Supplementary Information Characterizing the encroachment of juniper forests into sub-humid and semi-arid prairies from 1984 to 2010 using PALSAR and Landsat data Jie Wang, Xiangming Xiao, Yuanwei Qin, Russell B. Doughty, Jinwei Dong , Zhenhua Zou Inter-comparison of the juniper forest map in 2010 The comparison between the PALSAR/Landsat based juniper forest map in 2010 (PALSAR/Landsat-JF) (Fig. S11a) and the juniper forest and woodland map derived from the Oklahoma ecosystem map (OKESM-JWF) (Fig. S11b) showed good spatial consistency between these two products. Some differences occurred in two western counties highlighted by the black box in Fig. S11a,b. The area estimation at the county scale showed that these two products have a statistically significant linear correlation with R 2 of 0.74 (Fig. S11c). Two counties with large discrepancies are shown in red circles in Fig. S11c, which corresponds to Woods and Woodward counties shown in the black box in Fig. 6a, b. We 1
Transcript
Supplementary Information
Characterizing the encroachment of juniper forests into sub-humid and semi-arid prairies
from 1984 to 2010 using PALSAR and Landsat data
Jie Wang, Xiangming Xiao, Yuanwei Qin, Russell B. Doughty, Jinwei Dong, Zhenhua Zou
Inter-comparison of the juniper forest map in 2010
The comparison between the PALSAR/Landsat based juniper forest map in 2010
(PALSAR/Landsat-JF) (Fig. S11a) and the juniper forest and woodland map derived from the
Oklahoma ecosystem map (OKESM-JWF) (Fig. S11b) showed good spatial consistency between
these two products. Some differences occurred in two western counties highlighted by the black
box in Fig. S11a,b. The area estimation at the county scale showed that these two products have
a statistically significant linear correlation with R2 of 0.74 (Fig. S11c). Two counties with large
discrepancies are shown in red circles in Fig. S11c, which corresponds to Woods and Woodward
counties shown in the black box in Fig. 6a, b. We zoomed-in on these two counties (Fig. S11d,e),
selected three case regions, and examined the landscapes using the GE high resolution images
dated 03/23/2011 in Fig. S11f, g, h. These GE images show that PALSAR/Landsat-JF may miss
some pixels with small proportions of juniper trees within one Landsat pixel (Fig. S11f) or
woodlands with sparse juniper tree coverage (Fig. S11g,h). The OKESM-JWF map found
juniper woodlands and forests, whereas the PALSAR/Landsat-based juniper forest map
identified only the forests, thus causing some discrepancies between these two products as shown
in Fig. S11g,h.
1
Figure S1. (a) The distribution of Landsat tiles, Elevation, and county boundary of Oklahoma are shown in the figure. (b) The land cover map in Oklahoma from the 2011 National Land Cover Database (2011 NLCD). The pie figure shows the area percentage of different land cover types.
2
Figure S2. The numbers of Landsat images in the study area in a year by (a) sensors (Landsat 5, and Landsat 7), (b) path/row, and (c) months.
3
Figure S3. Numbers of observations for all Landsat images (Landsat 5 and 7) in 2010. Statistics at pixel level by (a) the number of total (all) observations during the year (TO), (b) the number of good-quality observations during the year (GO), (c) the number of total observations
4
in winter season (TOW), (d) the number of good-quality observations in winter (GOW). (e, f, g, h) are the histograms of Figure a, b, c, and d, respectively.
Figure S4. Percentage of good-quality observations (GObs) in (a) winter (Dec-Feb) and (b) whole year over the period of 1984 to 2010.
5
Figure S5. Work flow of this study. It includes three main sections shown in dark grey boxes. Abbreviations in the figure refer to Phased Array type L-band Synthetic Aperture Radar (PALSAR), Land Surface Water Index (LSWI), Normalized Difference Vegetation Index (NDVI), point of interests (POIs), and region of interests (ROIs).
6
Figure S6. (a) Spatial distribution of training samples or region of interests (ROIs) for red cedar forest, Ashe juniper forest, and deciduous forests. (b) Mean NDVI and standard deviation (SD) in winter calculated from Landsat 5/7 images in winters of 2009 and 2010 using the training ROIs for red cedar forest, Ashe juniper, forest, and deciduous forest.
7
Figure S7. (a) 30 m non-juniper evergreen forest/woodland map from the Oklahoma Ecosystem Map (OKESM). (b, c, d) are three zoom-in views for the case regions labeled as 1, 2, 3, in (a).
8
Figure S8. Ground reference samples from random points and high spatial resolution images in Google Earth in 2010-2011. The number of samples for juniper forest, non juniper forest and non forest are 105, 218 and 612, respectively.
Figure S9. (a) PALSAR-based forest map in 2010. (b, c, d) are three zoom-in views for the case regions labeled as 1, 2, 3, in (a).
9
Figure S10. (a) PALSAR/Landsat-based evergreen forest in 2010. (b, c, d) are three zoom-in views for the case regions labeled as 1, 2, 3, in (a).
10
Figure S11. (a) PALSAR/Landsat-based juniper forest map in 2010 (PALSAR/Landsat_JF2010) produced in this study. (b) the zoom-in view from the PALSAR/Landsat_JF2010 for the region shown in the box in (a). (c) the Oklahoma Ecosystem Map-based juniper woodland/forest map (OKESM_JWF) re-sampled at 30 m spatial resolution.
11
(d) the zoom-in view from the OKESM_JWF for the region shown in the box in (b). (e, f, g) the zoom-in views from Google Earth images for three regions labeled as e, f, g in (d), respectively. (h) area comparison of two products at county level. (i) a histogram of juniper forest (JF) coverage calculated from the 30 m OKESM_JWF in (c). This histogram shows (1) the percentage of JF coverage within all pixels of Oklahoma (Perc in all) and the percentage of JF coverage just within the pixels with juniper encroachment (Perc in JF cover).
12
Figure S12. The dynamics of juniper forest (JF) encroachment during two consecutive periods including the late 1980s to early 1990s (L80s-E90s), early 1990 to late 1990s (E90s-L90s), late 1990 to early 2000s (L90s-E00s), and early 2000s to late 2000s (E00s-L00s). (a-d) show the expansion, decrease, and unchanged areas of JF at the spatial distribution with full views, and (e-h) show the zoom-in views at the region highlighted by the blue box in (a). (i) is an area statistics of JF dynamics during each study period. This figure corresponds to Fig. 4 in the manuscript.
Figure S13. Area dynamics in five periods and average stand age of juniper forests for each county.
13
Figure S14. The ratio of average annual encroached juniper forest area to total county land area during 1984-2010 at county level.
Figure S15. Trend analysis of juniper forest encroachment in five periods analyzed by geographic regions of longitude, latitude and elevation. The slope and P-value of each statistics are shown in the figures. This figure corresponds to Fig. 7 in the manuscript.
14
Figure S16. Trend analysis of juniper forest encroachment (JFE) in five periods analyzed by different soil texture. The slope and P-value are shown in the figure. The orange dash line shows P<0.01. This statistics was based on the juniper forest pixels within each soil texture over fiver periods. The soil types include clay (Cl), silty clay (SiCl), sandy clay (SaCl), clay loam (ClLo), silty clay loam (SiClLo), sandy clay loam (SaClLo), loam (Lo), silty loam (SiLo), sandy loam (SaLo), silt (Si), loamy sand (LoSa), and sand (Sa) according to the soil classification system of USDA. This figure corresponds to Fig. 8 in the manuscript.
15
Figure S17. Trend analysis of juniper forest encroachment (JFE) in five periods analyzed by different soil depths and available water storage based on the juniper forest pixels. The slope and P-value are shown in the figure. This figure corresponds to Fig. 9 in the manuscript.
