tag:blogger.com,1999:blog-9093436161326155359.post2772224253576114961..comments2024-03-28T06:43:02.954+00:00Comments on Variable Variability: A framework for benchmarking of homogenisation algorithm performance on the global scale - Paper now publishedVictor Venemahttp://www.blogger.com/profile/02842816166712285801noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-9093436161326155359.post-65101176392090611422014-10-16T16:23:50.952+01:002014-10-16T16:23:50.952+01:00Yes, the trend in the data itself is also not rele...Yes, the trend in the data itself is also not relevant (for relative homogenization methods).<br /><br />I looked at Figure 2 again and think I now understand what you want to say. Yes, you need to take the uncertainty into account and because we work world wide, we probably also have to take the spatial variability in the climate and in the non-climatic changes into account. The results will depend on such considerations. <br /><br />We have not worked much on the validation part. We did formulate some principles, but no specific error measures yet.Victor Venemahttps://www.blogger.com/profile/02842816166712285801noreply@blogger.comtag:blogger.com,1999:blog-9093436161326155359.post-42953481172099874702014-10-16T16:05:15.290+01:002014-10-16T16:05:15.290+01:00I intend (not sure yet if I will manage to do it) ...I intend (not sure yet if I will manage to do it) to compete with HOMER, not Craddock, to get some more hints about the reliability of our results regarding Slovenian climate time series.<br /><br />I didn't mean the sign of the trend in the inhomogeneities, but the trend in data itself, i.e. trend in a homogenised time series vs. trend in a clean-world time series for the same station. For example, you get -0.3 °C/century for clean-world time series of Ljubljana and +0.2 °C/century after homogenisation. If the uncertainty at 5 % level is, let say +/- 1 °C/century (both trends insignificant), this is very different from the uncertainty of +/- 0.1 °C/century (both trends significant). This applies to Willet et al. (2014), Figure 2. You may consider statistics counting hits or faults regarding the sign & statistically significance of trends (e.g. trend is either positive significant or negative significant or insignificant).<br /><br />Best regards,<br /><br />Gregor Gregor Vertacniknoreply@blogger.comtag:blogger.com,1999:blog-9093436161326155359.post-1589727305965329412014-10-16T15:00:31.749+01:002014-10-16T15:00:31.749+01:00Gregor, does that mean that you would like to comp...Gregor, does that mean that you would like to compete? That would be great! Everyone is invited. <br /><br />I had thought that a global dataset is a bit too large to be homogenized with Craddock, though. :-) We will also select some smaller regions where people with less automatic and robust methods can show off their skills.<br /><br />Filling (and later gridding) will not be studied in this first cycle of the ISTI, is my current understanding. That is a pity, but we are quite limited in manpower, it is basically a volunteer project. Funding agencies find impact studies for cauliflower agriculture more important.<br /><br />You are right, the sign of the trend in the inhomogeneities does not matter. That should not have slipped through. :-(Victor Venemahttps://www.blogger.com/profile/02842816166712285801noreply@blogger.comtag:blogger.com,1999:blog-9093436161326155359.post-44595346572270057792014-10-16T14:48:49.897+01:002014-10-16T14:48:49.897+01:00I've just read the paper. Seems to be pretty i...I've just read the paper. Seems to be pretty interesting project and a chance for homogenisers to compete a little bit again :))<br /><br />I would like to ask if the final results ought to be interpolated (for missing values) or only homogenised values are going to be compared? <br /><br />The paper mentions that wrong trend sign could be problematic, but I would stress that only if the true and the homogenised trends are opposite and statistically significant. On the other hand, if both trends are insignificant, the sign doesn't matter very much. <br /><br />I'm looking forward to seeing the benchmark dataset :)<br /><br />Regards, <br /><br />GregorGregor Vertacniknoreply@blogger.com