Distance sampling with vehicle-mounted cameras for the estimation of farmland bird abundance over a wide area

Abstract Agricultural birds are declining due to farmland abandonment and agricultural intensification, and monitoring approaches, including efficient survey methods, are crucial for the development of effective conservation strategies. Roadside surveys (RS), especially those using vehicle-mounted video cameras, can be used to efficiently search for targets; however, they are limited by the unknown detection range. Distance sampling (DS), in which animal density or abundance is estimated from the distance between the observer and the detected individual, can provide a complimentary approach to address this limitation of RS. We developed a DS model robust to location uncertainty based on video-based RS. We integrated location errors determined by an independent field test using RS and unmanned aerial vehicles (UAVs) into a stochastic DS model. We estimated the abundance of herons and egrets in farmlands in and around the evacuation zone of the Fukushima Daiichi nuclear power plant accident by combining RS and DS. The video-based RS covered 7,031 km of roads and 24.41 km 2 of farmlands. All herons and egrets were observed around the evacuation zone and none were detected in the zone. The predicted abundance of herons and egrets differed substantially between areas inside (0.0158 ± 0.0174/km 2 ) and outside of the zone (4.62 ± 5.41/km 2 ). Integrating location uncertainty into the DS model did not affect predicted heron and egret population densities (4.62 ± 5.41 vs. 4.66 ± 5.45/km 2 , with or without integrating location uncertainty, respectively). Accordingly, our survey method combining RS and DS is robust to location uncertainty. The study system (i.e., herons and egrets in farmlands) and the inability of location error to exceed the size of one farmland separated by levees may contribute to the accuracy of RS in this study. Synthesis and Applications: Combining RS with vehicle-mounted cameras and DS considering location uncertainty is widely applicable to open land species and can improve the efficiency of monitoring. When large location errors are expected by conventional approaches, DS models incorporating location uncertainty determined by RS and UAVs could be effective for abundance estimation over wide areas.