Study on climate change in German television programming: Fraunhofer IDMT analyzes 20 TV channels

Press Release /

On October 24, 2023, the German actress and co-founder of the MaLisa Foundation, Maria Furtwängler, presented the results of the study "Climate change and biodiversity: What does television show - what do viewers want?" in Munich. The study was initiated jointly by the MaLisa Foundation and the four German major television stations ARD, ZDF, ProSiebenSat.1 and RTL Deutschland. The study investigated whether there is a discrepancy between the presentation of the climate crisis and biodiversity in German television programs and the perception of these topics by viewers. The Fraunhofer Institute for Digital Media Technology IDMT was responsible for the audio and video analysis of the television program as a research partner in the one-year project.

Screenshot der Audiovisual Identity Suite zeigt mit Heat Map, wie oft eine bestimmte Persönlichkeit im Erhebungszeitraum bei welchem Fernsehsender präsent war.
© Fraunhofer IDMT
A heat map was used to show how often a particular personality was present on which television channel during the survey period.

The Fraunhofer IDMT conducted the TV study together with two social scientists from Freie Universität (FU) Berlin and Ludwig-Maximilians-Universität (LMU) Munich. Various survey methods were combined for the evaluation. The Fraunhofer IDMT examined the programming of 20 German TV stations over a period of two months using automated content analysis. Professors Irene Neverla (FU Berlin) and Imke Hoppe (LMU Munich) conducted qualitative group discussions and panel interviews with around 2,600 people.

The research questions

Among other things, the study aimed to find out

  1. which representatives from politics, science, business, law, and culture are visible in the climate discourse on TV
  2. the proportion of men and women in the climate discourse is (compared to the non-climate discourse) and
  3. the thematic contexts in which climate change and biodiversity are discussed on television.

For the program analysis, the broadcasts were examined daily from 05:30 to 0:30 in the period from September 1 to October 31, 2022. For this purpose, Fraunhofer IDMT recorded the programs EPG-controlled and then analyzed them automatically. The gender of individuals was determined based on their facial and vocal characteristics and the recordings were then transcribed to assess the contextual content of television programs.

EPG means Electronic Program Guide and refers to electronically distributed information such as the title, time and duration of radio and television programs.

Identification of visible personalities in the climate discourse

To analyze which personalities are visible in the TV discourse, 146 people from politics, science, business, law and culture were first defined as reference persons. The complete video material was then searched using intelligent facial analysis. In a further analysis step, the faces found were classified according to the perceived gender and a facial comparison was carried out with the 146 reference persons. The results of this analysis will soon be presented in a scientific paper. 

To enhance gender identification accuracy, the data was also subjected to an auditory speaker recognition. This combined approach not only increases the accuracy of gender recognition but can also solve special cases in which a person is not visible but speaks.

A total of 37,476 items were analyzed and 4,655,706 different faces of 806,390 different people were recognized. The comparison of the facial data with the 146 predefined reference persons worked very reliably.

27.9 percent of all faces were classified as visually female and 72.1 percent as visually male. Through the additional comparison with the results of the auditory speaker recognition, the percentage distribution could be corrected to 35 percent female and 65 percent male.

This means that a cross-modal analysis of audio and video data enables a higher level of significance and significantly more accurate results in the identification of people in television programs.

Content analysis of the television programs

For analyzing the content of the recorded television broadcasts on the topics of climate and biodiversity, the language contributions were analyzed with the help of a trained speech recognition system and converted into evaluable text. The speech recognition system used was trained with over 17,000 hours of speech material and, specifically for this study, with over 40 terms relating to climate change and biodiversity. These included terms such as climate change, global warming, greenhouse effect, CO2 emissions, climate summit, biodiversity, species diversity and extinction. In an initial sample of 370 randomly selected broadcast minutes, the reliable and robust recognition of the keywords in the transcribed texts was tested. This was almost 100 percent.

As a result, despite the reliable analysis methods, the relative share of topics on climate change was still only 1.8 percent of all evaluated broadcast minutes, while the share on biodiversity was only 0.2 percent. 

Evaluation of the results

To ensure that the automatic analysis of the programs worked reliably, the Fraunhofer IDMT evaluated the developed algorithms regarding the error measures. Manually annotated data was used for comparison and, if necessary, the algorithms were improved.

Fast analysis of large amounts of program data

The research institute used five satellite transponders distributed across four PC workstations to record the program data. In total, the Fraunhofer IDMT recorded, transcribed, and analyzed around 1.4 million broadcast minutes and made them available in a dashboard for further evaluation steps – this corresponds to around 2.7 years of broadcast material and 63 terabytes of program data.

Project manager Dr.-Ing. Christian Weigel from Fraunhofer IDMT is very pleased with the reliability of the cross-modal analysis methods used: "Our analysis tools perform robustly and fast. As part of the MaLisa study, we were able to analyze very large amounts of data in a short time and thus provide valuable information for the MaLisa study".

You can find all the results of the current MaLisa study at:

Last modified: