A . Example map. A typical landscape in Southern Estonia with different types of green area indicated by different colours; pixel size is 100x100 m.

B . Sum of model weights for models relating atopic sensitization with combinations of green area types at different spatial scales. Fifteen models were constructed at each scale combining green area types in various ways. Models at the scale of 8 km produced highest weights.

C . The importance of different green area types at various spatial scales in models explaining atopic sensitization. Importance was calculated as the summed Akaike weight from models at a particular spatial scale where the green area type was included. 95% bootstrap confidence intervals are shown around mean importance.

D . Predicted probabilities with confidence intervals showing the likelihood that a child from a home surrounded by a certain amount of biodiverse nature will be classified to a particular atopic sensitization group. Ordinal regression model linked three groups of subjects based on atopic sensitization to the summed area of forests, grasslands and wetlands within a radius of 8 km of the subject home location (presented here as the proportion of the circle for better readability; the model used log-transformed values).

Table 1. Summary of best models from each spatial scale where atopic sensitization was linked to the areas of green habitat types. Each row represents a spatial scale from 1 to 10 km. The overall best model is in bold (8 km radius)

Estimate Std. Error z value Pr(>|z|)

(agricultural1 + forest1 + wetland1) 0.1080 0.0424 2.5467 0.0109

(agricultural2 + forest2 + wetland2) 0.1345 0.0526 2.5549 0.0106

(forest3) 0.1939 0.0694 2.7943 0.0052

(forest4) 0.2380 0.0880 2.7051 0.0068

(forest5) 0.2956 0.1092 2.7081 0.0068

(forest6 + wetland6) 0.3476 0.1287 2.7004 0.0069

(forest7 + wetland7) 0.4521 0.1624 2.7843 0.0054