Skip to main content
ARS Home » Research » Publications at this Location » Publication #139447

Title: TEST OF EQUATION ROBUSTNESS TO PREDICT FORAGE QUALITY FROM REMOTELY SENSED DATA

Author
item Starks, Patrick
item Brown, Michael
item APPEDDU, LISA - SW OKLA STATE UNIV

Submitted to: Society for Range Management Meeting Abstracts
Publication Type: Abstract Only
Publication Acceptance Date: 2/2/2003
Publication Date: 2/2/2003
Citation: STARKS, P.J., BROWN, M.A., APPEDDU, L.A. TEST OF EQUATION ROBUSTNESS TO PREDICT FORAGE QUALITY FROM REMOTELY SENSED DATA. SOCIETY FOR RANGE MANAGEMENT MEETING ABSTRACTS. 2003. v. 56. p. 106.

Interpretive Summary:

Technical Abstract: In an earlier study, calibration equations were developed to predict forage quality from hyperspectral reflectance data collected from monocultures of Midland, Midland 99, Worldfeeder and Ozarka bermudagrasses (Cynodon dactylon L. (Pears.). To test equation robustness, we applied the previously developed equations to four 1.6 ha mixed grass species paddocks consisting of senescent downy brome (Bromus tectorum L.), yellow bristlegrass (Setaria glauca (L.) Beauv.), and common bermudagrass. Once per week from late May to early August, an SE-590 hyperspectral radiometer was used to collect spectral reflectance data from 8 locations along a transect in each of the four paddocks. Forage samples were collected from each of the 8 sample locations in each paddock, and percent crude protein (CP), neutral detergent fiber (NDF), and acid detergent fiber (ADF) were determined using conventional laboratory techniques. Estimates of forage quality from the hyperspectral data were compared to laboratory values using conventional mixed model repeated measures analyses with a linear model that included effects of pasture, prediction method, time, and appropriate interactions. Analyses for CP gave little evidence of differences between prediction methods (P > .55), nor was there evidence of an interaction of method with time of estimate (P > .37). Predictably, there was strong evidence of time differences in CP (P < .01). Analyses for NDF were similar to CP with little evidence of differences between prediction methods (P > .98) or method by time interaction (P > .5); time differences were evident in NDF (P < .05). Analyses for ADF suggested differences between prediction methods were consistent over time (P > .33), but there was a trend for remotely sensed ADF to be larger than laboratory reference values (P < .11). These data suggest that remotely sensed estimates of percent CP and NDF accurately reflect laboratory reference values, but equations for percent ADF may need further refinement for more general applicability.