predictions.  Projected climate space losses (gains) from the BIOMOD consensus model are -40.84% (limited expansion) and -10.38% (full expansion) for Wet heath (Figure 1a); and -18.31% (limited expansion) and +28.17% (full expansion) for Alpine and Boreal heath (Figure 1b).

Fig. 1. Mapped BIOMOD consensus model outputs for (a) wet heath and (b) alpine and boreal heath habitats based on median probability ensemble forecasting method values using the true skill statistic threshold. Red squares denote projected losses of climate space for the A1B 2031−2060 scenario relative to the baseline; blue squares denote stable climate space grids (areas of suitable climate under a no dispersal—no habitat expansion—scenario); green squares denote potential climate space gains relative to the baseline; blue and green squares combined indicate areas of suitable climate under a full dispersal (habitat expansion) scenario.

The projected decline and fragmentation of the climate space associated with heath habitats would
have significant implications for the ecology of these complex upland ecosystems and their associated species.  Results indicate that the distribution of wet heath habitats in Ireland is regionally sensitive to climate change, most notably for lower-lying areas in the south and west of the country. Increasing temperature and precipitation changes may reduce and fragment the area that is suitable for heath development.  Degrading heaths will also have an impact on the wider structure and function of the uplands as the overall mosaic of habitat types respond to climate change. For example, drier and warmer summers may increase the frequency, size and severity of uncontrolled fires, and drought effects may become more common later in the year. This may have severe impacts in areas already subject to pressures such as overgrazing, inappropriate burning, and loss of vegetation cover combined with erosion of the peat or soil.

Some attempt has been made to deal with uncertainty, at least in relation to differing results between the model categories, by providing the results from the individual BEMs implemented in the BIOMOD framework alongside the ensemble projection. Certainly, there is substantial variation in the results between the individual BEM types when the A1B scenario data are projected through the models.  Although only the downscaled output from 1 GCM and scenario has been used to project climate space changes, the methods lend themselves to using different GCM and RCM outputs from a range of scenarios to better encapsulate uncertainty. Thus e.g., given the importance of mean winter precipitation in all the BEM model families, if a wetter or dryer model or scenario had been used from the ENSEMBLES RCMs, the results projected via the BEMs could have varied further. 

Such an expanded framework would allow the identification of adaptation strategies that are robust (i.e. insensitive) to climate change uncertainties, and would allow more confidence in identifying and targeting vulnerable areas of heath habitat for priority conservation management measures.  These sort of refinements would also help inform best practice conservation management, whereby limited resources could be directed to areas coincident with healthy and functional heath communities and projected future climate suitability.