Get reliable science

Remotely sensed vegetation indices are increasingly used to characterize vegetation growth dynamics and inform rangeland management decisions. Existing predictive models of livestock reproductive performance on rangelands have mostly relied on precipitation data. The value of vegetation green-up parameters as predictors of livestock reproductive performance has not been tested for New Mexico shortgrass rangelands. Our objectives were to explain and predict sheep reproductive performance using satellite-derived green-up parameters and to compare them to precipitation-based models for a semiarid rangeland in central New Mexico. A ten year (2003-2012) time series of 250 m 16 day MODIS Terra and Aqua EVI and NDVI images was processed in Timesat 3.1 to extract 11 green-up parameters (phenometrics) for each year. Annual averages of phenometrics and spring, fall, and growing season precipitation were used as predictors to model marking rate (lambs/ewes around July 1), weaning rate (lambs/ewes around October 1), and lamb weaning weights. Multiple regressions and adjusted R² model selection were used on reduced subsets of predictors to build candidate retrospective and prospective models with current or previous year predictors, respectively. Top models were selected from candidate models using combined criteria (P<0.1, AICc, VIF, residuals, and influential point analysis). We compared models including either phenometrics- or precipitation-based predictors. Green-up parameters were easier to interpret and explained a higher proportion of variation in marking and weaning rates compared with models that only included precipitation. Marking rate was positively correlated with greenness baseline value derived from EVI images (R²=0.47; P=0.03,) while weaning rate was negatively correlated with the greenness peak date derived from NDVI images (R²=0.54; P=0.08). Previous year senescence date and greenness baseline value derived from NDVI images were positively correlated with current year's marking rate (AdjR²=0.57; P=0.03). Our study suggests that phenometrics-based models may be a promising tool to predict livestock reproductive performance on rangelands.