- Email: [email protected]
The relationship between satellite-derived indices and species diversity across African savanna ecosystems
International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Contents lists available at ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
The relationship between satellite-derived indices and species diversity across African savanna ecosystems Ratidzo B. Mapfumo a,∗ , Amon Murwira a , Mhosisi Masocha a , R Andriani b a b
Department of Geography and Environmental Science, University of Zimbabwe, P.O. Box MP 167, Harare, Zimbabwe Centre for International Forestry Research (CIFOR), P.O. Box 0113 BOCBD, Bogor 16000, Indonesia
a r t i c l e
i n f o
Article history: Received 3 February 2016 Received in revised form 27 May 2016 Accepted 29 June 2016 Keywords: Species diversity Savanna Remote sensing Spectral variation hypothesis
a b s t r a c t The ability to use remotely sensed diversity is important for the management of ecosystems at large spatial extents. However, to achieve this, there is still need to develop robust methods and approaches that enable large-scale mapping of species diversity. In this study, we tested the relationship between species diversity measured in situ with the Normalized Difference Vegetation Index (NDVI) and the Coefficient of Variation in the NDVI (CVNDVI) derived from high and medium spatial resolution satellite data at dry, wet and coastal savanna woodlands. We further tested the effect of logging on NDVI along the transects and between transects as disturbance may be a mechanism driving the patterns observed. Overall, the results of this study suggest that high tree species diversity is associated with low and high NDVI and at intermediate levels is associated with low tree species diversity and NDVI. High tree species diversity is associated with high CVNDVI and vice versa and at intermediate levels is associated with high tree species diversity and CVNDVI. © 2016 Elsevier B.V. All rights reserved.
1. Introduction Savanna ecosystems constitute approximately 20% of the area of world’s terrestrial ecosystems (Sankaran and Anderson, 2009; Van Wilgen, 2009). One-fifth of the world’s population depends on the savanna ecosystem (Solbrig et al., 1996; Parr et al., 2014). These ecosystems also contain 5–20% of the world’s herbivore biomass (Owen-Smith, 1993; Cowling et al., 2004; Sankaran et al., 2005). Thus, the ability to understand patterns of species diversity in savanna ecosystems, especially at large spatial extents is important for their efficient management and conservation. In this regard, the need for methods that can achieve this large scale mapping is critical. The development of satellite remote sensing has provided an opportunity to map diversity at large spatial extents (Nagendra et al., 2010; Rocchini et al., 2015). In fact, remotely sensed data provide an effective and evident way to address ecological patterns at different multiple scales (Simova et al., 2013). However species distribution patterns are easier to map at a broader scale compared to fine-scale distributions (Kerr and Ostrovsky, 2003; Gillman et al., 2015). To this end, several studies have used remote sensing to map
∗ Corresponding author. E-mail address: [email protected] (R.B. Mapfumo). http://dx.doi.org/10.1016/j.jag.2016.06.025 0303-2434/© 2016 Elsevier B.V. All rights reserved.
species diversity in ecosystems (Scheiner and Jones, 2002; Rocchini et al., 2010b; Hernandez-Stefanoni et al., 2015). The main drawback with most of these studies has been their focus on one type of ecosystem, thus making generalizations difficult or impossible. Development of general understanding of the relationship between remotely sensed indices and diversity is, therefore imperative for enhancing the understanding of diversity in ecosystems. Savannas particularly exist in different states depending on moisture regimes (Devine et al., 2015). For example, Southern African savannas exist as dry savanna, wet savanna and coastal savannas (Scholes and Archer, 1997; Scholes and Walker, 2004; Devine et al., 2015). There are differences in species composition, structure and ecological processes between dry savannas, wet savannas and coastal savannas (Justice 1994). Thus, developing remote sensing models that can be applied generally across various ecosystems is crucial. The application of remote sensing in understanding ecological patterns in savanna woodlands principally depends on the suitability of the remotely sensed indices used to relate to ground measurements of biological diversity (Mutowo and Murwira, 2012; Hernandez-Stefanoni et al., 2015) as well as on whether these indices incorporate the range of variation in the savanna ecosystem. Satellite-derived vegetation indices, such as the Normalized Difference Vegetation Index (NDVI) have been shown to be useful estimates of productivity and can also be used to quantify vegetation-related spatial heterogeneity thereby shap-
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
307
Table 1 Description of the seven study sites, the soil data of Zimbabwe (Kutsaga, Shurugwi) was extracted from (Nyamapfene 1991), the soil data of Zambia (Kaoma, Kasenu, Simungoma) were extracted from (Trapnell et al., 2001) and the soil data of Mozambique (Miti, Mofid) were extracted from (Maria and Yost, 2006). Site
Location
Mean Annual Rainfall (mm)
Mean Annual Temperature
Soils
Vegetation data
Kutsaga
17◦ 55 S 31◦ 08 E
850
18.6 ◦ C
Ferallic cambisols
Shurugwi
19◦ 58 S 29◦ 51 E
800
17.6 ◦ C
Lithic leptosols
Kaoma
15◦ 70 S 24◦ 45 E
1100
21.5 ◦ C
Ferrallic arenosols
Kasenu Simungoma
15◦ 30 S 25 30 E
1000
27.7 ◦ C
Miombo woodlands dominated by Julbernadia globiflora, and Brachystegia spiciformis Miombo woodlands dominated by Julbernardia globiflora, Msasa Brachystegia spiciformis and Terminalia. Miombo woodlands characterized by Brachystegia, Julbenardia paniculta and Marquesia marcroura Isoberlinia Angolensis, The Kalahari woodland dominated by Cryptosepalum Miombo woodland dominated by Dry Deciduous Thicket with Guibourtia schliebenii Sclerophyllous species which include Manilkara sansibarensis, Warneckea sansibarica and Baphia macrocalyx
◦
Miti
11 39 S 39 33 E
1000
31.7 C
Ferrallic/ cambic arenosols Ultisols
Mofid
11◦ 43 S 39◦ 47 E
1400
26 ◦ C
Oxisols
◦
◦
◦
ing biodiversity patterns (Tucker and Sellers, 1986; Thiollay, 1997; Shimabukuro et al., 1998; Loboda et al., 2013; Girma et al., 2016). In addition, the variance in NDVI has also been used as a useful measure of diversity in ecosystems (Bongers et al., 2009). The variance in NDVI as an index for estimating diversity in ecosystems is well supported by the spectral variation hypothesis. The theory of Spectral Variation proposes that spatial variation on a remotely sensed image is related to spatial variations of the environment which reflects habitat heterogeneity (Rocchini et al., 2010a; Heumanna et al., 2015). Therefore, habitat heterogeneity is linked to the structural complex of habitats which may provide environmental resources leading to an increase in species diversity (Oldeland et al., 2010). Although, variance in NDVI has been used to successfully characterize tree species diversity in the dry savanna woodlands of Mukuvisi, Mutirikwi and Mabalauta in Zimbabwe (Mutowo and Murwira, 2012), this work did not include various types of savanna such as wet savanna sites, as well as, coastal savannas. Thus, the inclusion of different savanna sites in modeling the relationship between species diversity and remotely sensed indices could improve our capacity to map diversity at large spatial extents. Therefore, in this study, we test the hypothesis that tree species diversity and satellite-derived indices of NDVI should be positively correlated, and that tree species diversity will be positively correlated with the Coefficient of Variation of NDVI. The use of remote sensing in ecology is vital as it has been widely used to identify spatial and temporal patterns in biodiversity in different ecosystems over a period of time (Wiens et al., 2009). Remotely sensed spectral heterogeneity information also offers an inexpensive means to derive spatially complete environmental information for large areas in a consistent and systematic manner (Levin et al., 2007; Rocchini et al., 2010b). Ecologists may gain critical knowledge about the drivers of the spatial and temporal distribution of biodiversity at any given time (Pettorelli et al., 2014). Through the use remote sensed data, we can understand the positive as well as the negative impacts on biodiversity, predictions about the future and action to prevent ecological degradation strategy to mitigate adverse the impacts observed (Dodson et al., 2000; Zellweger et al., 2013). Principled and appropriate forest management can lead to increased biological diversity (Parma and Shataee, 2013).Explaining the mechanisms driving the observed relationships is of fundamental importance to understanding the determinants of biodiversity (Mittelbach et al., 2001).The biodiversity patterns observed may suggest dynamic, nonequilibrium community processes encountered in ecosystems (Graham and
Duda, 2011). Remote sensing approaches may provide planners and conservation biologists with an efficient and cost-effective method to study and estimate biodiversity different ecosystems (Levin et al., 2007). In this study, we tested the relationship between species diversity measured in situ with the NDVI and CVNDVI derived from high and medium spatial resolution satellite data in dry, wet and coastal savanna woodlands. We selected the study sites in the dry savanna of Zimbabwe, wet savanna of Zambia and coastal savanna in Mozambique because they experience different climatic conditions which as a result influence different ecological patterns. 2. Materials and methods 2.1. Study area The study was carried out at 7 study sites (Fig. 1a): Zimbabwe (Kutsaga and Shurugwi), Zambia (Kaoma, Sesheke Simungoma, and Sesheke Kasenu) and Mozambique (Miti and Mofid) (Table 1). Each of the six study sites covered 100 km2 in area. 3. Field data 3.1. Tree species data We randomly selected six transects using ArcView GIS 3.2 (ESRI, 2002) in each study site. The transects had a minimum length of 3 km and a maximum length of 6 km. Transect length was determined on how accessible were the sampling points, transects with a maximum length of 6 km were characterized by rugged terrain which makes the sampling points to be inaccessible (Marshall and Region, 2000). The starting point and the ending point of each transect were navigated using a handheld Global Positioning System (GPS) receiver. The locational accuracy of the field plots was within the 15 m error of the GPS. We defined sample plots of 15 m × 15 m which were north oriented at 500 m distances along each transect and in each sample plot, species names were recorded. Research has shown that sampling plot sizes widely used ranges between 25 and 200 m2 in tall shrub communities and 200–25000 m2 for trees in woods and forests (Sutherland and Krebs, 1997) and our plot size of 225 m2 falls within this range. A sampling distance of 500 m between the sampling plots was used because it is a distance that can capture the spatial variations in forest structure, species composition and vegetation density thereby reducing uniformity
308
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Mozambique
N
Zambia
0
0
400
800
1200 Kilometers
60
120
180 Kilometers
Zimbabwe
Study sites Kutsaga Shurugwi Kaoma Kasenu Simugoma Miti Mofid
0
200
400
600 Kilometers
Fig. 1a. The location of the study sites.
on plots (Lutz et al., 2013). Tree species that could not be identified in the field were taken for identification at the Herbariums in Lusaka for the Zambian sites, in Harare for the Zimbabwean sites and Maputo in Mozambique. Tree species data were collected in February 2012 and October 2012 in Shurugwi and Kutsaga respectively, in October 2013 in Sesheke Simungoma, Sesheke Kasenu, Kaoma, Miti and Mofid.
Table 2 The description of the dates acquired for Landsat 8 and Worldview-2 satellite images for the sites in Zimbabwe, Zambia and Mozambique. Country
Site
Date Acquired Landsat 8
Date Acquired Worldview-2
Zimbabwe
Kutsaga Shurugwi Kaoma Simungoma Kasenu Miti Mofid
11 July 2013 11 July 2013 07 July 2013 07 July 2013 07 July 2013 19 July 2013 19 July 2013
24 July 2012 24 July 2012 16 July 2012 30 June 2013 16 July 2013 – –
Zambia
Mozambique
3.2. Measures of diversity In this study, we used Shannon’s H and Simpson’s 1-D as measures of diversity. Simpson’s diversity index is based on the probability of any two individuals drawn at random from an infinity large community belonging to the same species (DeJong, 1975). The index ranges from 0 to 1 in ascending order with increased diversity with a greater probability of interspecific encounter (Magurran, 2013). Shannon’s diversity index is a measure of the average ‘uncertainty’ of predicting to what species an individual is chosen at random from a collection of species belong (Williams et al., 2005; Magurran 2013). There is an increase in the average uncertainty as the number of species also increases and the distribution of individuals among the species become more even (DeJong, 1975; Ludwig and Reynolds, 1988). In this study, we selected the Simpson diversity index (1-D) (equation 1) and Shannon diversity index (H) (equation 2) because these indices combine evaluations of abundance and evenness of the species present in an ecosystem (Elzinga and Evenden, 1997; Oldeland et al., 2010; Rocchini et al., 2010b; Duro et al., 2014). In addition, the Shannon’s diversity index is less affected by the presence of rare species (Nagendra, 2002).The Simpson’s (1-D) index of diversity is calculated as follows:
1−D=
n i=1
pi2
(1)
where pi is the proportion of the ith species in the sampling plot. The Shannon-Weaver index is calculated as follows: H=
n
− pi log pi
(2)
i=1
where pi is the proportion of the ith species in the sampling plot. 3.3. Remotely sensed data In this study, we used Landsat 8 imageries which are made freely available by the United States Geological Survey (USGS) and the imageries were downloaded from http://glovis.usgs.gov/ . We also used Worldview -2 imagery as a source of multispectral remotely sensed data (Table 2). Digital numbers (DN) of Landsat 8 satellite imagery were converted to top-of-atmosphere spectral reflectance in ENVI 5.1 image processing software (ITT, 2013) using the reflectance rescaling coefficients provided in the metadata files of the images while the Quick Atmospheric Correction (QUAC) algorithm was applied in ENVI 4.8 (ITT, 2008) for Worlview-2. We also Pan Sharpened the Landsat 8 imagery to 15 m × 15 m to coincide with the field plot size of 15 m × 15 m. Landsat 8 satellite images are associated with geometric errors of 12 m in CE90 which mean that 90% of the points has less than 12 m positional error (Roy et al.,
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
309
Fig. 1b. The spatial variations in canopy structure estimated by NDVI derived from Landsat 8 in (a) Kutsaga, (b) Shurugwi, (c) Kaoma, (d) Kasenu, (e) Simungoma, (f) Miti and (g) Mofid. The map units are Geographic coordinates.
2014).The variations in canopy structure was estimated by NDVI derived from Landsat 8 (Fig. 1b).
3.4. Remotely sensed indices NDVI derived from Landsat 8 and Worldview-2 was calculated in GIS by subtracting the value of the red band from the value of the NIR band and divide it by the sum of red and NIR band values (Eidenshink, 1992). The NDVI has a range of −1 to +1. NDVI was selected as an index since it is known to represent the greenness of active biomass (DeFries and Townshend, 1994; Parviainen et al.,
2010; Hall et al., 2012). In addition, NDVI has widely been used in the study of biodiversity in landscapes of spatial heterogeneity (Santin-Janin et al., 2009; Cui et al., 2013; Zhang et al., 2016). Next, we calculated the Coefficient of Variation in the NDVI (CVNDVI) derived from Landsat 8 and Worldview 2 images by dividing the standard deviation of NDVI with the average NDVI, expressed as a percentage using a moving window of 3pixels by 3 pixels. CVNDVI has been used in diversity studies and has been proven to be a simple and effective indicator of spectral heterogeneity (Lucas and Carter, 2008; Duro et al., 2014; Somers et al., 2015). The Coefficient of Variation was applied in the Great Basin
310
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Fig. 2. Relationships between tree species diversity estimated using Shannon’s H index and the Normalized Difference Vegetation Index (NDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
of western North America (Seto et al., 2004), to characterize the relationship between species diversity of butterflies and remotely sensed diversity. 4. Data analysis The mean NDVI, as well as mean CVNDVI, computed from Landsat-8 and Worldview 2 were regressed against Shannon’s H and Simpson’s 1-D diversity indices calculated from field data yielding a total of 10 relationships tested. Worldview-2 imagery for the Mozambican site was not available and therefore the imageries were not used in this study. Due to the curvature inherent in the diversity-productivity relationships, linear and non-linear regression based on the second-order polynomial model was used to test the consistency of the relationships across different ecosystems in this study. We tested whether NDVI, CVNDVI and tree diversity data followed a normal distribution using the Kolmogorov- Smirnov test and the results indicated that our data followed a normal distribution. The strength and significance of the relationship between field-based indices and satellite-derived indices were evaluated
using the R2 and P-values, respectively. We further tested the effect of disturbance on NDVI derived from Landsat along the transects and between transects of the same site using students t-test and two-way ANOVA respectively. Prior to analysis, we tested whether NDVI data significantly deviated from a normal distribution using Kolmogorov-Smirnov test and the results showed that the data did not significantly deviate from a normal distribution thus parametric tests could be used.
5. Results A consistent and significant U-shaped relationship is observed (Fig. 2) between Shannon’s H index of diversity and NDVI at the dry savanna of Zimbabwe regardless of whether NDVI was calculated from Landsat 8 or Worldview-2 and at the wet savanna sites derived from Worldview-2 a significant U-shaped relationship is observed. There is a positive linear and significant relationship observed at the wet savanna sites of Zambia and coastal savanna sites of Mozambique derived from Landsat 8. However, the relationship between
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
311
Fig. 3. Relationships between tree species diversity estimated using Simpson’s 1-D index and the Normalized Difference Vegetation Index (NDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
Shannon’s H index of diversity and NDVI is observed to be stronger when using Worldview-2 compared to Landsat-8. Fig. 3 shows that a is consistent U-shaped relationship between Simpson’s 1-D diversity index and NDVI derived from Landsat 8 and Worldview-2 at the dry savanna sites of Zimbabwe, wet savanna sites of Zambia and coastal savanna sites of Mozambique. A hump-shaped and significant relationship is observed between Shannon’s H diversity index and remotely sensed diversity (CVNDVI) at the dry savanna of Zimbabwe (Fig. 4) using remotely sensed diversity (CVNDVI) derived from both Landsat 8 and Worldview-2 and at the wet savanna of Zambia using Landsat 8 derived index. However, a positive linear relationship is observed at the wet savanna sites of Zambia and coastal savanna of Mozambique using remotely sensed diversity (CVNDVI) derived from Worldview-2 and Landsat 8 respectively. Hence, the results also indicate no significant relationship for the sites in Mozambique. We also observe a hump-shaped significant relationship between Simpson’s 1-D diversity index and remotely sensed diversity (CVNDVI) at the dry savanna of Zimbabwe (Fig. 5). At the
wet savanna sites of Zambia, we observe a hump-shaped and positive linear relationship between Simpson’s 1-D diversity index and remotely sensed diversity (CVNDVI) derived from both Landsat 8 and Worldview-2 respectively. A positive linear relationship at the sites in Mozambique using CVNDVI derived from Landsat 8 is observed. Hence, the results indicate no significant relationship for the sites in Mozambique. Regardless of the different patterns observed, the relationship between Simpson’s 1-D diversity index and remotely sensed diversity (CVNDVI) is stronger using Worldview-2.
5.1. Effect of logging on NDVI derived from Landsat 8 between logged and unlogged areas along and between transects in Zimbabwe, Zambia and Mozambique The overall results show no significant differences (p > 0.05) between logged and unlogged areas in mean NDVI along the transects (Table 3). However, there are significant differences (p < 0.05) between logged and unlogged areas along the transects in Kutsaga (transect 1and 2), Simungoma (transect 1) and Mofid (transect 1).
312
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Table 3 NDVI derived from Landsat 8 (mean ± standard error) in logged and unlogged plots along the transects in Zimbabwe, Zambia and Mozambique. Country
Site
Transect
Logged
t statistic
p-value
Zimbabwe
Kutsaga
1 2 3 4 5 6 1 2 3 4 5 6
0.36 0.46 0.35 0.43 0.44 0.42 0.62 0.70 0.66 0.59 0.67 0.63
0.03 0.014 0.03 0.06 0.02 0.05 0.09 0.07 0.02 0.14 0.07 0.04
0.5 0.29 0.34 0.31 0.38 0.31 0.67 0.55 0.62 0.40 0.62 0.61
± ± ± ± ± ± ± ± ± ± ± ±
0.05 0.2 0.03 0.02 0.003 0.03 0.02 0.08 0.05 0.05 0.01 0.04
−2.908 5.816 0.162 1.414 1.851 1.512 −0.374 1.439 0.703 0.952 −0.755 1.311
0.04* 0.01* 0.89 0.25 0.16 0.35 0.73 0.22 0.56 0.41 0.53 0.25
1 2 3 4 1 2 3 4 5 1 2 3 4 5
0.22 ± 0.04 0.24 ± 0.03 0.24 ± 0.02 0.18 ± 0.02 0.36 ± 0.001 0.34 ± 0.016 0.30 ± 0.039 0.36 ± 0.02 0.34 ± 0.02 0.33 ± 0.01 0.27 ± 0.04 0.33 ± 0.01 0.35 ± 0.03 0.38 ± 0.003
0.20 0.26 0.25 0.23 0.29 0.33 0.31 0.35 0.32 0.34 0.28 0.33 0.36 0.32
± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.05 0.02 0.05 0.05 0.02 0.03 0.06 0.06 0.01 0.02 0.008 0.03 0.12 0.03
0.225 −0.545 −0.222 −0.911 4.051 0.479 −0.114 −0.133 0.718 0.605 −0.215 0.225 −0.405 0.623
0.84 0.64 0.85 0.46 0.03* 0.68 0.92 0.90 0.53 0.59 0.85 0.84 0.73 0.60
1 2 3 1 2 3 4 5
0.63 0.38 0.45 0.25 0.19 0.19 0.23 0.47
± ± ± ± ± ± ± ±
0.45 0.37 0.49 0.27 0.35 0.21 0.32 0.47
± ± ± ± ± ± ± ±
0.38 0.07 0.008 0.05 0.08 0.09 0.09 0.03
4.908 0.055 −3.000 −0.298 −1.285 −0.138 −0.813 −0.041
0.02* 0.96 0.05 0.79 0.33 0.9 0.5 0.97
Shurugwi
Zambia
Kaoma
Simungoma
Kasenu
Mozambique
Mofid
Miti
*
± ± ± ± ± ± ± ± ± ± ± ±
Unlogged
0.02 0.15 0.009 0.05 0.06 0.07 0.07 0.03
<0.05.
The results from ANOVA showed no significant differences in NDVI derived from Landsat 8 (Fig. 6, Table 4) between logged and unlogged areas (p > 0.05) between the transects of the same site except for the site in Zimbabwe where there is a significant difference in NDVI between logged and unlogged areas (p < 0.05) between the transects. There is inconsistence in mean NDVI between logged and unlogged plots between the transects.
Table 4 The differences in NDVI derived from Landsat 8 between Logged and unlogged areas between transects in the sites of Zimbabwe, Zambia and Mozambique. Country
Site
F value
p-value
Zimbabwe
Kutsaga Shurugwi Kaoma Kasenu Simungoma Miti Mofid
5.448 0.756 0.243 0.244 0.421 0.313 2.113
0.001* 0.57 0.86 0.91 0.79 0.86 0.18
Zambia
6. Discussion
Mozambique
The results of this study suggest the existence of a U-shaped relationship between tree species diversity and NDVI at dry and wet savanna sites in Zimbabwe and Zambia, respectively. The results of this study are consistent with the findings where a U-shaped relationship between diversity and NDVI was found in semi natural savanna grassland in Sweden (Nagendra et al., 2013). In addition, U-shaped relationships between NDVI and diversity in Mukuvisi, Mutirikwi and Mabalauta sites in Zimbabwe were observed (Mutowo and Murwira, 2012). The mechanisms which explain this pattern are not clear. However, a proposed explanation of such a pattern can arise when the green cover of the competitively dominant tree and shrub tree species occur at intermediate times after disturbance (Gosper et al., 2013). In this regard, increased biomass production after logging may be triggering the U-shaped relationships. However, human induced disturbance of logging may not be the mechanism explaining the U-shaped relationships observed since there is no significant difference in NDVI between logged and unlogged areas along and between the transects of the study sites. Further research targeting non-logged (protected sites) adjacent to logged sites is needed to validate the effect of logging on the ecological patterns especially the U-shaped pat-
*
<0.05.
tern. Although these results are inconsistent with previous research where hump-shaped relationships between NDVI and tree diversity were reported (Waide et al., 1999), we deduce that disturbance may not be a mechanism driving U-shaped relationships in savanna ecosystems. There are consistent relationships between diversity and CVNDVI derived from Landsat 8 Worldview-2 in the study sites of Zimbabwe and Mozambique. The linear relationships observed in Zambia and Mozambique can be explained by the fact that wet and coastal ecosystems receive high rainfall above 1000 mm leading to high diversity which facilitates high CVNDVI (Kerr and Ostrovsky, 2003; Glenn et al., 2008; Fatehi et al., 2015). These results are interesting as they depict the importance of high spatial resolution in predicting species diversity in wet and coastal climates. Thus, results suggest that the use of high spatial resolution imagery may be important to allow for the capturing of more detailed information on the spectral variability of landscape features consistent with the spectral variability hypothesis (Rocchini et al., 2010b;
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
313
Fig. 4. Relationships between tree species diversity estimated using Shannon’s H index and the Coefficient of Variation in the Normalized Difference Vegetation Index (CVNDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
Turner et al., 2015).Our results also indicate the existence of a hump-shaped relationship between the tree species diversity and Coefficient of Variation of NDVI observed at both dry and wet savanna sites in Zimbabwe and Zambia, respectively. This result is consistent with the findings of a study conducted at Mount Hermon, located in NE Israel and a savanna ecosystem of Kenya where humpshaped relationships between species diversity and CVNDVI (Oindo and Skidmore, 2002; Levin et al., 2007) was observed. Thus, we deduce that we can predict species diversity using CVNDVI derived from both high and medium satellite resolution data. The results of this study further suggest stronger relationships between tree species diversity against NDVI and CVNDVI derived from Worldview-2. This could be a result of the higher spatial resolution of Worldview-2 imagery with a spatial resolution of ∼2 m compared with the Landsat 8 imagery with a spatial resolution of ∼30m. These results are not surprising as they are consistent with other findings where relationships between Landsat derived indices and diversity were found to be weaker than those derived
from Worldview-2 (Nagendra and Rocchini, 2008; Hall et al., 2012; Karteris et al., 2016). However, due to high expenses involved in acquiring high spatial resolution data; their use in spatial ecology remains limited. To this end, most vegetation diversity assessment studies use Landsat data that are more readily available and useful (St Louis et al., 2009; Oldeland et al., 2010). The high spatial resolution images such as Worldview-2 can be used for calibration and validation of regional remote sensing based on species diversity mapping (Immitzer et al., 2012; Heenkenda et al., 2014; Cho et al., 2015). This is because Worldview-2 images are not readily available (commercial) and costly while Landsat imageries are freely available. Thus, we deduce that although high spatial resolution satellite data are capable of predicting species diversity at finer scales which is important when monitoring savanna ecosystems, the use of medium spatial resolution imagery such as Landsat still remains invaluable. Where our study differs from previous studies (Levin et al., 2007; Saatchi et al., 2008; Rocchini et al., 2010b; Hall et al., 2012;
314
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Fig. 5. Relationships between tree species diversity estimated using Simpson’s 1-D index and the Coefficient of Variation in the Normalized Difference Vegetation Index (CVNDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
Turner et al., 2015) is the testing of the relationships between species diversity and satellite derived indices across three savanna ecosystems, i.e., dry, wet and coastal savanna woodlands. Previous research has reported the findings of the ecological patterns from single sites. For example, the relationship between diversity measures and spectral variability using hyperspectral remote sensing data has been tested on a single semi-arid savanna ecosystem of Central Namibia (Oldeland et al., 2010). In a study conducted in the southeast of the United States (Costanza et al., 2011), they also characterized the relationship between species diversity and remotely sensed data in a single coastal savanna plain. In a savanna ecosystem located to the southeast of Australia (Schmidt et al., 2015) also assessed the relationship between species diversity and satellite derived indices. In this regard, we deduce that the relationship between diversity and remotely sensed diversity becomes more robustly understood when applicable across different ecosystems. The main limitation of this study is that the findings are based on a single season and that the satellite data used was collected
when tree species were still in leaf (Chidumayo, 1997) while the grass species had senesced for all the study sites. Again, the sample size used in this present study is inadequate, therefore we recommend that future studies may increase the sample size so that strong conclusions on the prediction of species diversity can be made. Therefore, future studies could use remotely sensed data as well as tree inventory data at different times of the year in order to further our understanding of the patterns observed.
7. Conclusion From the results of this study, Firstly, we concluded that tree species diversity can be predicted using NDVI and the CVNDVI derived from both high and medium spatial resolution satellite data across different savanna ecosystems, i.e., dry, wet and coastal savanna woodlands. Secondly, we concluded that although the mechanism explaining the relationship between species diversity and NDVI is not well understood, however, high tree species diver-
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
315
Fig. 6. Mean NDVI derived from Landsat 8 along the transects between logged and unlogged areas in a) Kutsaga b) Shurugwi c) Kaoma d) Kasenu e) Simungoma f) Miti g) Mofid.
sity is associated with low and high NDVI and at intermediate levels is associated with low tree species diversity and NDVI. Thirdly, we also concluded that high tree species diversity is associated with high CVNDVI and vice versa and at intermediate levels is associated with high tree species diversity and CVNDVI. Finally, we concluded that although results from high and medium spatial resolution satellite indices gave consistent results, high spatial resolution imagery provided comparatively better results.
Acknowledgement We thank the Centre for International Forestry Research (CIFOR) for their financial support thus making data collection possible. Their dedication to developing policies and technologies for sustainable use and management of forests is appreciated.
References Bongers, F., Poorter, L., Hawthorne, W.D., Sheil, D., 2009. The intermediate disturbance hypothesis applies to tropical forests, but disturbance contributes little to tree diversity. Ecol. Lett. 12, 798–805. Chidumayo, E.N., 1997. Miombo Ecology and Management: An Introduction. Intermediate Technology Publications Ltd (ITP). Cho, M.A., Malahlela, O., Ramoelo, A., 2015. Assessing the utility WorldView-2 imagery for tree species mapping in South African subtropical humid forest and the conservation implications: Dukuduku forest patch as case study. Int. J. Appl. Earth Obs. Geoinf. 38, 349–357. Costanza, J.K., Moody, A., Peet, R.K., 2011. Multi-scale environmental heterogeneity as a predictor of plant species richness. Landscape Ecol. 26, 851–864. Cowling, R.M., Richardson, D.M., Pierce, S.M., 2004. Vegetation of Southern Africa. Cambridge University Press. Cui, X., Gibbes, C., Southworth, J., Waylen, P., 2013. Using remote sensing to quantify vegetation change and ecological resilience in a semi-arid system. Land 2, 108–130. DeFries, R., Townshend, J., 1994. NDVI-derived land cover classifications at a global scale. Int. J. Remote Sens. 15, 3567–3586. DeJong, T.M., 1975. A comparison of three diversity indices based on their components of richness and evenness. Oikos 26, 222–227.
316
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Devine, A.P., Stott, I., McDonald, R.A., Maclean, I., 2015. Woody cover in wet and dry African savannas after six decades of experimental fires. J. Ecol. 103, 473–478. Dodson, S.I., Arnott, S.E., Cottingham, K.L., 2000. The relationship in lake communities between primary productivity and species richness. Ecology 81, 2662–2679. Duro, D.C., Girard, J., King, D.J., Fahrig, L., Mitchell, S., Lindsay, K., Tischendorf, L., 2014. Predicting species diversity in agricultural environments using Landsat TM imagery. Remote Sens. Environ. 144, 214–225. ESRI, 2002. ArcView GIS 3.2. Environmental Systems Research Institute, California. Eidenshink, J.C., 1992. The 1990 conterminous U. S. AVHRR data set. Photogramm. Eng. Remote Sens. 58, 809–813. Elzinga, C.L., Evenden, A.G., 1997. Vegetation Monitoring: an Annotated Bibliography. DIANE Publishing. Fatehi, P., Damm, A., Schaepman, M.E., Kneubühler, M., 2015. Estimation of alpine forest structural variables from imaging spectrometer data. Remote Sens. 7, 16315–16338. Gillman, L.N., Wright, S.D., Cusens, J., McBride, P.D., Malhi, Y., Whittaker, R.J., 2015. Latitude, productivity and species richness. Global Ecol. Biogeogr. 24, 107–117. Girma, A., de Bie, C., Skidmore, A.K., Venus, V., Bongers, F., 2016. Hyper-temporal SPOT-NDVI dataset parameterization captures species distributions. Int. J. Geogr. Inf. Sci. 30, 89–107. Glenn, E.P., Huete, A.R., Nagler, P.L., Nelson, S.G., 2008. Relationship between remotely-sensed vegetation indices, canopy attributes and plant physiological processes: what vegetation indices can and cannot tell us about the landscape. Sensors 8, 2136–2160. Gosper, C.R., Yates, C.J., Prober, S.M., 2013. Floristic diversity in fire-sensitive eucalypt woodlands shows a ‘U’-shaped relationship with time since fire. J. Appl. Ecol. 50, 1187–1196. Graham, J.H., Duda, J.J., 2011. The humpbacked species richness-curve: a contingent rule for community ecology. Int. J. Ecol. Hall, K., Reitalu, T., Sykes, M.T., Prentice, H.C., 2012. Spectral heterogeneity of QuickBird satellite data is related to fine scale plant species spatial turnover in semi-natural grasslands. Appl. Veg. Sci. 15, 145–157. Heenkenda, M.K., Joyce, K.E., Maier, S.W., Bartolo, R., 2014. Mangrove species identification: comparing WorldView-2 with aerial photographs. Remote Sens. 6, 6064–6088. Hernandez-Stefanoni, J.L., Johnson, K.D., Cook, B.D., Dupuy, J.M., Birdsey, R., Peduzzi, A., Tun-Dzul, F., 2015. Estimating species richness and biomass of tropical dry forests using LIDAR during leaf-on and leaf-off canopy conditions. Appl. Veg. Sci. 50, 1–9. Heumanna, B.W., Hackettb, R.A., Monfilsb, A.K., 2015. Testing the spectral diversity hypothesis using spectroscopy data in a simulated wetland community. Ecol. Inf. 25. ITT, 2008. ENVI 4.8. ITT Industries Inc., Colorado. ITT, 2013. ENVI 5.1. ITT Industries Inc., Colorado. Immitzer, M., Atzberger, C., Koukal, T., 2012. Tree species classification with random forest using very high spatial resolution 8-band WorldView-2 satellite data. Remote Sens. 4, 2661–2693. Karteris, M., Theodoridou, I., Mallinis, G., Tsiros, E., Karteris, A., 2016. Towards a green sustainable strategy for Mediterranean cities: assessing the benefits of large-scale green roofs implementation in Thessaloniki, Northern Greece, using environmental modelling, GIS and very high spatial resolution remote sensing data. Renew. Sustain. Energy Rev. 58, 510–525. Kerr, J.T., Ostrovsky, M., 2003. From space to species: ecological applications for remote sensing. Trends Ecol. Evol. 18, 299–305. Levin, N., Shmida, A., Levanoni, O., Tamari, H., Kark, S., 2007. Predicting mountain plant richness and rarity from space using satellite-derived vegetation indices. Divers. Distrib. 13, 692–703. Loboda, T.V., Zhang, Z., O’Neal, K.J., Sun, G., Csiszar, I.A., Shugart, H.H., Sherman, N.J., 2013. Reconstructing disturbance history using satellite-based assessment of the distribution of land cover in the Russian Far East. Remote Sens. Environ. 118, 241–248. Lucas, K.L., Carter, G.A., 2008. The use of hyperspectral remote sensing to assess vascular plant species richness on Horn Island, Mississippi. Remote Sens. Environ. 112, 3908–3915. Ludwig, J.A., Reynolds, J.F., 1988. Statistical Ecology: A Primer in Methods and Computing. John Wiley & Sons. Lutz, J.A., Larson, A.J., Freund, J.A., Swanson, M.E., Bible, K.J., 2013. The importance of large-diameter trees to forest structural heterogeneity. PLoS One 8, e82784. Magurran, A.E., 2013. Ecological Diversity and Its Measurement. Springer Science & Business Media. Maria, R.M., Yost, R., 2006. A survey of soil fertility status of four agroecological zones of Mozambique. Soil Sci. 171, 902–914. Marshall, P.L., Region, V. F., 2000. Using line intersect sampling for coarse woody debris. [Research Section], Vancouver Forest Region. Mittelbach, G.G., Steiner, C.F., Scheiner, S.M., Gross, K.L., Reynolds, H.L., Waide, R.B., Willig, M.R., Dodson, S.I., Gough, L., 2001. What is the observed relationship between species richness and productivity? Ecology 82, 2381–2396. Mutowo, G., Murwira, A., 2012. Relationship between remotely sensed variables and tree species diversity in savanna woodlands of Southern Africa. Int. J. Remote Sen. 33, 6378–6402. Nagendra, H., Rocchini, D., 2008. High resolution satellite imagery for tropical biodiversity studies: the devil is in the detail. Biodivers. Conserv. 17, 3431–3442.
Nagendra, H., Rocchini, D., Ghate, R., Sharma, B., Pareeth, S., 2010. Assessing plant diversity in a dry tropical forest: comparing the utility of Landsat and IKONOS satellite images. Remote Sens. 2, 478–496. Nagendra, H., Lucasb, R., Honradoc, J.P., Jongmand, R.H.G., Tarantinoe, C., Adamoe, M., Mairotaf, P., 2013. Remote sensing for conservation monitoring: assessing protected areas, habitat extent, habitat condition, species diversity, and threats. Ecol. Indic. 33, 45–59. Nagendra, H., 2002. Opposite trends in response for the Shannon and Simpson indices of landscape diversity. Appl. Geogr. 22, 175–186. Nyamapfene, K.W., 1991. The Soils of Zimbabwe. Nehanda Publishers. Oindo, B.O., Skidmore, A.K., 2002. Interannual variability of NDVI and species richness in Kenya. Int. J. Remote Sens. 23, 285–298. Oldeland, J., Wesuls, D., Rocchini, D., Schmidt, M., Jürgens, N., 2010. Does using species abundance data improve estimates of species diversity from remotely sensed spectral heterogeneity? Ecol. Indic. 10, 390–396. Owen-Smith, N., 1993. Evaluating optimal diet models for an African browsing ruminant, the kudu: how constraining are the assumed constraints? Evol. Ecol. 7, 499–524. Parma, R., Shataee, S., 2013. Estimation of species diversity of trees and shrubs using ETM+ sensor data (case study of forests in Qalajeh Kermanshah province). Int. J. Adv. Biol. Biomed. Res. (IJABBR) 1, 71–78. Parr, C.L., Lehmann, C.E.R., Bond, W.J., Hoffmann, W.A., Andersen, A.N., 2014. Tropical grassy biomes: misunderstood, neglected, and under threat. Trends Ecol. Evol. 29, 205–213. Parviainen, M., Luoto, M., Heikkinen, R.K., 2010. NDVI-based productivity and heterogeneity as indicators of plant-species richness in boreal landscapes. Boreal Environ. Res. 15, 301–318. Pettorelli, N., Safi, K., Turner, W., 2014. Satellite remote sensing, biodiversity research and conservation of the future. Philos. Trans. R. Soc. Lond. B: Biol. Sci. 369, 20130190. Rocchini, D., Balkenhol, N., Carter, G.A., Foody, G.M., Gillespie, T.W., He, K.S., Kark, S., Levin, N., Lucas, K., Luoto, M., 2010a. Remotely sensed spectral heterogeneity as a proxy of species diversity: recent advances and open challenges. Ecol. Inf. 5, 318–329. Rocchini, D., Balkenhol, N., Carter, G.A., Foody, G.M., Gillespie, T.W., He, K.S., Kark, S., Levin, N., Lucas, K., Luoto, M., Nagendra, H., Oldeland, J., Ricotta, C., Southworth, J., Neteler, M., 2010b. Remotely sensed spectral heterogeneity as a proxy of species diversity: recent advances and open challenges. Ecol. Inf., 1–12. Rocchini, D., Hernández-Stefanoni, J.L., He, K.S., 2015. Advancing species diversity estimate by remotely sensed proxies: a conceptual review. Ecol. Inf. 25, 22–28. Roy, D.P., Wulder, M., Loveland, T., Woodcock, C., Allen, R., Anderson, M., Helder, D., Irons, J., Johnson, D., Kennedy, R., 2014. Landsat-8: Science and product vision for terrestrial global change research. Remote Sens. Environ. 145, 154–172. Saatchi, S., Buermann, W., t. Steege, H., Mori, S., Smith, T.B., 2008. Modeling distribution of Amazonian tree species and diversity using remote sensing measurements. Remote Sens. Environ. 112, 2000–2017. Sankaran, M., Anderson, T.M., 2009. Management and restoration in African savannas: interactions and feedbacks. New Models Ecosyst. Dyn., 136–155. Sankaran, M., Hanan, N.P., Scholes, R.J., Ratnam, J., Augustine, D.J., Cade, B.S., Gignoux, J., Higgins, S.I., Le Roux, X., Ludwig, F., 2005. Determinants of woody cover in African savannas. Nature 438, 846–849. Santin-Janin, H., Garel, M., Chapuis, J.L., Pontier, D., 2009. Assessing the performance of NDVI as a proxy for plant biomass using non-linear models: a case study on the Kerguelen archipelago. Polar Biol. 32, 861–871. Scheiner, S.M., Jones, S., 2002. Diversity, productivity and scale in Wisconsin vegetation. Evol. Ecol. Res. 4, 1097–1117. Schmidt, M., Lucas, R., Bunting, P., Verbesselt, J., Armston, J., 2015. Multi-resolution time series imagery for forest disturbance and regrowth monitoring in Queensland, Australia. Remote Sens. Environ. 158, 156–168. Scholes, R., Archer, S., 1997. Tree-grass interactions in savannas. Annu. Rev. Ecol. Syst., 517–544. Scholes, R.J., Walker, B.H., 2004. An African Savanna: Synthesis of the Nylsvley Study. Cambridge University Press. Seto, K., Fleishman, E., Fay, J., Betrus, C., 2004. Linking spatial patterns of bird and butterfly species richness with Landsat TM derived NDVI. Int. J. Remote Sens. 25, 4309–4324. Shimabukuro, Y.E., Batista, G.T., Mello, E.M.K., Moreira, J.C., Duarte, V., 1998. Using shade fraction image segmentation to evaluate deforestation in Landsat Thematic Mapper images of the Amazon Region. Int. J. Remote Sens. 19, 535–541. Simova, I., Li, Y.M., Storch, D., 2013. Relationship between species richness and productivity in plants: the role of sampling effect, heterogeneity and species pool. J. Ecol. 101, 161–170. Solbrig, O.T., Medina, E., Silva, J., 1996. Biodiversity and tropical savanna properties: a global view. Scope Sci. Comm. Prob. Environ. Int. Counc. Sci. Unions 55, 185–211. Somers, B., Asner, G.P., Martin, R.E., Anderson, C.B., Knapp, D.E., JosephWright, S., Kerchove, R.V.D., 2015. Mesoscale assessment of changes in tropical tree species richness across a bioclimatic gradient in Panama using airborne imaging spectroscopy. Remote Sens. Environ. 167, 111–120. St Louis, V., Pidgeon, A.M., Clayton, M.K., Locke, B.A., Bash, D., Radeloff, V.C., 2009. Satellite image texture and a vegetation index predict avian biodiversity in the Chihuahuan Desert of New Mexico. Ecography 32, 468–480. Sutherland, W.J., Krebs, C.J., 1997. Ecological census techniques. Trends Ecol. Evol. 12, 81–81.
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317 Thiollay, J.-m., 1997. Disturbance, selective logging and bird diversity: a Neotropical forest study. Biodivers. Conserv. 6, 1155–1173. Trapnell, C.G., Martin, J. D., Allan, W., 2001. Vegetation-soil Map of Zambia. Royal Botanic Gardens. Tucker, C.J., Sellers, P.J., 1986. Satellite remote sensing of primary production. Int. J. Remote Sens. 7, 1395–1416. Turner, W., Rondinini, C., Pettorelli, N., Mora, B., Leidner, A.K., Szantoi, Z., Buchanan, G., Dech, S., Dwyer, J., Herold, M., Koh, L.P., Leimgruber, P., Taubenboeck, H., l, M.W., Wikelskim, M., Woodcock, C., 2015. Free and open-access satellite data are key to biodiversity conservation. Biological Conservatio. Biol. Conserv., 173–176. Van Wilgen, B., 2009. The evolution of fire management practices in savanna protected areas in South Africa. S. Afr. J. Sci. 105, 343–349. Waide, R., Willig, M., Steiner, C., Mittelbach, G., Gough, L., Dodson, S., Juday, G., Parmenter, R., 1999. The relationship between productivity and species richness. Annu. Rev. Ecol. Syst., 257–300.
317
Wiens, J., Sutter, R., Anderson, M., Blanchard, J., Barnett, A., Aguilar-Amuchastegui, N., Avery, C., Laine, S., 2009. Selecting and conserving lands for biodiversity: the role of remote sensing. Remote Sens. Environ. 113, 1370–1381. Williams, V.L., Witkowski, E.T., Balkwill, K., 2005. Application of diversity indices to appraise plant availability in the traditional medicinal markets of Johannesburg, South Africa. Biodivers. Conserv. 14, 2971–3001. Zellweger, F., Braunisch, V., Baltensweiler, A., Bollmann, K., 2013. Remotely sensed forest structural complexity predicts multi species occurrence at the landscape scale. For. Ecol. Manag. 307, 303–312. Zhang, C., Lu, D., Chen, X., Zhang, Y., Maisupova, B., Tao, Y., 2016. The spatiotemporal patterns of vegetation coverage and biomass of the temperate deserts in Central Asia and their relationships with climate controls. Remote Sens. Environ. 175, 271–281.
Contents lists available at ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
The relationship between satellite-derived indices and species diversity across African savanna ecosystems Ratidzo B. Mapfumo a,∗ , Amon Murwira a , Mhosisi Masocha a , R Andriani b a b
Department of Geography and Environmental Science, University of Zimbabwe, P.O. Box MP 167, Harare, Zimbabwe Centre for International Forestry Research (CIFOR), P.O. Box 0113 BOCBD, Bogor 16000, Indonesia
a r t i c l e
i n f o
Article history: Received 3 February 2016 Received in revised form 27 May 2016 Accepted 29 June 2016 Keywords: Species diversity Savanna Remote sensing Spectral variation hypothesis
a b s t r a c t The ability to use remotely sensed diversity is important for the management of ecosystems at large spatial extents. However, to achieve this, there is still need to develop robust methods and approaches that enable large-scale mapping of species diversity. In this study, we tested the relationship between species diversity measured in situ with the Normalized Difference Vegetation Index (NDVI) and the Coefficient of Variation in the NDVI (CVNDVI) derived from high and medium spatial resolution satellite data at dry, wet and coastal savanna woodlands. We further tested the effect of logging on NDVI along the transects and between transects as disturbance may be a mechanism driving the patterns observed. Overall, the results of this study suggest that high tree species diversity is associated with low and high NDVI and at intermediate levels is associated with low tree species diversity and NDVI. High tree species diversity is associated with high CVNDVI and vice versa and at intermediate levels is associated with high tree species diversity and CVNDVI. © 2016 Elsevier B.V. All rights reserved.
1. Introduction Savanna ecosystems constitute approximately 20% of the area of world’s terrestrial ecosystems (Sankaran and Anderson, 2009; Van Wilgen, 2009). One-fifth of the world’s population depends on the savanna ecosystem (Solbrig et al., 1996; Parr et al., 2014). These ecosystems also contain 5–20% of the world’s herbivore biomass (Owen-Smith, 1993; Cowling et al., 2004; Sankaran et al., 2005). Thus, the ability to understand patterns of species diversity in savanna ecosystems, especially at large spatial extents is important for their efficient management and conservation. In this regard, the need for methods that can achieve this large scale mapping is critical. The development of satellite remote sensing has provided an opportunity to map diversity at large spatial extents (Nagendra et al., 2010; Rocchini et al., 2015). In fact, remotely sensed data provide an effective and evident way to address ecological patterns at different multiple scales (Simova et al., 2013). However species distribution patterns are easier to map at a broader scale compared to fine-scale distributions (Kerr and Ostrovsky, 2003; Gillman et al., 2015). To this end, several studies have used remote sensing to map
∗ Corresponding author. E-mail address: [email protected] (R.B. Mapfumo). http://dx.doi.org/10.1016/j.jag.2016.06.025 0303-2434/© 2016 Elsevier B.V. All rights reserved.
species diversity in ecosystems (Scheiner and Jones, 2002; Rocchini et al., 2010b; Hernandez-Stefanoni et al., 2015). The main drawback with most of these studies has been their focus on one type of ecosystem, thus making generalizations difficult or impossible. Development of general understanding of the relationship between remotely sensed indices and diversity is, therefore imperative for enhancing the understanding of diversity in ecosystems. Savannas particularly exist in different states depending on moisture regimes (Devine et al., 2015). For example, Southern African savannas exist as dry savanna, wet savanna and coastal savannas (Scholes and Archer, 1997; Scholes and Walker, 2004; Devine et al., 2015). There are differences in species composition, structure and ecological processes between dry savannas, wet savannas and coastal savannas (Justice 1994). Thus, developing remote sensing models that can be applied generally across various ecosystems is crucial. The application of remote sensing in understanding ecological patterns in savanna woodlands principally depends on the suitability of the remotely sensed indices used to relate to ground measurements of biological diversity (Mutowo and Murwira, 2012; Hernandez-Stefanoni et al., 2015) as well as on whether these indices incorporate the range of variation in the savanna ecosystem. Satellite-derived vegetation indices, such as the Normalized Difference Vegetation Index (NDVI) have been shown to be useful estimates of productivity and can also be used to quantify vegetation-related spatial heterogeneity thereby shap-
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
307
Table 1 Description of the seven study sites, the soil data of Zimbabwe (Kutsaga, Shurugwi) was extracted from (Nyamapfene 1991), the soil data of Zambia (Kaoma, Kasenu, Simungoma) were extracted from (Trapnell et al., 2001) and the soil data of Mozambique (Miti, Mofid) were extracted from (Maria and Yost, 2006). Site
Location
Mean Annual Rainfall (mm)
Mean Annual Temperature
Soils
Vegetation data
Kutsaga
17◦ 55 S 31◦ 08 E
850
18.6 ◦ C
Ferallic cambisols
Shurugwi
19◦ 58 S 29◦ 51 E
800
17.6 ◦ C
Lithic leptosols
Kaoma
15◦ 70 S 24◦ 45 E
1100
21.5 ◦ C
Ferrallic arenosols
Kasenu Simungoma
15◦ 30 S 25 30 E
1000
27.7 ◦ C
Miombo woodlands dominated by Julbernadia globiflora, and Brachystegia spiciformis Miombo woodlands dominated by Julbernardia globiflora, Msasa Brachystegia spiciformis and Terminalia. Miombo woodlands characterized by Brachystegia, Julbenardia paniculta and Marquesia marcroura Isoberlinia Angolensis, The Kalahari woodland dominated by Cryptosepalum Miombo woodland dominated by Dry Deciduous Thicket with Guibourtia schliebenii Sclerophyllous species which include Manilkara sansibarensis, Warneckea sansibarica and Baphia macrocalyx
◦
Miti
11 39 S 39 33 E
1000
31.7 C
Ferrallic/ cambic arenosols Ultisols
Mofid
11◦ 43 S 39◦ 47 E
1400
26 ◦ C
Oxisols
◦
◦
◦
ing biodiversity patterns (Tucker and Sellers, 1986; Thiollay, 1997; Shimabukuro et al., 1998; Loboda et al., 2013; Girma et al., 2016). In addition, the variance in NDVI has also been used as a useful measure of diversity in ecosystems (Bongers et al., 2009). The variance in NDVI as an index for estimating diversity in ecosystems is well supported by the spectral variation hypothesis. The theory of Spectral Variation proposes that spatial variation on a remotely sensed image is related to spatial variations of the environment which reflects habitat heterogeneity (Rocchini et al., 2010a; Heumanna et al., 2015). Therefore, habitat heterogeneity is linked to the structural complex of habitats which may provide environmental resources leading to an increase in species diversity (Oldeland et al., 2010). Although, variance in NDVI has been used to successfully characterize tree species diversity in the dry savanna woodlands of Mukuvisi, Mutirikwi and Mabalauta in Zimbabwe (Mutowo and Murwira, 2012), this work did not include various types of savanna such as wet savanna sites, as well as, coastal savannas. Thus, the inclusion of different savanna sites in modeling the relationship between species diversity and remotely sensed indices could improve our capacity to map diversity at large spatial extents. Therefore, in this study, we test the hypothesis that tree species diversity and satellite-derived indices of NDVI should be positively correlated, and that tree species diversity will be positively correlated with the Coefficient of Variation of NDVI. The use of remote sensing in ecology is vital as it has been widely used to identify spatial and temporal patterns in biodiversity in different ecosystems over a period of time (Wiens et al., 2009). Remotely sensed spectral heterogeneity information also offers an inexpensive means to derive spatially complete environmental information for large areas in a consistent and systematic manner (Levin et al., 2007; Rocchini et al., 2010b). Ecologists may gain critical knowledge about the drivers of the spatial and temporal distribution of biodiversity at any given time (Pettorelli et al., 2014). Through the use remote sensed data, we can understand the positive as well as the negative impacts on biodiversity, predictions about the future and action to prevent ecological degradation strategy to mitigate adverse the impacts observed (Dodson et al., 2000; Zellweger et al., 2013). Principled and appropriate forest management can lead to increased biological diversity (Parma and Shataee, 2013).Explaining the mechanisms driving the observed relationships is of fundamental importance to understanding the determinants of biodiversity (Mittelbach et al., 2001).The biodiversity patterns observed may suggest dynamic, nonequilibrium community processes encountered in ecosystems (Graham and
Duda, 2011). Remote sensing approaches may provide planners and conservation biologists with an efficient and cost-effective method to study and estimate biodiversity different ecosystems (Levin et al., 2007). In this study, we tested the relationship between species diversity measured in situ with the NDVI and CVNDVI derived from high and medium spatial resolution satellite data in dry, wet and coastal savanna woodlands. We selected the study sites in the dry savanna of Zimbabwe, wet savanna of Zambia and coastal savanna in Mozambique because they experience different climatic conditions which as a result influence different ecological patterns. 2. Materials and methods 2.1. Study area The study was carried out at 7 study sites (Fig. 1a): Zimbabwe (Kutsaga and Shurugwi), Zambia (Kaoma, Sesheke Simungoma, and Sesheke Kasenu) and Mozambique (Miti and Mofid) (Table 1). Each of the six study sites covered 100 km2 in area. 3. Field data 3.1. Tree species data We randomly selected six transects using ArcView GIS 3.2 (ESRI, 2002) in each study site. The transects had a minimum length of 3 km and a maximum length of 6 km. Transect length was determined on how accessible were the sampling points, transects with a maximum length of 6 km were characterized by rugged terrain which makes the sampling points to be inaccessible (Marshall and Region, 2000). The starting point and the ending point of each transect were navigated using a handheld Global Positioning System (GPS) receiver. The locational accuracy of the field plots was within the 15 m error of the GPS. We defined sample plots of 15 m × 15 m which were north oriented at 500 m distances along each transect and in each sample plot, species names were recorded. Research has shown that sampling plot sizes widely used ranges between 25 and 200 m2 in tall shrub communities and 200–25000 m2 for trees in woods and forests (Sutherland and Krebs, 1997) and our plot size of 225 m2 falls within this range. A sampling distance of 500 m between the sampling plots was used because it is a distance that can capture the spatial variations in forest structure, species composition and vegetation density thereby reducing uniformity
308
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Mozambique
N
Zambia
0
0
400
800
1200 Kilometers
60
120
180 Kilometers
Zimbabwe
Study sites Kutsaga Shurugwi Kaoma Kasenu Simugoma Miti Mofid
0
200
400
600 Kilometers
Fig. 1a. The location of the study sites.
on plots (Lutz et al., 2013). Tree species that could not be identified in the field were taken for identification at the Herbariums in Lusaka for the Zambian sites, in Harare for the Zimbabwean sites and Maputo in Mozambique. Tree species data were collected in February 2012 and October 2012 in Shurugwi and Kutsaga respectively, in October 2013 in Sesheke Simungoma, Sesheke Kasenu, Kaoma, Miti and Mofid.
Table 2 The description of the dates acquired for Landsat 8 and Worldview-2 satellite images for the sites in Zimbabwe, Zambia and Mozambique. Country
Site
Date Acquired Landsat 8
Date Acquired Worldview-2
Zimbabwe
Kutsaga Shurugwi Kaoma Simungoma Kasenu Miti Mofid
11 July 2013 11 July 2013 07 July 2013 07 July 2013 07 July 2013 19 July 2013 19 July 2013
24 July 2012 24 July 2012 16 July 2012 30 June 2013 16 July 2013 – –
Zambia
Mozambique
3.2. Measures of diversity In this study, we used Shannon’s H and Simpson’s 1-D as measures of diversity. Simpson’s diversity index is based on the probability of any two individuals drawn at random from an infinity large community belonging to the same species (DeJong, 1975). The index ranges from 0 to 1 in ascending order with increased diversity with a greater probability of interspecific encounter (Magurran, 2013). Shannon’s diversity index is a measure of the average ‘uncertainty’ of predicting to what species an individual is chosen at random from a collection of species belong (Williams et al., 2005; Magurran 2013). There is an increase in the average uncertainty as the number of species also increases and the distribution of individuals among the species become more even (DeJong, 1975; Ludwig and Reynolds, 1988). In this study, we selected the Simpson diversity index (1-D) (equation 1) and Shannon diversity index (H) (equation 2) because these indices combine evaluations of abundance and evenness of the species present in an ecosystem (Elzinga and Evenden, 1997; Oldeland et al., 2010; Rocchini et al., 2010b; Duro et al., 2014). In addition, the Shannon’s diversity index is less affected by the presence of rare species (Nagendra, 2002).The Simpson’s (1-D) index of diversity is calculated as follows:
1−D=
n i=1
pi2
(1)
where pi is the proportion of the ith species in the sampling plot. The Shannon-Weaver index is calculated as follows: H=
n
− pi log pi
(2)
i=1
where pi is the proportion of the ith species in the sampling plot. 3.3. Remotely sensed data In this study, we used Landsat 8 imageries which are made freely available by the United States Geological Survey (USGS) and the imageries were downloaded from http://glovis.usgs.gov/ . We also used Worldview -2 imagery as a source of multispectral remotely sensed data (Table 2). Digital numbers (DN) of Landsat 8 satellite imagery were converted to top-of-atmosphere spectral reflectance in ENVI 5.1 image processing software (ITT, 2013) using the reflectance rescaling coefficients provided in the metadata files of the images while the Quick Atmospheric Correction (QUAC) algorithm was applied in ENVI 4.8 (ITT, 2008) for Worlview-2. We also Pan Sharpened the Landsat 8 imagery to 15 m × 15 m to coincide with the field plot size of 15 m × 15 m. Landsat 8 satellite images are associated with geometric errors of 12 m in CE90 which mean that 90% of the points has less than 12 m positional error (Roy et al.,
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
309
Fig. 1b. The spatial variations in canopy structure estimated by NDVI derived from Landsat 8 in (a) Kutsaga, (b) Shurugwi, (c) Kaoma, (d) Kasenu, (e) Simungoma, (f) Miti and (g) Mofid. The map units are Geographic coordinates.
2014).The variations in canopy structure was estimated by NDVI derived from Landsat 8 (Fig. 1b).
3.4. Remotely sensed indices NDVI derived from Landsat 8 and Worldview-2 was calculated in GIS by subtracting the value of the red band from the value of the NIR band and divide it by the sum of red and NIR band values (Eidenshink, 1992). The NDVI has a range of −1 to +1. NDVI was selected as an index since it is known to represent the greenness of active biomass (DeFries and Townshend, 1994; Parviainen et al.,
2010; Hall et al., 2012). In addition, NDVI has widely been used in the study of biodiversity in landscapes of spatial heterogeneity (Santin-Janin et al., 2009; Cui et al., 2013; Zhang et al., 2016). Next, we calculated the Coefficient of Variation in the NDVI (CVNDVI) derived from Landsat 8 and Worldview 2 images by dividing the standard deviation of NDVI with the average NDVI, expressed as a percentage using a moving window of 3pixels by 3 pixels. CVNDVI has been used in diversity studies and has been proven to be a simple and effective indicator of spectral heterogeneity (Lucas and Carter, 2008; Duro et al., 2014; Somers et al., 2015). The Coefficient of Variation was applied in the Great Basin
310
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Fig. 2. Relationships between tree species diversity estimated using Shannon’s H index and the Normalized Difference Vegetation Index (NDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
of western North America (Seto et al., 2004), to characterize the relationship between species diversity of butterflies and remotely sensed diversity. 4. Data analysis The mean NDVI, as well as mean CVNDVI, computed from Landsat-8 and Worldview 2 were regressed against Shannon’s H and Simpson’s 1-D diversity indices calculated from field data yielding a total of 10 relationships tested. Worldview-2 imagery for the Mozambican site was not available and therefore the imageries were not used in this study. Due to the curvature inherent in the diversity-productivity relationships, linear and non-linear regression based on the second-order polynomial model was used to test the consistency of the relationships across different ecosystems in this study. We tested whether NDVI, CVNDVI and tree diversity data followed a normal distribution using the Kolmogorov- Smirnov test and the results indicated that our data followed a normal distribution. The strength and significance of the relationship between field-based indices and satellite-derived indices were evaluated
using the R2 and P-values, respectively. We further tested the effect of disturbance on NDVI derived from Landsat along the transects and between transects of the same site using students t-test and two-way ANOVA respectively. Prior to analysis, we tested whether NDVI data significantly deviated from a normal distribution using Kolmogorov-Smirnov test and the results showed that the data did not significantly deviate from a normal distribution thus parametric tests could be used.
5. Results A consistent and significant U-shaped relationship is observed (Fig. 2) between Shannon’s H index of diversity and NDVI at the dry savanna of Zimbabwe regardless of whether NDVI was calculated from Landsat 8 or Worldview-2 and at the wet savanna sites derived from Worldview-2 a significant U-shaped relationship is observed. There is a positive linear and significant relationship observed at the wet savanna sites of Zambia and coastal savanna sites of Mozambique derived from Landsat 8. However, the relationship between
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
311
Fig. 3. Relationships between tree species diversity estimated using Simpson’s 1-D index and the Normalized Difference Vegetation Index (NDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
Shannon’s H index of diversity and NDVI is observed to be stronger when using Worldview-2 compared to Landsat-8. Fig. 3 shows that a is consistent U-shaped relationship between Simpson’s 1-D diversity index and NDVI derived from Landsat 8 and Worldview-2 at the dry savanna sites of Zimbabwe, wet savanna sites of Zambia and coastal savanna sites of Mozambique. A hump-shaped and significant relationship is observed between Shannon’s H diversity index and remotely sensed diversity (CVNDVI) at the dry savanna of Zimbabwe (Fig. 4) using remotely sensed diversity (CVNDVI) derived from both Landsat 8 and Worldview-2 and at the wet savanna of Zambia using Landsat 8 derived index. However, a positive linear relationship is observed at the wet savanna sites of Zambia and coastal savanna of Mozambique using remotely sensed diversity (CVNDVI) derived from Worldview-2 and Landsat 8 respectively. Hence, the results also indicate no significant relationship for the sites in Mozambique. We also observe a hump-shaped significant relationship between Simpson’s 1-D diversity index and remotely sensed diversity (CVNDVI) at the dry savanna of Zimbabwe (Fig. 5). At the
wet savanna sites of Zambia, we observe a hump-shaped and positive linear relationship between Simpson’s 1-D diversity index and remotely sensed diversity (CVNDVI) derived from both Landsat 8 and Worldview-2 respectively. A positive linear relationship at the sites in Mozambique using CVNDVI derived from Landsat 8 is observed. Hence, the results indicate no significant relationship for the sites in Mozambique. Regardless of the different patterns observed, the relationship between Simpson’s 1-D diversity index and remotely sensed diversity (CVNDVI) is stronger using Worldview-2.
5.1. Effect of logging on NDVI derived from Landsat 8 between logged and unlogged areas along and between transects in Zimbabwe, Zambia and Mozambique The overall results show no significant differences (p > 0.05) between logged and unlogged areas in mean NDVI along the transects (Table 3). However, there are significant differences (p < 0.05) between logged and unlogged areas along the transects in Kutsaga (transect 1and 2), Simungoma (transect 1) and Mofid (transect 1).
312
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Table 3 NDVI derived from Landsat 8 (mean ± standard error) in logged and unlogged plots along the transects in Zimbabwe, Zambia and Mozambique. Country
Site
Transect
Logged
t statistic
p-value
Zimbabwe
Kutsaga
1 2 3 4 5 6 1 2 3 4 5 6
0.36 0.46 0.35 0.43 0.44 0.42 0.62 0.70 0.66 0.59 0.67 0.63
0.03 0.014 0.03 0.06 0.02 0.05 0.09 0.07 0.02 0.14 0.07 0.04
0.5 0.29 0.34 0.31 0.38 0.31 0.67 0.55 0.62 0.40 0.62 0.61
± ± ± ± ± ± ± ± ± ± ± ±
0.05 0.2 0.03 0.02 0.003 0.03 0.02 0.08 0.05 0.05 0.01 0.04
−2.908 5.816 0.162 1.414 1.851 1.512 −0.374 1.439 0.703 0.952 −0.755 1.311
0.04* 0.01* 0.89 0.25 0.16 0.35 0.73 0.22 0.56 0.41 0.53 0.25
1 2 3 4 1 2 3 4 5 1 2 3 4 5
0.22 ± 0.04 0.24 ± 0.03 0.24 ± 0.02 0.18 ± 0.02 0.36 ± 0.001 0.34 ± 0.016 0.30 ± 0.039 0.36 ± 0.02 0.34 ± 0.02 0.33 ± 0.01 0.27 ± 0.04 0.33 ± 0.01 0.35 ± 0.03 0.38 ± 0.003
0.20 0.26 0.25 0.23 0.29 0.33 0.31 0.35 0.32 0.34 0.28 0.33 0.36 0.32
± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.05 0.02 0.05 0.05 0.02 0.03 0.06 0.06 0.01 0.02 0.008 0.03 0.12 0.03
0.225 −0.545 −0.222 −0.911 4.051 0.479 −0.114 −0.133 0.718 0.605 −0.215 0.225 −0.405 0.623
0.84 0.64 0.85 0.46 0.03* 0.68 0.92 0.90 0.53 0.59 0.85 0.84 0.73 0.60
1 2 3 1 2 3 4 5
0.63 0.38 0.45 0.25 0.19 0.19 0.23 0.47
± ± ± ± ± ± ± ±
0.45 0.37 0.49 0.27 0.35 0.21 0.32 0.47
± ± ± ± ± ± ± ±
0.38 0.07 0.008 0.05 0.08 0.09 0.09 0.03
4.908 0.055 −3.000 −0.298 −1.285 −0.138 −0.813 −0.041
0.02* 0.96 0.05 0.79 0.33 0.9 0.5 0.97
Shurugwi
Zambia
Kaoma
Simungoma
Kasenu
Mozambique
Mofid
Miti
*
± ± ± ± ± ± ± ± ± ± ± ±
Unlogged
0.02 0.15 0.009 0.05 0.06 0.07 0.07 0.03
<0.05.
The results from ANOVA showed no significant differences in NDVI derived from Landsat 8 (Fig. 6, Table 4) between logged and unlogged areas (p > 0.05) between the transects of the same site except for the site in Zimbabwe where there is a significant difference in NDVI between logged and unlogged areas (p < 0.05) between the transects. There is inconsistence in mean NDVI between logged and unlogged plots between the transects.
Table 4 The differences in NDVI derived from Landsat 8 between Logged and unlogged areas between transects in the sites of Zimbabwe, Zambia and Mozambique. Country
Site
F value
p-value
Zimbabwe
Kutsaga Shurugwi Kaoma Kasenu Simungoma Miti Mofid
5.448 0.756 0.243 0.244 0.421 0.313 2.113
0.001* 0.57 0.86 0.91 0.79 0.86 0.18
Zambia
6. Discussion
Mozambique
The results of this study suggest the existence of a U-shaped relationship between tree species diversity and NDVI at dry and wet savanna sites in Zimbabwe and Zambia, respectively. The results of this study are consistent with the findings where a U-shaped relationship between diversity and NDVI was found in semi natural savanna grassland in Sweden (Nagendra et al., 2013). In addition, U-shaped relationships between NDVI and diversity in Mukuvisi, Mutirikwi and Mabalauta sites in Zimbabwe were observed (Mutowo and Murwira, 2012). The mechanisms which explain this pattern are not clear. However, a proposed explanation of such a pattern can arise when the green cover of the competitively dominant tree and shrub tree species occur at intermediate times after disturbance (Gosper et al., 2013). In this regard, increased biomass production after logging may be triggering the U-shaped relationships. However, human induced disturbance of logging may not be the mechanism explaining the U-shaped relationships observed since there is no significant difference in NDVI between logged and unlogged areas along and between the transects of the study sites. Further research targeting non-logged (protected sites) adjacent to logged sites is needed to validate the effect of logging on the ecological patterns especially the U-shaped pat-
*
<0.05.
tern. Although these results are inconsistent with previous research where hump-shaped relationships between NDVI and tree diversity were reported (Waide et al., 1999), we deduce that disturbance may not be a mechanism driving U-shaped relationships in savanna ecosystems. There are consistent relationships between diversity and CVNDVI derived from Landsat 8 Worldview-2 in the study sites of Zimbabwe and Mozambique. The linear relationships observed in Zambia and Mozambique can be explained by the fact that wet and coastal ecosystems receive high rainfall above 1000 mm leading to high diversity which facilitates high CVNDVI (Kerr and Ostrovsky, 2003; Glenn et al., 2008; Fatehi et al., 2015). These results are interesting as they depict the importance of high spatial resolution in predicting species diversity in wet and coastal climates. Thus, results suggest that the use of high spatial resolution imagery may be important to allow for the capturing of more detailed information on the spectral variability of landscape features consistent with the spectral variability hypothesis (Rocchini et al., 2010b;
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
313
Fig. 4. Relationships between tree species diversity estimated using Shannon’s H index and the Coefficient of Variation in the Normalized Difference Vegetation Index (CVNDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
Turner et al., 2015).Our results also indicate the existence of a hump-shaped relationship between the tree species diversity and Coefficient of Variation of NDVI observed at both dry and wet savanna sites in Zimbabwe and Zambia, respectively. This result is consistent with the findings of a study conducted at Mount Hermon, located in NE Israel and a savanna ecosystem of Kenya where humpshaped relationships between species diversity and CVNDVI (Oindo and Skidmore, 2002; Levin et al., 2007) was observed. Thus, we deduce that we can predict species diversity using CVNDVI derived from both high and medium satellite resolution data. The results of this study further suggest stronger relationships between tree species diversity against NDVI and CVNDVI derived from Worldview-2. This could be a result of the higher spatial resolution of Worldview-2 imagery with a spatial resolution of ∼2 m compared with the Landsat 8 imagery with a spatial resolution of ∼30m. These results are not surprising as they are consistent with other findings where relationships between Landsat derived indices and diversity were found to be weaker than those derived
from Worldview-2 (Nagendra and Rocchini, 2008; Hall et al., 2012; Karteris et al., 2016). However, due to high expenses involved in acquiring high spatial resolution data; their use in spatial ecology remains limited. To this end, most vegetation diversity assessment studies use Landsat data that are more readily available and useful (St Louis et al., 2009; Oldeland et al., 2010). The high spatial resolution images such as Worldview-2 can be used for calibration and validation of regional remote sensing based on species diversity mapping (Immitzer et al., 2012; Heenkenda et al., 2014; Cho et al., 2015). This is because Worldview-2 images are not readily available (commercial) and costly while Landsat imageries are freely available. Thus, we deduce that although high spatial resolution satellite data are capable of predicting species diversity at finer scales which is important when monitoring savanna ecosystems, the use of medium spatial resolution imagery such as Landsat still remains invaluable. Where our study differs from previous studies (Levin et al., 2007; Saatchi et al., 2008; Rocchini et al., 2010b; Hall et al., 2012;
314
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Fig. 5. Relationships between tree species diversity estimated using Simpson’s 1-D index and the Coefficient of Variation in the Normalized Difference Vegetation Index (CVNDVI) derived from Landsat-8 (left panels) and Worldview-2 (right panels) at dry savanna sites in Zimbabwe (a, b), wet savanna sites in Zambia (c, d) and at the coastal savanna sites in Mozambique (e).
Turner et al., 2015) is the testing of the relationships between species diversity and satellite derived indices across three savanna ecosystems, i.e., dry, wet and coastal savanna woodlands. Previous research has reported the findings of the ecological patterns from single sites. For example, the relationship between diversity measures and spectral variability using hyperspectral remote sensing data has been tested on a single semi-arid savanna ecosystem of Central Namibia (Oldeland et al., 2010). In a study conducted in the southeast of the United States (Costanza et al., 2011), they also characterized the relationship between species diversity and remotely sensed data in a single coastal savanna plain. In a savanna ecosystem located to the southeast of Australia (Schmidt et al., 2015) also assessed the relationship between species diversity and satellite derived indices. In this regard, we deduce that the relationship between diversity and remotely sensed diversity becomes more robustly understood when applicable across different ecosystems. The main limitation of this study is that the findings are based on a single season and that the satellite data used was collected
when tree species were still in leaf (Chidumayo, 1997) while the grass species had senesced for all the study sites. Again, the sample size used in this present study is inadequate, therefore we recommend that future studies may increase the sample size so that strong conclusions on the prediction of species diversity can be made. Therefore, future studies could use remotely sensed data as well as tree inventory data at different times of the year in order to further our understanding of the patterns observed.
7. Conclusion From the results of this study, Firstly, we concluded that tree species diversity can be predicted using NDVI and the CVNDVI derived from both high and medium spatial resolution satellite data across different savanna ecosystems, i.e., dry, wet and coastal savanna woodlands. Secondly, we concluded that although the mechanism explaining the relationship between species diversity and NDVI is not well understood, however, high tree species diver-
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
315
Fig. 6. Mean NDVI derived from Landsat 8 along the transects between logged and unlogged areas in a) Kutsaga b) Shurugwi c) Kaoma d) Kasenu e) Simungoma f) Miti g) Mofid.
sity is associated with low and high NDVI and at intermediate levels is associated with low tree species diversity and NDVI. Thirdly, we also concluded that high tree species diversity is associated with high CVNDVI and vice versa and at intermediate levels is associated with high tree species diversity and CVNDVI. Finally, we concluded that although results from high and medium spatial resolution satellite indices gave consistent results, high spatial resolution imagery provided comparatively better results.
Acknowledgement We thank the Centre for International Forestry Research (CIFOR) for their financial support thus making data collection possible. Their dedication to developing policies and technologies for sustainable use and management of forests is appreciated.
References Bongers, F., Poorter, L., Hawthorne, W.D., Sheil, D., 2009. The intermediate disturbance hypothesis applies to tropical forests, but disturbance contributes little to tree diversity. Ecol. Lett. 12, 798–805. Chidumayo, E.N., 1997. Miombo Ecology and Management: An Introduction. Intermediate Technology Publications Ltd (ITP). Cho, M.A., Malahlela, O., Ramoelo, A., 2015. Assessing the utility WorldView-2 imagery for tree species mapping in South African subtropical humid forest and the conservation implications: Dukuduku forest patch as case study. Int. J. Appl. Earth Obs. Geoinf. 38, 349–357. Costanza, J.K., Moody, A., Peet, R.K., 2011. Multi-scale environmental heterogeneity as a predictor of plant species richness. Landscape Ecol. 26, 851–864. Cowling, R.M., Richardson, D.M., Pierce, S.M., 2004. Vegetation of Southern Africa. Cambridge University Press. Cui, X., Gibbes, C., Southworth, J., Waylen, P., 2013. Using remote sensing to quantify vegetation change and ecological resilience in a semi-arid system. Land 2, 108–130. DeFries, R., Townshend, J., 1994. NDVI-derived land cover classifications at a global scale. Int. J. Remote Sens. 15, 3567–3586. DeJong, T.M., 1975. A comparison of three diversity indices based on their components of richness and evenness. Oikos 26, 222–227.
316
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317
Devine, A.P., Stott, I., McDonald, R.A., Maclean, I., 2015. Woody cover in wet and dry African savannas after six decades of experimental fires. J. Ecol. 103, 473–478. Dodson, S.I., Arnott, S.E., Cottingham, K.L., 2000. The relationship in lake communities between primary productivity and species richness. Ecology 81, 2662–2679. Duro, D.C., Girard, J., King, D.J., Fahrig, L., Mitchell, S., Lindsay, K., Tischendorf, L., 2014. Predicting species diversity in agricultural environments using Landsat TM imagery. Remote Sens. Environ. 144, 214–225. ESRI, 2002. ArcView GIS 3.2. Environmental Systems Research Institute, California. Eidenshink, J.C., 1992. The 1990 conterminous U. S. AVHRR data set. Photogramm. Eng. Remote Sens. 58, 809–813. Elzinga, C.L., Evenden, A.G., 1997. Vegetation Monitoring: an Annotated Bibliography. DIANE Publishing. Fatehi, P., Damm, A., Schaepman, M.E., Kneubühler, M., 2015. Estimation of alpine forest structural variables from imaging spectrometer data. Remote Sens. 7, 16315–16338. Gillman, L.N., Wright, S.D., Cusens, J., McBride, P.D., Malhi, Y., Whittaker, R.J., 2015. Latitude, productivity and species richness. Global Ecol. Biogeogr. 24, 107–117. Girma, A., de Bie, C., Skidmore, A.K., Venus, V., Bongers, F., 2016. Hyper-temporal SPOT-NDVI dataset parameterization captures species distributions. Int. J. Geogr. Inf. Sci. 30, 89–107. Glenn, E.P., Huete, A.R., Nagler, P.L., Nelson, S.G., 2008. Relationship between remotely-sensed vegetation indices, canopy attributes and plant physiological processes: what vegetation indices can and cannot tell us about the landscape. Sensors 8, 2136–2160. Gosper, C.R., Yates, C.J., Prober, S.M., 2013. Floristic diversity in fire-sensitive eucalypt woodlands shows a ‘U’-shaped relationship with time since fire. J. Appl. Ecol. 50, 1187–1196. Graham, J.H., Duda, J.J., 2011. The humpbacked species richness-curve: a contingent rule for community ecology. Int. J. Ecol. Hall, K., Reitalu, T., Sykes, M.T., Prentice, H.C., 2012. Spectral heterogeneity of QuickBird satellite data is related to fine scale plant species spatial turnover in semi-natural grasslands. Appl. Veg. Sci. 15, 145–157. Heenkenda, M.K., Joyce, K.E., Maier, S.W., Bartolo, R., 2014. Mangrove species identification: comparing WorldView-2 with aerial photographs. Remote Sens. 6, 6064–6088. Hernandez-Stefanoni, J.L., Johnson, K.D., Cook, B.D., Dupuy, J.M., Birdsey, R., Peduzzi, A., Tun-Dzul, F., 2015. Estimating species richness and biomass of tropical dry forests using LIDAR during leaf-on and leaf-off canopy conditions. Appl. Veg. Sci. 50, 1–9. Heumanna, B.W., Hackettb, R.A., Monfilsb, A.K., 2015. Testing the spectral diversity hypothesis using spectroscopy data in a simulated wetland community. Ecol. Inf. 25. ITT, 2008. ENVI 4.8. ITT Industries Inc., Colorado. ITT, 2013. ENVI 5.1. ITT Industries Inc., Colorado. Immitzer, M., Atzberger, C., Koukal, T., 2012. Tree species classification with random forest using very high spatial resolution 8-band WorldView-2 satellite data. Remote Sens. 4, 2661–2693. Karteris, M., Theodoridou, I., Mallinis, G., Tsiros, E., Karteris, A., 2016. Towards a green sustainable strategy for Mediterranean cities: assessing the benefits of large-scale green roofs implementation in Thessaloniki, Northern Greece, using environmental modelling, GIS and very high spatial resolution remote sensing data. Renew. Sustain. Energy Rev. 58, 510–525. Kerr, J.T., Ostrovsky, M., 2003. From space to species: ecological applications for remote sensing. Trends Ecol. Evol. 18, 299–305. Levin, N., Shmida, A., Levanoni, O., Tamari, H., Kark, S., 2007. Predicting mountain plant richness and rarity from space using satellite-derived vegetation indices. Divers. Distrib. 13, 692–703. Loboda, T.V., Zhang, Z., O’Neal, K.J., Sun, G., Csiszar, I.A., Shugart, H.H., Sherman, N.J., 2013. Reconstructing disturbance history using satellite-based assessment of the distribution of land cover in the Russian Far East. Remote Sens. Environ. 118, 241–248. Lucas, K.L., Carter, G.A., 2008. The use of hyperspectral remote sensing to assess vascular plant species richness on Horn Island, Mississippi. Remote Sens. Environ. 112, 3908–3915. Ludwig, J.A., Reynolds, J.F., 1988. Statistical Ecology: A Primer in Methods and Computing. John Wiley & Sons. Lutz, J.A., Larson, A.J., Freund, J.A., Swanson, M.E., Bible, K.J., 2013. The importance of large-diameter trees to forest structural heterogeneity. PLoS One 8, e82784. Magurran, A.E., 2013. Ecological Diversity and Its Measurement. Springer Science & Business Media. Maria, R.M., Yost, R., 2006. A survey of soil fertility status of four agroecological zones of Mozambique. Soil Sci. 171, 902–914. Marshall, P.L., Region, V. F., 2000. Using line intersect sampling for coarse woody debris. [Research Section], Vancouver Forest Region. Mittelbach, G.G., Steiner, C.F., Scheiner, S.M., Gross, K.L., Reynolds, H.L., Waide, R.B., Willig, M.R., Dodson, S.I., Gough, L., 2001. What is the observed relationship between species richness and productivity? Ecology 82, 2381–2396. Mutowo, G., Murwira, A., 2012. Relationship between remotely sensed variables and tree species diversity in savanna woodlands of Southern Africa. Int. J. Remote Sen. 33, 6378–6402. Nagendra, H., Rocchini, D., 2008. High resolution satellite imagery for tropical biodiversity studies: the devil is in the detail. Biodivers. Conserv. 17, 3431–3442.
Nagendra, H., Rocchini, D., Ghate, R., Sharma, B., Pareeth, S., 2010. Assessing plant diversity in a dry tropical forest: comparing the utility of Landsat and IKONOS satellite images. Remote Sens. 2, 478–496. Nagendra, H., Lucasb, R., Honradoc, J.P., Jongmand, R.H.G., Tarantinoe, C., Adamoe, M., Mairotaf, P., 2013. Remote sensing for conservation monitoring: assessing protected areas, habitat extent, habitat condition, species diversity, and threats. Ecol. Indic. 33, 45–59. Nagendra, H., 2002. Opposite trends in response for the Shannon and Simpson indices of landscape diversity. Appl. Geogr. 22, 175–186. Nyamapfene, K.W., 1991. The Soils of Zimbabwe. Nehanda Publishers. Oindo, B.O., Skidmore, A.K., 2002. Interannual variability of NDVI and species richness in Kenya. Int. J. Remote Sens. 23, 285–298. Oldeland, J., Wesuls, D., Rocchini, D., Schmidt, M., Jürgens, N., 2010. Does using species abundance data improve estimates of species diversity from remotely sensed spectral heterogeneity? Ecol. Indic. 10, 390–396. Owen-Smith, N., 1993. Evaluating optimal diet models for an African browsing ruminant, the kudu: how constraining are the assumed constraints? Evol. Ecol. 7, 499–524. Parma, R., Shataee, S., 2013. Estimation of species diversity of trees and shrubs using ETM+ sensor data (case study of forests in Qalajeh Kermanshah province). Int. J. Adv. Biol. Biomed. Res. (IJABBR) 1, 71–78. Parr, C.L., Lehmann, C.E.R., Bond, W.J., Hoffmann, W.A., Andersen, A.N., 2014. Tropical grassy biomes: misunderstood, neglected, and under threat. Trends Ecol. Evol. 29, 205–213. Parviainen, M., Luoto, M., Heikkinen, R.K., 2010. NDVI-based productivity and heterogeneity as indicators of plant-species richness in boreal landscapes. Boreal Environ. Res. 15, 301–318. Pettorelli, N., Safi, K., Turner, W., 2014. Satellite remote sensing, biodiversity research and conservation of the future. Philos. Trans. R. Soc. Lond. B: Biol. Sci. 369, 20130190. Rocchini, D., Balkenhol, N., Carter, G.A., Foody, G.M., Gillespie, T.W., He, K.S., Kark, S., Levin, N., Lucas, K., Luoto, M., 2010a. Remotely sensed spectral heterogeneity as a proxy of species diversity: recent advances and open challenges. Ecol. Inf. 5, 318–329. Rocchini, D., Balkenhol, N., Carter, G.A., Foody, G.M., Gillespie, T.W., He, K.S., Kark, S., Levin, N., Lucas, K., Luoto, M., Nagendra, H., Oldeland, J., Ricotta, C., Southworth, J., Neteler, M., 2010b. Remotely sensed spectral heterogeneity as a proxy of species diversity: recent advances and open challenges. Ecol. Inf., 1–12. Rocchini, D., Hernández-Stefanoni, J.L., He, K.S., 2015. Advancing species diversity estimate by remotely sensed proxies: a conceptual review. Ecol. Inf. 25, 22–28. Roy, D.P., Wulder, M., Loveland, T., Woodcock, C., Allen, R., Anderson, M., Helder, D., Irons, J., Johnson, D., Kennedy, R., 2014. Landsat-8: Science and product vision for terrestrial global change research. Remote Sens. Environ. 145, 154–172. Saatchi, S., Buermann, W., t. Steege, H., Mori, S., Smith, T.B., 2008. Modeling distribution of Amazonian tree species and diversity using remote sensing measurements. Remote Sens. Environ. 112, 2000–2017. Sankaran, M., Anderson, T.M., 2009. Management and restoration in African savannas: interactions and feedbacks. New Models Ecosyst. Dyn., 136–155. Sankaran, M., Hanan, N.P., Scholes, R.J., Ratnam, J., Augustine, D.J., Cade, B.S., Gignoux, J., Higgins, S.I., Le Roux, X., Ludwig, F., 2005. Determinants of woody cover in African savannas. Nature 438, 846–849. Santin-Janin, H., Garel, M., Chapuis, J.L., Pontier, D., 2009. Assessing the performance of NDVI as a proxy for plant biomass using non-linear models: a case study on the Kerguelen archipelago. Polar Biol. 32, 861–871. Scheiner, S.M., Jones, S., 2002. Diversity, productivity and scale in Wisconsin vegetation. Evol. Ecol. Res. 4, 1097–1117. Schmidt, M., Lucas, R., Bunting, P., Verbesselt, J., Armston, J., 2015. Multi-resolution time series imagery for forest disturbance and regrowth monitoring in Queensland, Australia. Remote Sens. Environ. 158, 156–168. Scholes, R., Archer, S., 1997. Tree-grass interactions in savannas. Annu. Rev. Ecol. Syst., 517–544. Scholes, R.J., Walker, B.H., 2004. An African Savanna: Synthesis of the Nylsvley Study. Cambridge University Press. Seto, K., Fleishman, E., Fay, J., Betrus, C., 2004. Linking spatial patterns of bird and butterfly species richness with Landsat TM derived NDVI. Int. J. Remote Sens. 25, 4309–4324. Shimabukuro, Y.E., Batista, G.T., Mello, E.M.K., Moreira, J.C., Duarte, V., 1998. Using shade fraction image segmentation to evaluate deforestation in Landsat Thematic Mapper images of the Amazon Region. Int. J. Remote Sens. 19, 535–541. Simova, I., Li, Y.M., Storch, D., 2013. Relationship between species richness and productivity in plants: the role of sampling effect, heterogeneity and species pool. J. Ecol. 101, 161–170. Solbrig, O.T., Medina, E., Silva, J., 1996. Biodiversity and tropical savanna properties: a global view. Scope Sci. Comm. Prob. Environ. Int. Counc. Sci. Unions 55, 185–211. Somers, B., Asner, G.P., Martin, R.E., Anderson, C.B., Knapp, D.E., JosephWright, S., Kerchove, R.V.D., 2015. Mesoscale assessment of changes in tropical tree species richness across a bioclimatic gradient in Panama using airborne imaging spectroscopy. Remote Sens. Environ. 167, 111–120. St Louis, V., Pidgeon, A.M., Clayton, M.K., Locke, B.A., Bash, D., Radeloff, V.C., 2009. Satellite image texture and a vegetation index predict avian biodiversity in the Chihuahuan Desert of New Mexico. Ecography 32, 468–480. Sutherland, W.J., Krebs, C.J., 1997. Ecological census techniques. Trends Ecol. Evol. 12, 81–81.
R.B. Mapfumo et al. / International Journal of Applied Earth Observation and Geoinformation 52 (2016) 306–317 Thiollay, J.-m., 1997. Disturbance, selective logging and bird diversity: a Neotropical forest study. Biodivers. Conserv. 6, 1155–1173. Trapnell, C.G., Martin, J. D., Allan, W., 2001. Vegetation-soil Map of Zambia. Royal Botanic Gardens. Tucker, C.J., Sellers, P.J., 1986. Satellite remote sensing of primary production. Int. J. Remote Sens. 7, 1395–1416. Turner, W., Rondinini, C., Pettorelli, N., Mora, B., Leidner, A.K., Szantoi, Z., Buchanan, G., Dech, S., Dwyer, J., Herold, M., Koh, L.P., Leimgruber, P., Taubenboeck, H., l, M.W., Wikelskim, M., Woodcock, C., 2015. Free and open-access satellite data are key to biodiversity conservation. Biological Conservatio. Biol. Conserv., 173–176. Van Wilgen, B., 2009. The evolution of fire management practices in savanna protected areas in South Africa. S. Afr. J. Sci. 105, 343–349. Waide, R., Willig, M., Steiner, C., Mittelbach, G., Gough, L., Dodson, S., Juday, G., Parmenter, R., 1999. The relationship between productivity and species richness. Annu. Rev. Ecol. Syst., 257–300.
317
Wiens, J., Sutter, R., Anderson, M., Blanchard, J., Barnett, A., Aguilar-Amuchastegui, N., Avery, C., Laine, S., 2009. Selecting and conserving lands for biodiversity: the role of remote sensing. Remote Sens. Environ. 113, 1370–1381. Williams, V.L., Witkowski, E.T., Balkwill, K., 2005. Application of diversity indices to appraise plant availability in the traditional medicinal markets of Johannesburg, South Africa. Biodivers. Conserv. 14, 2971–3001. Zellweger, F., Braunisch, V., Baltensweiler, A., Bollmann, K., 2013. Remotely sensed forest structural complexity predicts multi species occurrence at the landscape scale. For. Ecol. Manag. 307, 303–312. Zhang, C., Lu, D., Chen, X., Zhang, Y., Maisupova, B., Tao, Y., 2016. The spatiotemporal patterns of vegetation coverage and biomass of the temperate deserts in Central Asia and their relationships with climate controls. Remote Sens. Environ. 175, 271–281.