Evaluation of bilberry and cowberry yield models


Forest owners and stakeholders need increased amounts of information about forests to support decision making. Besides timber, many forest owners also value forest products and services such as berries, mushrooms, biodiversity, recreation and carbon sequestration. Nowadays forest planning can consider simultaneously several forest products and services. This kind of forest planning requires numerical models to predict the development of different ecosystem services, such as berry yields.

Research on yield models for berries has been particularly active in Finland. However, the evaluation of the existing berry models using an independent data set has not been done so far. The aims of this study, carried out by the researchers of Natural Resources Institute Finland, were to find out 1) how well the models are able to predict the observed berry yields in the study area, 2) whether the models can be applied to locate the most productive berry habitats observed and 3) how sensitive the berry yield predictions are to the input data coming from different forest inventory sources.

Bilberry (Vaccinium myrtillus)

The study was conducted in 230 sample plots in North Karelia, Finland. Different variables were determined for each plot, such as main site class, dominant tree species, earlier cuttings and stand age. The percentage coverages of bilberries and cowberries were determined, and the numbers of bilberries and cowberries were counted. Furthermore, the bilberry and cowberry yields on the sample plots were predicted by four models for bilberry and four models for cowberry. Then it was possible to compare yield models to measures from the sample plots.

The bias and precision of the models varied highly. The average bilberry and cowberry yields based on the measurements in the sample plots were 32.7 kg/ha and 24.0 kg/ha, respectively. The average bilberry and cowberry yields predicted for the sample plots were 8.1-41.7 kg/ha and 2.7-61.5kg/ha, respectively. The predictions of all models were biased, and RMSEs (root-mean-square deviation) were high. According to results, the models for bilberry located the best berry stands more reliably than did those for cowberry. The berry pickers would be guided to pick bilberries in a few very good berry stands but would also be guided to moderate or even poor ones. Both forest inventory data measured in the field and national forest inventory data were utilized in the study. Both data sets were suitable to be used for the berry models.

It must be considered that the data consisted of berry yields only from one berry season (year 2014). Bilberry and cowberry yields in 2014 were slightly worse and better than average yields during the last years. Also weather conditions may affect the berry yields.

Lingonberry (cowberry) (Vaccinium vitis-idaea)

According to the study, all the berry yield models need to be calibrated with field sampled berry yield data collected from the area and time period in hand before applying them in practical applications. All the bilberry yield models could be used to locate the good bilberry stands, but cowberry yield models were not able to reliably detect the good cowberry stands. Instead of predicting the berry yields (kg/ha), the models can be competent in locating potentially good berry picking places.

Source: Kilpeläinen, H., Miina, J., Store, R., Salo, K. & Kurttila, M. (2016): Evaluation of bilberry and cowberry yield models by comparing model predictions with field measurements from North Karelia. Forest Ecology and Management 363: 120-129.


Anni Koskela, Arctic Flavours Association

+358 40 164 6177


päätöksentekoa tukevat ratkaisut  mallin testaaminen  metsien monikäyttö  luonnontuotteet  villit marjat  multiple use forest management  model validation  decision support  non-wood forest products  wild berries