Mark Hebblewhite, Evelyn Merrill, and Greg McDermid. 2008. A multi-scale test of the forage maturation hypothesis in a partially migratory ungulate population. Ecological Monographs 78:141–166.

Appendix B. Modeling forest maturation using NDVI and ground maturation models.

Herein a summary is provided for: (1) MODIS-NDVI data acquisition and processing details, (2) NDVI-time series smoothing approaches, and a (3) evaluation of the relationship between the MODIS NDVI index in relation to open habitat herbaceous biomass in BNP for the development of a forage growth/maturation model for open habitats.

NDVI data for the 2004 growing season was acquired by the Moderate Resolution Imaging Spectrometer (MODIS) on board NASA’s Terra and Aqua satellites. The data comprise a portion of the MOD-13 Vegetation Indices suite of Huete et al. (2002): a series of high-quality data products available to researchers through the MODIS Land Products program (Justice et al. 2002). Specifically, we selected the 16-d, 250-m NDVI product (MOD13Q1), derived from a series of composite imagery selected across 16-d time intervals spanning from 22 April to 30 October. MODIS sensors image any given portion of the earth twice per day, so the compositing process draws on as many as 32 individual scenes per 16-d time interval, selecting pixels designed to minimize the impacts of atmospheric contamination, off-nadir viewing geometry, and other sources of error (Huete et al. 2002). The MODIS sensors offer substantial technical improvements over alternative AVHRR imagery which, combined with more sophisticated processing and compositing algorithms, results in data sets that are significantly improved over previous-generation products from the Pathfinder program (Huete et al. 2002).

NDVI measures the normalized ratio of the near-infrared (r_NIR) and red reflectance (r_NRR) where NDVI = (r_{NIR -} r_NIR) _/r_{NIR +} r_NIR) and ranges from –1 to 1, where negative values represent absence of vegetation (Tucker and Sellers 1986). We removed all values <0 so that NDVI represented only vegetation productivity ranging from 0–1. Composite images were downloaded from NASA’s Land Process Distributed Active Archive Center (MODIS scenes H10v03 and H11v03) in the integerized sinusoidal projection. Image tiles were mosaicked and re-projected to UTM zone 11 (NAD83) using the MODIS Reprojection Tool. MODIS Science Data Sets (SDS), including NDVI values, cloud cover estimates, snow masks, and the combined quality data field, were extracted from NASA’s Hierarchical Data Format into ESRI ArcMap and clipped for analyses.

TABLE B1. MODIS 16-d composite intervals for 2004.

Interval	Start date	Julian-mid date	Interval	Start date	Julian-mid date
1	22 April	121	7	27 July	217
2	8 May	137	8	12 August	233
3	24 May	153	9	28 August	249
4	9 June	169	10	13 September	265
5	25 June	185	11	29 September	281
6	11 July	201	12	15 October	297

Despite improved compositing approaches (Huete et al. 2002), substantial errors remain in the MOD 13 NDVI time-series products, reflecting the impacts of clouds, snow, and other external factors which bias NDVI low, or rarer extreme-off nadir values which bias NDVI high. Numerous approaches have been developed for smoothing NDVI time-series (reviewed in Pettorelli et al. 2005). Time series used in this study were smoothed using an adaptation of Kawamura et al. (2005) ’s temporal window operation (TWO) approach. First, we identified a threshold MODIS – quality data field (Huete et al. 2002) value above which NDVI was no longer negatively biased to identify poor quality pixels using linear regression. We found this threshold to be the quality value of 7 (below average quality according to Huete et al. (2002). On average, 22% (range 0.05% to 63% on June 25) of all composite images had quality thresholds >7 (higher values represent poorer quality). Thus, we removed pixels with quality field >7 and used linear interpolation to calculate NDVI values for these poor-quality pixels (Chen et al. 2004). Second, from the start of any temporal sequence of NDVI values we assumed plant growth increased to some peak and then declined. Large declines in NDVI in this sequence were identified using the guideline of removing, and interpolating, any decline >25% (Kawamura et al. 2005; Jobbagy et al. 2002). Figure B1 shows an example of this two-pass (Kawamura et al. 2005) noise-reduction approach with first, the removal of poor quality pixels (e.g., 8 May), and second, the correction of remaining anomalously low NDVI values (e.g., 25 June). Following noise reduction, final NDVI layers were produced for 16-d composite time periods during the growing season starting 22 April until the final start date of 15 October – 31 October (Table B1). In a strategy similar to the statistical forage biomass maturation model (see Appendix A), for each pixel we calculated the % of maximum NDVI to derive the percent adjustment for rescaling forage biomass in open habitats (see Appendix C).The relationship between forage biomass and NDVI was then determined for the entire growing season (Fig. B2, Table B2). In addition to simple linear regression between NDVI and forage biomass, we included the covariates of elevation and distance to continental divide in multiple linear regression following Thoma et al. (2002). For individual regressions between forage quality and forage biomass for each 16-day MODIS interval, see Appendix D.

FIG. B1. Example of the two-step smoothing algorithm used to improve NDVI time-series pixel quality for repeat sample plot S3, an alpine grassland in Snow creek, Banff National Park, from April to October 2004. Raw refers to the raw NDVI data, Quality means just the quality screen of a score of 7 was used to clean the data, and the final time series includes the decision rule to interpolate large declines in NDVI. See text for details.

FIG. B2. Relationships during the growing season May – Oct 2004 between MODIS-NDVI and (a) total standing biomass and total green biomass and (b) total forb and graminoid biomass (g/m2) from open habitats in BNP, Alberta.

TABLE B2. Relationships between MODIS-NDVI and open habitat ground biomass plot through the growing season, 1 May to 30 September 2004, BNP, Alberta.

Model	N (plots)†‡			0 (SE)	NDVI (SE)	ELEV (SE)	DIST (SE)	Pearson's-rS	r2	P	AICc‡
Total biomass
Linear		45(11)		-20.0 (17.47)	128.9 (27.30)	--	--	0.59	0.34	<0.005	1.9
Multiple		40(9)		135.8 (33.15)	84.5 (23.78)	-0.075 (0.015)	0.370 (0.189)	0.86	0.75	0.006	0
Green biomass
Linear		33(11)		-28.6 (12.73)	122.0 (25.57)	--	--	0.65	0.42	0.0008	1.35
Multiple		40(9)		66.7 (31.76)	115.7 (24.59)	-0.053 (0.014)	0.233(0.126)	0.80	0.65	0.003	0
Forb
Linear		45(11)		-22.2 (10.38)	79.5 (56.23)	--	--	0.60	0.36	<0.005	1.06
Multiple		40(9)		4.4  (25.70)	68.3 (20.87)	-0.016 (0.012)	0.21 (0.100	0.67	0.45	0.028	0
Graminoid
Linear		45(11)		-1.30 (11.15)	56.23 (20.38)	--	--	0.33	0.12	0.021	0.5
Multiple		40(9)		128.6 (28.35)	23.2 (15.6)	-0.06 (0.019)	0.18 (0.15)	0.73	0.54	0.025	0
Dead biomass
Linear			33(11)	24.9 (11.12)	-23.9 (14.47)	--	--	-0.27	0.08	0.08	--
Multiple			40(9)	--	--	--	--	--	--

^† Simple and multiple linear regression models clustered on number of plots.

^‡ AIC_c indicating the best model with AIC_c=0.

LITERATURE CITED

Chen, J., P.Jonsson, M. Tamura, Z. H. Gu, B. Matsushita, and L. Eklundh. 2004. A simple method for reconstructing a high-quality NDVI time-series data set based on the Savitzky-Golay filter. Remote Sensing of Environment 91:332–344.

Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the radiometric and biophysical performance of the Modis Vegetation Indices. Remote Sensing of Environment 83:195–213.

Jobbagy, E. G., O. E. Sala, and J. M. Paruelo. 2002. Patterns and controls of primary production in the Patagonian Steppe: a remote sensing approach. Ecology 83:307–319.

Justice, C. O., L. Giglio, S. Korontzi, J. Owens, J. T. Morisette, D. Roy, J. Descloitres, S. Alleaume, F. Petitcolin, and Y. Kaufman. 2002. The Modis fire products. Remote Sensing of Environment 83:244–262.

Kawamura, K., T. Akiyama, H. Yokota, M. Tsutsumi, T. Yasuda, O. Watanabe, G. Wang, and S. Wang. 2005. Monitoring of forage conditions with MODIS imagery in the Xilingol steppe, Inner Mongolia. International Journal of Remote Sensing 26:1423–1436.

Petorrelli, N., J. O. Vik, A. Mysterud, J.-M. Gaillard, C. J. Tucker, and N.-C. Stenseth. 2005. Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends in Ecology and Evolution 20:503–510.

Thoma, D. P., D. W. Bailey, D. S. Long, G. A. Nielsen, M. P. Henry, M. C. Breneman, and C. Montagne. 2002. Short-term monitoring of rangeland forage conditions with AVHRR imagery. Journal of Range Management 55:383–389.