Recovery and disclosure of historical meteorological observations (Hisklim)

The aim of the programme is to make historical land and sea climate data from Dutch sources digitally accessible, with the highest possible time resolution and quality. The data extent back into the 17th century and stem from a variety of ship and station observations (see Figure 1 for the main types of data and time periods covered by Hisklim). The resulting high-quality datasets are needed to properly assess climate change and variability. Moreover, the datasets are also required to validate climate models.

The output of these models is the basis for the development of climate change policies and climate scenarios for the 21st century, which are increasingly being used in climate change impacts and adaptation studies. All historical data disclosed by Hisklim is (will be) freely available for research, engineering, and the general public from the KNMI website at http://www.knmi.nl/klimatologie.

In the first part of this highlight we present examples of the digitisation projects. Digitisation is a labour-intensive task that needs to be carried out with care. Figure 2 shows a specimen of an 18th century hardcopy data source available in the archive of KNMI. This kind of data is digitised by making an electronic copy (jpeg-images), which is used to type the data into a spreadsheet. Both the images and the data are made accessible via the internet. The work is embedded internationally: from 2001-2004 the digitisation of pre-1850 ship logs got a boost from the KNMI participation in the EU-funded project CLIWOC. From March 2005 until March 2009, part of the work is being funded through the KNMI participation in the National project Climate changes Spatial Planning. 

In the second part of the highlight, we present some examples of current work on homogenisation of climatological time series and datasets. Inhomogeneities result from e.g. changes in instrumentation, repositioning of instruments (both horizontally and vertically), changes in the surroundings like the growth of trees and the expansion of cities, and changes in measurement practices. We focus on improving homogenisation methods by studying the weather-dependence of inhomogeneities. An important goal is to obtain a homogeneous version of the so-called Zwanenburg/De Bilt subdaily temperature time series that covers the period 1706 until present. The digitisation efforts presented here are partly meant to obtain parallel time series for homogenising that series. To address deficiencies in present-day homogenisation methods, an EU COST action ES0601started in 2007. KNMI participates in this action. 
 

Digitisation

KNMI yearbooks
The 19th century KNMI yearbooks are an important source of information. The meteorological observations in these books constitute a crucial link between the pre-1850 measurements (before the founding of KNMI) and the modern post-1900 observations. The observations consist of (mostly) three measurements a day of air temperature, air pressure, humidity, wind speed and direction, cloudiness, and of daily measurements of precipitation of a varying number of stations (23 in total) distributed across the Netherlands. In the year 2006 we completed the digitisation of these data. About 0.6 million records (550 station-years) were digitised and quality controlled. About 40% of the data is now available via the KNMI website and the remainder will become available in 2007.

Amsterdam City Water Office
From about the year 1675 onwards, the Amsterdam City Water Office was responsible for the water management of the canals in the city. At that time, Amsterdam was situated along the open sea, and the canals in the city, which also served as sewer system, had to be flushed using the tide. Meteorological measurements were made by employees of the office to support this work. These measurements are unique in the sense that in the 1784-1963 period the measurements were taken hourly. In the beginning air temperature, windspeed and direction were measured. From 1824 onwards, air pressure was added. In the 1784-1963 period, the office moved three times to a new location. The handwritten observer logbooks were put on microfilms in 1984 by the Amsterdam City Archive and KNMI obtained a copy. Recently, we transformed the films into jpeg-images which were subsequently used as a basis for digitizing all 1.6 million observations. The first part of the work was finished in 20042), the second part, including Quality Control (QC), in 2006. In 2007 the data will become available.

Daily rainfall 1850-1950
KNMI measures daily rainfall since about 1850. In the 1850-1950 period, the network gradually increased to its present density of about 300 rain gauges. The measurements are taken by voluntary observers every morning. In the first decade of the 20th century a standardization of the measurements was implemented. In contrast to the post-1950 period, where all observations are digitally available, only about 10% of the observations were digitally available in the 1850-1950 period. In the years 2005 and 2006 we digitized all pre-1951 daily data, amounting to about 4.7 million observations (13500 station-years). It took two man-years of typing in the data. The data will become available in 2008 after QC and homogeneity testing.

Rainfall strip charts and paper rolls 
Self-recording rain gauges (Figure 3) have been applied for continuous rainfall measurements at a selected set of KNMI stations since the end of the 19thcentury. At first, rainfall was recorded on daily (Figure 4) and sometimes weekly rainfall strip charts. Thereafter, from about 1980 through 1993, paper rolls were used to register rainfall for about 10-20 days per roll. From 1994 onwards, rainfall measurements are transferred electronically and operationally stored at 10-minutes resolution (for some selected stations at 1-minute resolution). Until now, the strip charts and paper rolls have been used mainly for extracting hourly values. In infrastructural design (e.g. sewer systems, tunnel drainage) there is, however, a need for long rainfall series with much higher resolution than 1 hour. Fortunately, the charts and rolls can be used to extract rainfall with a time resolution of about 5 to 10 minutes.

We are developing a procedure that largely automates the labor-intensive extraction work for rainfall strip charts and paper rolls. Although developed for rainfall, it can be applied to other elements as well. The procedure consists of four basic steps: (1) scanning of the charts and rolls to high-resolution digital images, (2) applying automatic curve extraction software in a batch process to determine the coordinates of cumulative rainfall lines on the images, (3) visually inspecting the results of the curve extraction, (4) post-processing of the curves that were not correctly determined in step (3). Although KNMI is still perfecting the software, several tens of station-years have succesfully been digitised. The time resolution is about 5 minutes. In total 321 station-years are being digitised using the stations De Bilt, Eelde, Den Helder/De Kooy, Vlissingen, Beek and Amsterdam. When the digitisation is completed, the number of station-years with digital 5-minute rainfall series will be increased by a factor of 25. The data will become available in 2009.

Homogenisation and quality control

Thermometer screen intercomparison
To obtain a homogeneous version of the Zwanenburg/De Bilt temperature series (1706 until present), it is important to know and understand the effects of changes in screen types. One of our activities to study this topic has been to reanalyse the results of a thermometer screen intercomparison that took place at the KNMI terrain in the period 1989-1995. The objective of the analysis was to obtain weather-dependent corrections for past changes in screen types3,4). Ten screens (Figure 5) were compared. All were equipped with fast-responding sensors. Figure 6 presents an example of inter-screen temperature differences for the summer. For some of the screens, the figure clearly shows that the differences strongly depend on the weather. Despite the sometimes large temperature differences on specific days, the annual mean differences for daily maximum, minimum and mean temperature rarely exceed 0.1°C. The strongly aspirated Young screen (Young.aspII) is an exception, with annual mean maximum temperatures 0.28°C below the reference. Comparison of modern-day screens with older (e.g. 19th century) screens, however, reveals annual mean differences of the order of 1.0°C5). For a transition of the natural ventilated synthetic Stevenson screen to the present round-shaped multi-plate operational screen, we successfully developed transfer functions. These functions may be used for correcting inhomogeneities in times series resulting from screen changes.