Patterns of greenness (NDVI) in the southern Great Plains and their influence on the habitat quality and reproduction of a declining prairie grouse

Abstract

Patterns of vegetative greenness and timing of greenness events have been a strong predictor of habitat availability and space use for several species of wildlife and may be a particularly useful tool for imperiled grassland species such as the lesser prairie-chicken (Tympanuchus pallidicinctus). I evaluated the utility of Normalized Difference Vegetation Index (NDVI) and NDVI-based vegetation phenology metrics in estimating lesser prairie-chicken habitat availability, habitat quality, and space use during the reproductive season. I captured, marked with GPS and VHF transmitters, and monitored lesser prairie-chicken nest and brood locations during the reproductive season in Kansas during 2013-2015. I acquired Landsat 8 and vegetation phenology metric data from eMODIS Remote Sensing Phenology (RSP) satellite imagery during 2013-2015. Using NDVI and vegetation phenology data at nest and brood locations, I first examined the use of these remotely sensed tools to model habitat selection and predicted habitat availability. I then tested relationships among phenology metrics and reproductive success (e.g., nest and brood success) and assessed timing of nest initiation and hatch relative to patterns of greenness. Last, I investigated correlations between phenology metrics and in-situ vegetation measurements and stocking density. Nest site-selection was best predicted by Time Integrated NDVI (TIN), with probability of use increasing as values of TIN increased ([beta] = 2.897, SE = 1.049). The TIN metric is a proxy for the density of overhead vegetation cover. Brood site-selection was best predicted by an Amplitude (AMP) * Year model ([beta][subscript AMPscale1] = 7.76, SE = 4.81, [beta][subscript 2014] = 0.99, SE = 2.065, [beta][subscript 2015] = -1.78, SE = 2.17, [beta][subscript AMPscale1:2014] = -1.79, SE = 5.12, [beta][subscript AMPscale1:2015] = 6.32, SE = 5.47), with probability of use varying among years but increasing as values of AMP increased. The AMP metric describes the total increase in productivity from the start of the growing season to the peak of the growing season. Areas experiencing greater increases in productivity were more likely to be used by brood-rearing females. To predict nesting and brood-rearing habitat abundance in Kansas, I used a random forest approach. Ultimately, I was unable to predict nesting habitat availability using phenology metrics due to high out-of-bag error (30.48%) and high class error rates, with non-habitat predicted as habitat ~63% of the time. Fortunately, I was able to predict brood-rearing habitat abundance. Informative brood habitat variables selected by the random forest model included the End of Growing Season Time (EOST) at the 1-km scale, TIN at the 1-km scale, AMP at the 370-m scale, percent grassland within 5-km, End of Season NDVI (EOSN) at the 1-km scale, density of county roads within 2-km, density of oil wells within 2-km, Time of Maximum NDVI (MAXT) at the 1-km scale, Start of Growing Season Time (SOST) at the 250-m scale, and the density of transmission lines within 2-km. Using the selected variables, I identified priority habitat using the Kappa threshold and high priority habitat using the Sensitivity Specificity Sum Maximizer threshold. Habitat availability was variable between years, with a 71% and 51% decrease in priority and high priority habitat, respectively, from 2014 to 2015. I identified 2,154,137.5 ha of priority habitat and 8,225 ha of high priority habitat for 2014. I identified 636,493.75 ha of priority habitat and 3,993.75 ha of high priority habitat for 2015. Nest survival was best predicted by MAXT at the 500-m scale, with nest survival maximized when MAXT was Day-of-Year (DOY) 160 (June 9) and decreasing linearly as MAXT increased ([beta] = -0.009, SE = 0.004). Similarly, I identified phenological differences at successful and unsuccessful nest and brood sites. At successful nest sites, TIN was greater than at unsuccessful nests (p = 0.05), and MAXT occurred earlier than at unsuccessful nests (p = 0.04). At successful brood sites, MAXT occurred later and EOSN was greater than at unsuccessful brood sites (p = 0.003). The EOSN metric was also significantly different, with EOSN greater at successful brood sites than at unsuccessful brood sites (p = 0.01). Timing of nest initiation and hatch relative to patterns of greenness indicated that first nests were initiated within ~20 days of SOST. All hatch dates occurred before the peak of the growing season date (MAXT). Ultimately, lesser prairie-chickens time nest initiation and hatch between the start of the growing season and peak of the growing season. I also tested correlations among vegetation phenology metrics to in-situ vegetation measurements and stocking densities. Correlations with phenology metrics and in-situ vegetation measurements varied among years, but TIN and AMP were often positively correlated with measures of visual obstruction at multiple scales and cover of forbs and grasses (r = 0.02 – 0.51). The TIN, AMP, and Maximum NDVI (MAXN) metrics were often negatively correlated (r = -0.02 – -0.15) with cover of bare ground, litter depth, litter cover, and shrub cover. Last, I evaluated linkages between vegetation phenology metrics and cattle stocking density. Correlations varied among years (2014 and 2015). The TIN and AMP metrics were positively correlated with stocking density in 2014 (r = 0.13, r = 0.07, respectively); yet TIN was negatively correlated with stocking density in 2015 (r = -0.17) and AMP was not correlated with stocking density in 2015. Ultimately, I provide evidence that NDVI-based vegetation phenology metrics can be used to model habitat use and predict habitat availability for lesser prairie-chickens in Kansas. My predictions from phenology-based metrics indicated that the availability of high priority habitat may be limited. I also provided evidence that phenology metrics correlate to in-situ vegetation measurements and stocking densities, making phenology metrics a promising tool for monitoring lesser prairie-chicken habitat remotely.