USGS - science for a changing world

Northern Prairie Wildlife Research Center

  Home About NPWRC Our Science Staff Employment Contacts Common Questions About the Site

Improving Prairie Pond Counts with Aerial Video
and Global Positioning Systems

Methods: Statistical Analyses


We performed several analyses to examine accuracy of pond counts from interpreted video images. The most informative analysis was a comparison of the video interpretation of ponds with ground BGS for each pond on air-ground segments. Comparison of individual ponds enabled calculation of omission and commission errors and identification of errors related to determining transect boundaries. This analysis assumed no errors in ground BGS. Video interpretation was identified as correct if greater than or equal to 1 pond was detected in a wetland basin that the ground crew recorded as wet. An omission error occurred if video interpretation failed to detect a pond observed by the ground BGS. We calculated omission error as the number of omission errors divided by number of ponds from the ground BGS. A commission error occurred if the video interpreter identified an object as a pond and a pond was not observed at the location of the object by the ground BGS. We calculated commission error as the number of commission errors divided by total number of video-interpreted ponds. We also recorded boundary error, which occurred when a pond was not counted in either ground BGS or the video interpretation because of an error in determining transect boundaries.

We used data collected from our roadside survey and aerial photographs of the Hendon, Nut Mountain, and Peterson air-ground segments to examine ponds identified as commission or omission errors from comparison of the video interpretation with ground BGS. These ponds may have been incorrectly identified as errors because of a change in pond conditions between the date of video and ground BGS or an error in ground BGS. A pond initially identified as a commission error was changed to a correct video interpretation of a pond if water was observed at the location during the roadside survey or photo interpretation supported the presence of water. A pond initially identified as an omission error was excluded from analysis if interpretation of the 1:12,000 scale photographs did not support the presence of water.

We also calculated omission error for the video interpretation of ponds by comparison with ponds located during our roadside survey. Because less time elapsed between videography and roadside counts than between videography and ground BGS, ponds were less likely to have changed condition.

We used a Chi-square test of homogeneity to determine if accuracy of the video interpretation of ponds was dependent on wetland type and water level. The test compared pond distribution among wetland type and water level for ponds that were identified as correct from the video interpretation with pond distribution among wetland type and water level from ground BGS. Because of small sample sizes, data from streams were excluded, Type IV and V wetlands were combined, and full, flooded, and overflowing water levels were combined for the test of homogeneity. For each cell of the contingency table, we used a Z-test to determine if the proportions of ponds from the video interpretation differed from that of ground BGS.

We used several regression analyses to examine the relationship between the number of ponds from the video interpretation and number of ponds from ground and aerial BGS. We tested the slope and intercept of regression equations for departure from an exact correspondence between video pond counts and BGS pond counts(i.e., slope = 1 and intercept = 0). Because the purpose of regression analyses was not to develop predictive equations, we determined specification of explanatory and response variables by the suspected magnitude of measurement error in pond counts from the different survey methods. We used pond counts from the survey method that we believed had the smallest measurement error as the explanatory variable. We performed regression analyses using pond counts for 3 datasets: (1) video interpretation and ground BGS for both sides of the 12 air-ground segments, (2) video interpretation, ground BGS, and aerial BGS for the 1 side of the 12 air-ground segments used in aerial BGS, and (3) video interpretation and aerial BGS for the 1 side of 30 transect segments used in aerial BGS.

The regression of the number of ponds from the video interpretation on number of ponds from ground BGS for both sides of the 12 air-ground segments used the total number of ponds (includes Type I wetlands). This was the best dataset for examining the relationship of pond counts from video interpretation with pond counts from the ground BGS.

We performed 3 regression analyses using pond counts from the 1 side of the air-ground segment counted in aerial BGS. These regressions enabled a comparison of the relationships among pond counts from video and ground and aerial BGS. We excluded ponds in Type I wetlands from ground BGS pond counts for these regressions because ponds are not counted in aerial BGS. We excluded ponds in Type I wetlands identified by ground BGS and detected by video interpretation from the video pond count.

Pond counts from the 1 side of the 30 transect segments counted in aerial BGS provided the largest sample size for examining the relationship of pond counts from aerial BGS with pond counts from video interpretation. We subjectively excluded ponds in Type I wetlands when interpreting video images.

We used paired t-tests to compare mean pond counts from video interpretation and ground and aerial BGSs. We tested the null hypothesis that the mean difference in pond counts between 2 methods was equal to zero.


Previous Section -- Methods: Analysis of Video Images
Return to Contents
Next Section -- Results: Errors in Video Interpretation of Ponds

Accessibility FOIA Privacy Policies and Notices

Take Pride in America logo USA.gov logo U.S. Department of the Interior | U.S. Geological Survey
URL: http://www.npwrc.usgs.gov/resource/methods/pndcount/statanal.htm
Page Contact Information: Webmaster
Page Last Modified: Saturday, 02-Feb-2013 06:02:48 EST
Reston, VA [vaww55]