Sampling for Validation of Ecotope Maps of Floodplains in the Netherlands
Martin Knotters, Dick J. Brus
Soil Science Centre Alterra, Wageningen University and Research Centre Wageningen, the Netherlands
Abstract: Ecotope maps of five districts of main water courses in the Netherlands were validated by independent field observations of ecotopes. The map quality was quantified by the overall map purity, and by the user's and producer's accuracies of the map units. In four districts the validation locations were selected by purposive (targeted) sampling. In these purposively sampled districts the sampling points were clustered in a limited number of compact validation areas, in order to reduce travel costs. For the fifth district a stratified two-stage probability sample was designed, such that the spatial pattern resembles that of the purposive samples. In this way the same practical and budgetary constraints are met as in the purposively sampled districts. In the first stage the district was divided into eight geographical strata, representing the main river branches, and each stratum was divided into validation areas of approximately constant size (primary sampling units). In each geographical stratum two validation areas were selected by simple random sampling without replacement. In the second stage, in each selected validation area a simple random sample of points was selected (the secondary sampling units). At these locations ecotopes were observed in the field. For the maps validated by purposive sampling the quality measures were estimated by model-based inference based on a stochastic model for the spatial variation of classification errors. For the map validated by probability sampling the quality measures were estimated by design-based inference based on the inclusion probabilities of the validation locations. The total map purities varied from 56 to 76 % among the five districts. Both user's and producer's accuracies showed large variation among the map units, depending on the contribution of several sources of error in the mapping process. Stratified two-stage sampling combined with a design-based estimation method results in model-free estimates of total map purity, user's and producer's accuracies. This is an important advantage in validation, because the results do not depend on the quality of model assumptions. This means that the validity of the estimated map purities, user's and producer's accuracies is beyond discussion if a design-based approach is followed.
Keywords: map accuracy; ratio estimator; two-stage sampling