The northerly extreme of the mixed grass prairie is predicted to be extremely sensitive to a changing climate. Such changes will likely be hastened through the interaction of climate warming with other commonly occurring grassland disturbances. The changing community composition of northern prairie is expected to have significant effects on the populations of prairie fauna, such as the Greater Sage Grouse (Centrocerus urophasianus). In Canada, Sage Grouse are found almost exclusively within the Silver Sagebrush range of the semi-arid mixed-grass prairie regions of southeastern Alberta and southwestern Saskatchewan. However, the population of Sage Grouse in these Provinces has declined significantly over the past few decades. This decline has mainly been attributed to the loss of suitable habitat, resulting in reduced nesting success.

Restoring Canadian Sage Grouse populations requires the identification of critical habitat features such as forb cover, at landscape scales. Forb cover is an important food resource for Sage Grouse. We assessed the ability of spectral measurements to predict forb cover in Canadian prairie. We collected forb cover and spectral data from sampling plots (0.5 m) in Grasslands National Park, Saskatchewan, in June 2004, then used a spatially nested sampling design to scale these observations “up” to coarser observational scales (10 m and 30 m). Our results showed that (a) the ability to predict forb cover using spectral observations is temporally-dependent; (b) the near infrared and shortwave infrared wavebands are better predictors of forbcover than the spectral vegetation indices tested; (c) forb cover is best predicted over grass-dominated communities; and (d) the predictive ability of all spectral approaches double as data are aggregated to coarser sampling resolutions. The temporal dependence in forb cover predictability is due to differences in plant phenology and surface water conditions among sampling dates. The less accurate predictions of forb cover in shrubbed communities occur because shrub species dominate the spectral signal recorded by the radiometer. Refinements to this approach may permit land managers to (a) identify sites for recovery and reintroduction based on forb density potential as well as linking the fate of extant Sage Grouse metapopulations to features of their critical habitat, and (b) assess the longer-term effects of climate on Sage Grouse recovery, by allowing climate-driven changes in forb cover to be monitored, and the identification of sites that are declining or not declining in productivity and forb density.

AbstractAbstract

� The changing community composition of northern prairie is expected to have significant effects on the populations of prairie fauna such as Greater Sage Grouse (Centrocerus urophasianus).

� The population of Sage Grouse in Alberta and Saskatchewan has declined significantly over the past few decades. This decline has mainly been attributed to the loss of suitable habitat, resulting in reduced nesting success.

� Because Sage Grouse survival is intimately tied to Sagebrush, attempts to map the critical habitat of Sage Grouse have generally been limited to studies of Sagebrush density and structure. However, there have been no attempts to map other critical habitat features to which Sage Grouse may be responding.

� One such feature is forb cover, which is important because it (a) is an important food resource for Sage Grouse and (b) plays a structural role in home range selection by Sage Grouse.

� Here, we assess the ability of spectral measurements to predict forb cover in Canadian prairie. We hypothesize that since forb cover in upland grassland in southern Saskatchewan has been shown to correlate relatively well with sward productivity, productivity-related spectral indices may be well correlated with forb abundance.

1. Introduction and 1. Introduction and AImsAIms

The assessment of critical The assessment of critical forbforb habitat of Greater Sage Grouse habitat of Greater Sage Grouse ((Centrocerus urophasianusCentrocerus urophasianus) in northern prairie using field and ) in northern prairie using field and remote sensing dataremote sensing data

A. DavidsonA. Davidson1 1 , S. Wang, S. Wang11, , J. WilmshurstJ. Wilmshurst22 and J. Thorpeand J. Thorpe33

1 1 Canada Centre for Remote Sensing, Natural Resources Canada, OttaCanada Centre for Remote Sensing, Natural Resources Canada, Ottawa, ON, Canada.wa, ON, Canada.2 2 Parks Canada Agency, Western Canada Service Centre, Winnipeg, MBParks Canada Agency, Western Canada Service Centre, Winnipeg, MB, Canada., Canada.33 Saskatchewan Research Council, Saskatoon, Saskatchewan, Canada, Saskatchewan Research Council, Saskatoon, Saskatchewan, Canada, S7N 2X8S7N 2X8

This project was funded by the This project was funded by the Interdepartmental RecoveryInterdepartmental Recovery Fund (IRF) (Grant 379).Fund (IRF) (Grant 379).

2. Study Site and Sampling Strategy2. Study Site and Sampling Strategy

Figure 1. Grasslands National Park and sample site locations

UUG

UAG UAL

UAH

GUG

GAG

GAL

U ngraze d Alluvial Lo w-density shrubla nd (U A L)

Ungrazed A lluvia l H igh-density shrubla nd (U AH )

Grazed Alluvial Grass land (GAG )

U ngrazed Upland Grassland (U U G)

Un grazed Alluv ial Grass land (U AG ) Grazed Upland Grass land (GUG )

G raze d Alluvial Lo w-d ensi ty shrubla nd (G A L)

Figure 2. Sample locations in

ungrazed and grazed areas of GNP.

Ungrazed Upland Grassland (UUG) Ungrazed Alluvial Low-density Shrub (UAL) Ungrazed Alluvial High-density Shrub (UAH)

Ungrazed Alluvial Grassland (UAG) Grazed Upland Grassland (GUG) Grazed Alluvial Grassland (GAG)

Grazed Alluvial Low-density Shrub (GAL)

3. Statistical Methods3. Statistical MethodsParameter selection for regression models� The combination of individual spectral bands and derivatives to be used to predict water content were

selected using the bootstrap approach for model selection described by Olden and Jackson (2000).

� Separate regressions were carried out using (a) the selected combination of individual bands; (b) the selected combination of derivatives, and (c) individual spectral vegetation indices.

� The best approach was identified using a leave-one-out cross-validation approach.

Validation� Each model was validated using a leave-one-out cross-validation approach (Olden and Jackson,

2000). This approach has been shown to be superior to split-sample validation, particularly for small sample sizes (Goutte 1997).

� The predictive ability of each model was characterized using (a) the root-mean-square error of prediction (RMSE), and (b) the correlation (r) between the observed and predicted responses (also called the cross-validated r).

� Plant biophysical and spectral data were collected during June 2004 at seven sites within and around GNP (Fig. 1).

� At each of the seven sites, 72 0.5 x 0.5m sampling plots were arranged in a nested design (total n = 504). We used a nested sampling scheme because it allows plot-level observations to be scaled to coarser sampling resolutions (Davidson and Csillag 2001).

� Forb cover estimates were derived directly from non-destructive measurements taken in these plots using a modified Daubenmire procedure (Daubenmire 1959).

� Spectrally reflected radiation in Landsat Thematic Mapper (TM) bands 1 (0.45-0.52µm), 2 (0.52-0.60µm), 3 (0.63-0.69µm), 4 (0.76-0.90µm) and 5 (1.55-1.75µm) was measured over each plot on three separate occasions during the sampling period using a Cropscan Model MSR5 Multispectral Radiometer (0.5 m spatial resolution).

� Spectral data was collected on three occasions so that a range of surface conditions could be adequately sampled. These conditions corresponded to (a) early June, when forbs were just starting their annual growth cycles and the upper soil layers were dry; (b) mid-June, when forb growth had increased, but heavy rains had saturated the upper soil layers, leaving standing water in places; and (c) late June, whenforb growth was at its greatest and the upper soil layers were again dry.

� All spectral measurements were collected within 2h of solar noon under cloud-free sky conditions. We used the mean of three separate reflectance measurements, each taken 5s apart, as a representative measure of plot reflectance.

� Plot reflectances were then transformed into various vegetation indices. These indices were chosen because they have been shown to correlate best to live biomass and vegetation water content of grassland-shrubland communities in the GNP region (Davidson, 2002; Davidson et al. in press).

0 5 10 15 20 2 5

0

5

10

15

20

25

Obs

erve

d (%

)

(A) 0 .5m

P redic ted (% )

0 5 10 15 20 25

0

5

1 0

1 5

2 0

2 5

Obs

erve

d (%

)

(C) 30m

Predic ted (% )

Figure 3. The scale dependence of the correspondence between predicted andobserved forb cover for the best Band Combination Approach (using TM bands 4

and 5).

0 5 10 1 5 20 25

0

5

10

15

20

25

Obs

erve

d (%

)

(B) 10m

Predic ted (% )

4. Results4. Results� The results of our 0.5 m, 10 m and 30 m-resolution studies highlight four important trends:

(a) that the ability to predict forb cover using spectral observations depends on the time of spectral data acquisition;

(b) that the “best” Band Combination Approach (using TM bands 4 and 5) predicts forb cover of grassland-shrubland communities with greater accuracy and precision than the “best”Vegetation Index Approach (using the NDII, MSI or GVMI);

(c) that forb cover is predicted more accurately over grassland dominated-communities than shrubland-dominated communities, but predicted only slightly more accurately over grazed communities than non-grazed communities; and

(d) that the predictive ability of the Band Combination and Vegetation Index approaches double when plot-level data are aggregated to 10 m and 30 m sampling resolutions.

5. Conclusions5. Conclusions� Research presented here is encouraging for the prospect of monitoring and mapping forb cover of

northern prairie grassland-shrublands using medium-resolution multispectral satellite data.

� Because forb cover is a critical component to Sage Grouse fledgling survival, a spectrally-based spectral monitoring tool will aid in the identification of Sage Grouse critical habitat at landscape scales and allow land managers to identify sites for recovery and reintroduction based on their forb cover potential, and attribute the fate of extant Sage Grouse metapopulations to features of their critical habitat.

� The ability to predict forb cover from spectral data may provide a way for assessing the longer-term effects of climate on Sage Grouse recovery. Our spectrally-based spectral monitoring tool may allow for climate-driven changes in forb cover to be monitored, and the identification of sites that are declining or not declining in productivity and forb density.

� Daubenmire, R. 1959. A canopy coverage method of vegetation analysis. Northwest Science. 33:43-64.

� Davidson A and Csillag F. 2001. The influence of vegetation index and spatial resolution on a two-date remote sensing derived relation to C4 species coverage. Remote Sensing of Environment. Vol 75(1), pp138-151.

� Davidson A, Wang S and Wilmshurst J. Remote sensing of grassland-shrubland vegetation water content in the shortwave domain. International Journal of Applied Earth Observation and GeoInformation(in press).

� Davidson A. 2002. Integrating field sampling and remotely sensed data for monitoring the function and composition of northern Mixed Grass prairie. A thesis submitted in conformity with the requirements for the Degree of Doctor of Philosophy, Graduate Department of Geography, University of Toronto. National Library of Canada. Ottawa, Canada.

� Olden, J.D., and Jackson, D.A. (2000), Torturing data for the sake of generality: How valid are our regression models? Écoscience, 7(4), 501-510.

� Goutte, C. (1997), Note on free lunches and cross-validation. Neural Computation, 9, 12111-11215.

6. References6. References