Independent reevaluation of the Mann-Bradley-Hughes Nothern Hemisphere Climate Reconstruction

Goal

The general goals of these pages were stated above (see Millennium Home Page). In short, we want to provide the available climate reconstruction methods in an accessible format to the climate research community so that everybody can not only reproduce the individually published reconstructions, but that everybody can study the method's behavior and evaluate strength and weaknesses. The example shown here employs the Mann-Bradley-Hughes climate field reconstruction. All evaluations below are based on this method only and test reproducibility and a small number of performance tests tailored after recent criticisms.

We are working on an update of this site and a cleaned-up version of the code since the Wahl and Ammann paper is accepted by Climatic Change and in press. Once GRL has reached a decision, we will also post the appropriate codes and illustrations here. (Note, the current version is fully functional, and no results will change.)

Brief Introduction

The MBH reconstruction is one of a growing number of climate reconstructions based on high-resolution proxy data from past centuries. Compared to others, the MBH method employs the dominant scale temperature patterns to make use of spatially coherent variations in the proxies rather than treating each record independently. Using the spatial information, this reconstruction can be used to imply climate variations in remote areas where no proxies are available. It is also the only climate reconstruction that provides climate at regional scales, a key strength when it comes to studies that focus at impacts of climate at the most relevant spatial scales for most of the decision makers as well as for individual societies in general. The global or hemispheric temperature series are, therefore, somewhat of a 'side product', although of key interest, no question. A further strength of the MBH method is also its potential to be robust against noise.

Primary Code

NOTE:The version of our code as posted is preliminary as the accompanying paper is under review. However, this code is complete and is fully functioning. It has been applied to reproduce the MBH original. Please refer to Wahl and Ammann for further detail (available when in press). We want the users also to be aware of two important simplifications:

The instrumental data is decomposed on the annually averaged level rather than on the monthly as in MBH98.

The weights of the individual proxies have been dropped.
The code written in the freely available and platform independent R-programming language is separated into three primary files: (full code incl. input data is here)

0. Generate a new subdirectory (e.g., WA_original), and save the code incl. datafiles (unpack full tar)

1. The CONTROL-File (control_WA_original.txt) file that contains all the information about input files, length of the proxy records, number of reconstruction periods, etc. Each descriptor line is followed by the actual entry that is to be used in the reconstruction. For example, the years to be used as the calibration period are by default "1902 1980" but could be changed to something different, e.g., "1902 1971"

2. The MASTER-File (mann_masterfile.r) is the actual run code that guides the reconstruction by calling the control-file for instructions and then by looping over the reconstruction periods as specified in the control-file. Note, in the current version, this code is ready for a UNIX-based system. Read the comments to make the code compatible with Windows. The only issue is with system commands that clean up or tar-up the output directory. The code is tested on MAC OS-X (10.3.x).

3. The core reconstruction is done in the FUNCTIONS-File (mann_Rfunction.r). This file contains all the functions necessary both for I/O (read.control) as well as the core of the reconstruction.

Please visit again after full release of the Wahl and Ammann paper for a detailed description of the code and the individual scenarios discussed in the paper.

Data Structure

The input data is currently structured so that the Wahl-Ammann reproduction of MBH can be performed. The data is stored in a subdirectory called manndata (contained in full tar-file). Currently, we release a code that reads separate ASCII-input files for each of the reconstruction periods. We are working on providing a version that read large data-tables as well as self-referenced netCDF files (e.g. from GCM output).

Applications: Verification of Primary Results by McIntyre and McKitrick

Currently, we can show results from four primary scenarios of interest:

(a) Wahl-Ammann Primary Reconstruction of Mann-Bradley-Hughes (WA-Original).
(b) Scenario with complete omission of N-American data (cf. McIntyre and McKitrick 2003)
(c) Scenario where Bristlecone/Foxtail Pine records have been removed prior to PC calculation for the N-American ITRDB dataset. Records have been centered but not standardized, according to McIntyre and McKitrick (GRL 2005a). The reconstructed global temperature PCs are not rescaled to the variance of the instrumental PCs (equiv. to McIntyre and McKitrick 2005b, Energy and Environment). Two network resolutions are shown: 1400 and 1450. The Gaspe series has been removed.
(d) Scenario with McIntyre and McKitrick-suggested centering of N-American ITRDB data (centering over full data length rather than the MBH method of centering relative to the restricted calibration period), however with standardization of individual records. The Gaspe series has been removed.

Results are shown in the figure below. In (b) and (d) both the 1400 network (1400-1449) and 1450 network (1450-1980) are shown; in (c) both networks are shown at full length. Note: All curves shown in pink are statistically and climatologically NOT meaningful because they fail validation. It is imperative that the validation statistics are presented to gauge the meaning of a computed climate reconstruction. The calibration and verification RE for the shown scenarios are listed in the Table 1 below. The underperformance of (b) and (c)-1400-network is clearly visible and contrasts strongly against the other reconstruction models, as well as the instrumental record of the 20th century. Independent verification period applied was 1854-1901.

Click on Image for large version

Figure 1: Comparison of different reconstruction models with WA_Original and with Instrumental data (1902-1980).(a) shows the successful reproduction of the original MBH climate reconstruction using the code provided above. (b) shows the statistically and climatologically meaningless reconstruction based on the omission of the N-American proxy information, combined with lack of rescaling of the reconstructed global temperature PCs (equivalent to McIntyre and McKitrick, 2003). (c) shows two networks (1400 and 1450) for series where the N-American ITRDB data was summarized by PCs, using McIntyre and McKitrick-suggested centering without standardization of records (2005a), incl. lack of scaling of reconstructed global temperature PCs (McIntyre and McKitrick, 2005b). The 1400 network is statistically and climatologically meaningless, while the 1450 network, which performs very close to WA_Original, provides a valid reconstruction. (d) a valid reconstruction based on MM-centering, including full standardization of the ITRDB data.

Table 1: Validation Statistics of reconstruction models shown in Figure 1

Reconstruction:	Compare to:	Calibration-RE	Verification-RE	Pass/Fail
(a) WA_Original 1400 network	Wahl and Ammann	0.39	0.48	Pass
(a) WA_Original 1450 network	Wahl and Ammann	0.47	0.44	Pass
(b) Omission N-American records 1400 network	McIntyre and McKitrick (2003)	-0.42	-0.57	Fail
(b) Omission N-American records 1450 network	McIntyre and McKitrick (2003)	-0.65	-2.71	Fail
(c) Omission BC/FT Pine unscaled: 1400 network	McIntyre and McKitrick (2005b)	-0.34	-0.56	Fail
(c) Omission BC/FT Pine unscaled: 1450 network	McIntyre and McKitrick (2005b)	0.32	0.14	Pass
(d) MM-centering, standardized: 1400 network	Wahl and Ammann	0.32	0.18	Pass
(d) MM-centering, standardized: 1450 network	Wahl and Ammann	0.49	0.46	Pass

In Summary, it can be clearly stated that none of the warm 15th century reconstructions turn out to be statistically and climatologically meaningful. The centering issue raised by McIntyre and McKitrick in GRL (2005a), if approached properly (i.e. using full standardization of individual records), is influencing the reconstruction in a minor way and is in fact confirming the robustness of the MBH reconstruction within its own framework.

Home to Millennium Main Page

Caspar Ammann

Last modified: Mon Mar 25 13:57:32 MDT 2006