Quantifying sample biases of inland lake sampling programs in relation to lake surface area and land use/cover

We quantified potential biases associated with lakes monitored using non-probability based sampling by six state agencies in the USA (Michigan, Wisconsin, Iowa, Ohio, Maine, and New Hampshire). To identify biases, we compared state-monitored lakes to a census population of lakes derived from the National Hydrography Dataset. We then estimated the probability of lakes being sampled using generalized linear mixed models. Our two research questions were: (1) are there systematic differences in lake area and land use/land cover (LULC) surrounding lakes monitored by state agencies when compared to the entire population of lakes? and (2) after controlling for the effects of lake size, does the probability of sampling vary depending on the surrounding LULC features? We examined the biases associated with surrounding LULC because of the established links between LULC and lake water quality. For all states, we found that larger lakes had a higher probability of being sampled compared to smaller lakes. Significant interactions between lake size and LULC prohibit us from drawing conclusions about the main effects of LULC; however, in general lakes that are most likely to be sampled have either high urban use, high agricultural use, high forest cover, or low wetland cover. Our analyses support the assertion that data derived from non-probability-based surveys must be used with caution when attempting to make generalizations to the entire population of interest, and that probability-based surveys are needed to ensure unbiased, accurate estimates of lake status and trends at regional to national scales.