Performance Assessment Of GPS Collar Over Mountainous Region

GPS collars have been widely used for large and medium-sized wildlife monitoring.However,the data returned by GPS collars may contain positioning errors and missing bias in positioning data,which could impact associated ecological studies(e.g.,daily travel distance,habitat selection,home range,etc.).Especially,missing bias and location error of GPS location are large in mountainous terrain.As a result,evaluating positioning performance is critical when employing GPS collars to monitor animals.In this study,we assessed the performance of GPS collars over rugged terrain in Northwest Yunnan,China.Firstly,we analyzed the influence of topography on positioning performance.Secondly,we proposed an approach to removing abnormal positioning data using a random forest algorithm.Finally,we removed abnormal GPS fixes from a grazing sheep based on estimated elevation error,and tested by maximum nocturnal distance and monthly home range.Firstly,We set two stationary points on the ridge and another two stationary points in the valley of Cangshan Mountain,Dali,Yunnan.All these four points were set up at the same altitude above sea level(approximate 2600 m).In addition,to measure the performance of GPS collars under ideal conditions,we also placed two GPS collars over an open site.The positioning performance varied according to different topographic positions.First,the fix success rate ranged from 21.7% to 21.9% in the two valley points,increased to 29.2%-37.8% on the ridge points,and reached 79.2% in the open site.Second,the location errors of the valley points were much larger than those of the ridge points.The 50% circular error probable(CEP)of the 3D fixes were 21 m and28 m at the valley points,13 m and 20 m at the ridge points,and 5 m at the open site.However,the 90% CEP of the 3D fixes were 843 m and 2837 m at the valley points,95 m and 690 m at the ridge points,and 14 m at the open site.Third,the precision of GPS collars in the valley was 56.5 m and 106.3 m,while on the ridge of the mountain it was21.6 m and 40.9 m,and only 6.9 m in the open site.Our results suggest that the performance of the GPS collar was severely influenced by the rugged terrain.Thus,it is essential to evaluate both the missing data bias and the location error of GPS fixes over mountainous regions before performing an ecological analysis with the obtained GPS data.Secondly,we proposed an approach to removing anomalous data from GPS collars using a random forest algorithm.The results showed that the GPS fixes having larger location error could be recognized.Furthermore,we also verified that the estimated elevation error could explain 36-74% of the variation of the horizontal positioning error.Finally,to verify the effectiveness of estimated elevation error(EEE)in rejecting anomalous data in practical applications,a GPS collar was placed on a sheep in the Rasha Mountains,and positioning data were collected continuously for a whole year.The sheep's 3D data were screened for outliers using PDOP and EEE,and the changes in daily distance and monthly home range were compared before and after screening out the outliers.The findings revealed that 1)the monthly home range of sheep decreased from 27.6 km2 to a reasonable range(mean value 3.06)after screening out the anomalies by EEE,and 2)the distance of sheep's nocturnal activity declined significantly and stabilized after screening out the anomalies by EEE.In practical application,EEE can screen out the anomalous data with larger errors more effectively than traditional methods.In summary,when using GPS collars in mountainous regions,the influence of terrain factors on the positioning performance of GPS collars should be considered,and for the returned positioning data,it is suggested to use the estimated elevation error method(EEE)to screen out points with large horizontal positioning errors,which can efficiently improve the accuracy and precision of positioning data.