- Email: [email protected]
Predicting forest structural attributes using ancillary data and ASTER satellite data
International Journal of Applied Earth Observation and Geoinformation 12S (2010) S23–S26
Contents lists available at ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Predicting forest structural attributes using ancillary data and ASTER satellite data M.T. Gebreslasie a,c,*, F.B. Ahmed a,c, Jan A.N. van Aardt b,c a
University of KwaZulu Natal, School of Environmental Sciences, King George V Avenue, Glenwood, Durban 4041, South Africa Council for Scientific and Industrial Research, Natural Resources and the Environment, Ecosystems, Earth Observation, P.O. Box 395, Pretoria 0001, South Africa c Rochester Institute of Technology, Center for Imaging Science – Laboratory for Imaging Algorithms and Systems, 54 Lomb Memorial Drive, Rochester, NY 14623, USA b
A R T I C L E I N F O
A B S T R A C T
Article history: Received 12 November 2008 Accepted 13 November 2009
This study assessed the suitability of both visible and shortwave infrared ASTER data and vegetation indices for estimating forest structural attributes of Eucalyptus species in the southern KwaZulu Natal, South Africa. The empirical relationships between forest structural attributes and ASTER data were derived using stepwise multiple regression analysis; Modified Soil Adjusted Vegetation Index (MSVI) and band 3 were selected for analysis as it showed best relationships with forest structural attributes. The ancillary data such as age and site index were also included in the analysis. Although the results of this study have indicated statistically significant relationships between the forest structural attributes and the ASTER data in the plantation forests stands with adjusted R2-values for volume, basal area (BA), stem per hectare (SPHA), and tree height of 0.51, 0.67, 0.65, and 0.52, respectively, but these results are not suitable for operational purpose in a forest company. However, the structural forest attribute predictions were markedly improved after incorporating age and site index as predictor variable. R2values for the stands increased by 42%, 20.2%, 16.8%, and 42.2% for volume, basal area, SPHA, and tree height, respectively. These results imply that ASTER satellite data alone are not applicable to forest structural attribute estimation; however, ASTER data can provide useful information if it is used in conjunction with age and site index data for forest structural attribute estimation in plantation forests. ß 2009 Elsevier B.V. All rights reserved.
Keywords: ASTER dataset Spectral vegetation indices Forest structural attributes
1. Introduction Forest structural attributes, such as volume, BA, SPHA, and tree height are important data needed for effective forest management. Currently, in South Africa, manual field surveys are used to gather information regarding the forest structural attributes, which will be referred as ‘‘plantations/plantation attributes’’ for the purpose of this study. Even though this method provides highly accurate measurements of plantation attributes, it is costly and time consuming (Trotter et al., 1997). Many researchers, e.g. Wulder (1998), Hyyppa¨ et al. (2000), Lu et al. (2004), and McRoberts and Tomppo (2007), have recommended that remotely sensed data be investigated as an alternative means of acquiring information about plantations resources. A large number of remote sensing studies has shown that prediction of forest structural attributes using optical remote sensing has been based on empirical relationships established between the field measured data and remote sensing data, such as wavelength bands and vegetation
* Corresponding author at: University of KwaZulu Natal, School of Environmental Sciences, King George V Avenue, Glenwood, Durban 4041, South Africa. Tel.: +27 72 744 9253; fax: +27 31 261 1216. E-mail address: [email protected] (M.T. Gebreslasie). 0303-2434/$ – see front matter ß 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jag.2009.11.006
indices. A variety of vegetation indices have been developed using broad-band remotely sensed data based on the spectral features of green vegetation (Birth and Mcvey, 1968; Rouse et al., 1973; Tucker, 1979; Spanner et al., 1990; Nemani et al., 1993; Qi et al., 1994; Brown et al., 2000). In addition, remote sensing data transformation methods like Principal Components Analysis appear to provide acceptable estimation results (Lu et al., 2004). Remote sensing studies, conducted in forested areas, have indicated that relationships between forest structural attributes and remote sensing data differ depending on the geographic settings of the study sites and level of management. In a managed forest plantation in northern Wisconsin, USA, Zheng et al. (2004) have found that diameter at breast height (DBH) estimates for hardwood forest were strongly related to stand age and near-infrared reflectance adjusted R2-values of 0.77, while for softwood forests the estimates were strongly related to the Normalised Difference Vegetation Index (NDVI) adjusted R2-values of 0.79. Another study in mountain birch forests northernmost Finland conducted by Heiskanen (2006) indicated that NDVI, MSAVI and simple ratio (SR) of ASTER data showed a significant relationship with biomass and LAI. In addition, a study in west-central Alberta, Canada by Hall et al. (2006) suggested that red and near-infrared bands returned the strongest relationships with above ground biomass and volume adjusted R2-values of 0.67 and 66, respectively. These studies in the
S24
M.T. Gebreslasie et al. / International Journal of Applied Earth Observation and Geoinformation 12S (2010) S23–S26
way indicated that satellite data in general are potentially valuable for characterizing forest structural variables in a variety of environments, differently. The objectives of this study was (i) to analyse the potential of ASTER data sets for estimating plantation attributes such as volume, BA, SPHA, and tree height and (ii) to assess the degree to which ancillary data i.e. age and site index impact estimations in plantation stands in the southern part of KwaZulu Natal, South Africa.
DBH and tree height. These variables were in turn used to derive plot-level basal area and volume. Site index, which was incorporated as an ancillary independent variable, is a measurement, commonly used by foresters, of the quality of an even-aged site for growing trees and is easily extracted from forestry records for even-aged, planted stands.
2. Materials and methods
An ASTER scene acquired on November 2006 and processed to level 1A product (Abrams, 2000) was used in this study. ASTER has three subsystems operating in different spectral regions, namely the visible and near-infrared (VNIR), shortwave infrared (SWIR), and thermal infrared (TIR) regions. The spatial resolution is 15 m, 30 m, and 90 m for VNIR, SWIR and TIR, respectively. The thermal data were not used in this study, given the focus on forest attribute assessment using relatively standard VNIR to SWIR wavelengths. These bands were normalised to absolute radiance. Spectral vegetation indices were calculated (refer to Table 1) in order to evaluate their potential for the prediction of plantation attributes.
2.1. Study area The study area is located in the Midlands southern KwaZulu Natal, South Africa. The sites chosen for this study are managed by MONDI-SA. The study area is bounded by 298460 31.4100 S and 30820 57.0700 E on the northwest and 298560 32.4300 S and 308190 18.0600 E on the southeast (refer to Fig. 1). The terrain in the study area ranges from gently undulating to highly dissected, strongly rolling, and hilly topography. Elevations range between 800 m and 1400 m above-mean-sea-level. Plantation forestry is a major land use in the study area due to the suitable climate and soils. Rainfall ranges from 820 mm to 1300 mm, but averages 1000 mm per annum mostly falling between October and April. Temperatures vary between 24 8C and 26 8C in summer, but drop to between 5 8C and 14 8C in winter. 2.2. Field measurements A Geographical Information System (GIS), compiled and provided by MONDI-SA, was consulted in order to select stands of interest. Attributes used in the selection procedure were the spatial location and extent of each forest compartment or stand, species type, age, site index, planting and felling dates, and coppice status. Site index is a quantitative measure of site in even-aged stands; it is calculated from tree height. Only Eucalyptus species such as Eucalyptus grandis and Eucalyptus nitens were considered from this dataset. These species were chosen given the need identified by the forestry sector to increase their productivity, which was in turn driven by the growing demands for its various end products. The field data collection was conducted in November 2006. Forest structural attributes measured during the field surveys were
2.3. Remote sensing data and processing
2.4. Plot-level spectral data extraction A plot-level average reflectance value was derived from the corresponding image for each plot subsequent to spectral bands and calculated of vegetation indices. ASTER band 2 was assigned to the ‘‘Red’’, band 3 to the ‘‘NIR’’, and band 4 to the ‘‘SWIR’’ variable in the index equations. The slope and intercept of the soil line required for the derivation of the PVI and TSAVI were determined from a scatter plot of the red and NIR ASTER reflectance values. 2.5. Statistical analysis For these cases the best predictors were selected from spectral bands and vegetation indices using stepwise regression analysis, while model selection was based on fit criteria. The results with the highest adjusted R2-values and p values equal to or less than 0.05 were selected from cases with both spectral bands and vegetation indices as independent variables. The final selection of the most appropriate models was based on the computation of adjusted R2, low RMSE, and the number of independent variables. This latter aspect relates to model robustness while avoiding over-fitting
Fig. 1. Map showing the location of the study sites.
M.T. Gebreslasie et al. / International Journal of Applied Earth Observation and Geoinformation 12S (2010) S23–S26
S25
Table 1 Spectral vegetation indices examined in this study. Spectral index NDVI MSAVI PVI TSAVI RSR
Equations
Reference
(NIR Red)/(NIR + Red) qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðNIR þ 0:5Þ2 2ðNIR RedÞ
Rouse et al. (1973)
NIR þ 0:5
Qi et al. (1994)
NIR a Red þ b pffiffiffiffiffiffiffiffiffiffiffiffiffiffi a2 þ 1 NIR a Red b a Red þ a2 Þ þ aðNIR bÞ þ 0:08ð1 NIR SWIR SWIRmin 1 Red SWIRmax SWIRmin
Baret and Guyot (1991)
Richardson and Wiegand (1977)
Brown et al. (2000)
SWIRmin and SWIRmax in the RSR are the minimum and maximum reflectances observed in the field plots; a (0.8452) and b (0.1932) in the PVI and TSAVI formulae represents the soil gradient and intercept, respectively.
concerns. Additional performance assessment was based on the evaluation of scatter plots in order to gauge field measured versus predicted residual properties. Its correlation coefficient (r) is reported. 3. Results Multiple regression model based on MSAVI and band 3 (red band) was selected as the best predictor of volume with the adjusted R2 = 0.51; RMSE = 0.14 m3 (refer to Table 2). However, with the addition of ancillary data i.e. age and site index as an independent variable, the adjusted R2-value increased to 0.88, and the RMSE decreased to 0.11 m3 as shown in Table 3. This model exhibited a negative relationship between MSAVI values and volume, whereas a direct relationship was observed between ASTER band 3 reflectance values and volume as shown in Table 3. Comparatively, a multiple regression model using spectral vegetation indices i.e. MSAVI and ASTER spectral band 3 explained more variability in basal area (i.e. adjusted R2 = 0.67; RMSE = 9.2 m2), as shown in Table 2, than using other predictors. The MSAVI and band 3, with the addition of age and site index as independent variables, have exhibited an improved basal area prediction (adjusted R2 = 0.84; RMSE = 7.39 m2) than the other predictor variables. A direct relationship was observed between the predictor values and basal area (refer to Tables 2 and 3). A multiple regression model consisting of ASTER spectral vegetation indices MSAVI and ASTER band 3 were shown as the best predictors of SPHA when compared to other predictor combinations. The incorporation of age and site index to the model improved the relationships from adjusted R2 = 0.65; RMSE 168 to adjusted R2 = 0.81 and RMSE = 124 (refer to Tables 2 and 3). The reflectance in band 3 increased with SPHA, while reflectance in MSVI decreased as shown in Table 3. Lastly, a multiple regression model, with MSAVI and band 3 as independent variables, proved to
be the model best suited to estimation of tree height. An improved relationship (adjusted R2 = 0.90, and RMSE = 1.98 m) was observed by adding age and site index as independent variable (refer to Tables 2 and 3). However, it can be justly argued that plantation attributes and site index are two related variables and with inclusion of age as independent variable, they can logically result in an improved prediction model. The ease with which current site index can be derived from plantation company database and the existence of age in plantation records were considered as valid reasons for their inclusion in the modeling effort. 4. Discussion The near-infrared spectral range (band 3) consistently returned a strong (positive) correlation with volume, basal area, SPHA, and tree height in the plantation forest. This was attributed to a homogenous full canopy cover, implying that as canopy cover increases the amount of bright smooth canopy surface increases sharply, resulting in an increased near-infrared reflectance. Nearinfrared reflectance has been found to be a significant predictor of forest structural variables in other studies that use correlation and regression analyses (Hall et al., 2006; Zheng et al., 2004). Vegetation indices MSAVI and NDVI showed the strongest correlation to the plantation attributes. MSAVI has been shown to be particularly useful in more closed canopy plantation forest stands (Heiskanen, 2006). MASVI and NDVI predictor also appeared to have a strong linear relationship with forest structural variables as suggested in studies by Ingram et al. (2005), Freitas et al. (2005), and Heiskanen (2006). The models using ASTER satellite data only resulted in satisfactory estimations of plantation structural attributes; however if used in conjunction with age and site index, data could be used for forest structural attribute estimation in the plantation forest context. The incorporation of site index as an ancillary data
Table 2 Models without ancillary data for the estimation of plantation attribute.
3
Volume (m ) Basal area (m2) SPHA Tree height (m) a
Modela
R2
RMSE
r
0.479 7.11(MSAVI) + 8.65(Band 3) 659 + 5066(MSAVI) + 5716(Band 3) 13.64 27.4(MSAVI) + 151(Band 3) 16.47 + 139.8(MSAVI) 278 (Band 3)
0.51 0.67 0.65 0.52
0.14 9.2 168 2.85
0.553 0.729 0.705 0.566
SI, site index; r, coefficient of correlation between field observed data and predicted.
Table 3 Models with ancillary data for the estimation of plantation attribute.
Volume (m3) Basal area (m2) SPHA Tree height (m) a
Modela
R2
RMSE
r
0.196 + 0.03497(Age) + 0.01568(SI) 3.9(MSAVI) + 5.38(Band 3) 229 + 26.58(Age) 9.02(SI) + 916(MSAVI) + 2981(Band 3) 13.64 + 1.049(Age) + 0.36(SI) 27.4(MSAVI) + 151.6(Band 3) 10.87 + 1.924(Age) + 0.555(SI) 183.9(MSAVI) 34.6(Band 3)
0.88 0.84 0.81 0.90
0.11 7.39 124 1.98
0.958 0.911 0.881 0.976
SI, site index; r, coefficient of correlation between field observed data and predicted.
S26
M.T. Gebreslasie et al. / International Journal of Applied Earth Observation and Geoinformation 12S (2010) S23–S26
source seemed to improve forest structural attribute prediction based on ASTER data for all models. This is partly due to the determination of site quality classes being based on tree height, which is an important structural-related variable. It could be argued that site index as an independent variable requires too much additional calculation that are independent of spectral data, thereby negating an approach based only on single-source reflectance data. However, the inclusion of especially site index in this research (i) has shown the extent to which the stand-alone spectral data are bolstered by ancillary variables for forest structural assessment and (ii) is based on the presumption that site index information is readily available for intensively managed plantations. 5. Conclusions The relationships between reflectance data recorded by the ASTER sensor and forest structural attributes of plantation forests were analysed through stepwise regression techniques in this study. A combination of ASTER data (i.e. MSAVI and band 3 with ancillary data stand age and site index) were the best model-based predictors of volume, BA, SPHA, and tree height. The combination of these two ASTER variables in conjunction with age and site index yielded adjusted R2-values of 0.88, 0.84, 0.81, and 0.90 for the respective dependent variables. Results from this analysis demonstrated that ASTER satellitebased reflectance values are relatively weak for prediction of plantation forest structural attributes, which requires high estimation accuracy. However, with the incorporation of age and site index the models have exhibited improved model fit metrics. When one considers the accuracy of the estimates and the spatial range of the input satellite data, the developed models are applicable to estimate forest structural attributes for plantation forests and therefore can be potentially employed to map these variables for similar areas in the future. However, further research is required to document the performance of the retrieval under different environmental conditions and topographical changes, as well as for other species. The associated model accuracy and precision could be considered trade-offs relative to the spatial extent and cost of such an approach. Factors affecting the reflectance of plantation forests also should be studied further through sensitivity analysis using a suitable canopy reflectance model. Acknowledgements This study was conducted as part of a remote sensing cooperative program between MONDI SA, the Council for Scientific and Industrial Research (CSIR), and the University of Kwa-Zulu Natal to investigate the potential of satellite remote sensing imagery for prediction of plantation forest structural attributes.
The authors would like to thank Wesley Roberts, Solomon Gebremariam, Thamsanqa Mzinyane, and Wesley Naidoo for their assistance during field work. Financial support for this study was provided by the CSIR and MONDI SA. References Abrams, M., 2000. The advanced spaceborne thermal emission and reflection radiometer (ASTER): data products for the high spatial resolution imager on NASA’s Terra platform. International Journal of Remote Sensing 21, 847–859. Baret, F., Guyot, G., 1991. Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sensing of Environment 35, 161–173. Birth, G.S., Mcvey, G.R., 1968. Measuring the colour of growing turf with a reflectance spectrophotometer. Agronomy Journal 60, 640–643. Brown, L., Chen, J.M., Leblanc, S.G., Cihlar, J., 2000. A shortwave infrared modification to the simple ratio for LAI retrieval in boreal forests: an image and model analysis. Remote Sensing of Environment 71, 16–25. Freitas, S.R., Mello, M.C.S., Cruz, C.B.M., 2005. Relationships between forest structure and vegetation indices in Atlantic rainforest. Forest Ecology and Management 218, 353–362. Hall, R.J., Skakun, R.S., Arsenault, E.J., Case, B.S., 2006. Modelling forest stand structure attributes using Landsat ETM+ data: application to mapping of aboveground biomass and stand volume. Forest Ecology and Management 225, 378– 390. Heiskanen, J., 2006. Estimating aboveground tree biomass and leaf area index in a mountain birch forest using ASTER satellite data. International Journal of Remote Sensing 27 (6), 1135–1158. Hyyppa¨, H., Inkinen, M., Engdahl, M., Linko, S., Zhu, Y.H., 2000. Accuracy comparison of various remote sensing data sources in the retrieval of forest stand attributes. Forest Ecology and Management 128 (1–2), 109–120. Ingram, J.C., Dawson, T.P., Whittaker, R.J., 2005. Mapping tropical forest structure in south-eastern Madagascar using remote sensing and artificial neural networks. Remote Sensing of Environment 94, 491–507. Lu, D., Mausel, P., Brondizio, E., Moran, E., 2004. Relationships between forest stand parameters and Landsat TM spectral response in Brazilian Amazon Basin. Forest Ecology and Management 198, 149–167. McRoberts, R.E., Tomppo, E.O., 2007. Remote sensing support for national forest inventories. Remote Sensing of Environment 110, 412–419. Nemani, R., Pierce, L., Running, S.W., Band, L., 1993. Forest ecosystem processes at the watershed scale: sensitivity to remotely sensed leaf area index estimates. International Journal of Remote sensing 14 (13), 2519–2534. Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H., Sorooshian, S., 1994. A modified soil adjusted vegetation index. Remote Sensing of Environment 48, 119–126. Richardson, A.J., Wiegand, C.L., 1977. Distinguishing vegetation from soil background information. Photogrammetry of Engendering and Remote Sensing 43, 1541–1552. Rouse, J., Haas, R., Schell, J., Deering, D., Harlan, J., 1973. Monitoring the vernal advancements and retrogradation (greenwave effect) of nature vegetation. NASA/GSFC Final Report, NASA, Greenbelt, MD. Spanner, M.A., Pierce, L.L., Peterson, D.L., Running, S.W., 1990. Remote sensing of temperate coniferous forest leaf area index: the influence of canopy closure, understory vegetation, and background reflectance. International Journal of Remote Sensing 11, 95–111. Trotter, C.M., Dymond, J.R., Goulding, C.J., 1997. Estimation of timber volume in a coniferous plantation forest using Landsat TM. International Journal of Remote Sensing 18, 2209–2223. Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment 8, 127–150. Wulder, M.A., 1998. Optical remote sensing techniques for the assessment of forest inventory and biophysical parameters. Progress in Physical Geography 22 (4), 449–476. Zheng, D., Rademacher, J., Chen, J., Crow, T., Bressee, M., Moine, J.L., Ryu, S., 2004. Estimating aboveground biomass using Landsat ETM+ data across a managed landscape in northern Wisconsin, USA. Remote Sensing of Environment.
Contents lists available at ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Predicting forest structural attributes using ancillary data and ASTER satellite data M.T. Gebreslasie a,c,*, F.B. Ahmed a,c, Jan A.N. van Aardt b,c a
University of KwaZulu Natal, School of Environmental Sciences, King George V Avenue, Glenwood, Durban 4041, South Africa Council for Scientific and Industrial Research, Natural Resources and the Environment, Ecosystems, Earth Observation, P.O. Box 395, Pretoria 0001, South Africa c Rochester Institute of Technology, Center for Imaging Science – Laboratory for Imaging Algorithms and Systems, 54 Lomb Memorial Drive, Rochester, NY 14623, USA b
A R T I C L E I N F O
A B S T R A C T
Article history: Received 12 November 2008 Accepted 13 November 2009
This study assessed the suitability of both visible and shortwave infrared ASTER data and vegetation indices for estimating forest structural attributes of Eucalyptus species in the southern KwaZulu Natal, South Africa. The empirical relationships between forest structural attributes and ASTER data were derived using stepwise multiple regression analysis; Modified Soil Adjusted Vegetation Index (MSVI) and band 3 were selected for analysis as it showed best relationships with forest structural attributes. The ancillary data such as age and site index were also included in the analysis. Although the results of this study have indicated statistically significant relationships between the forest structural attributes and the ASTER data in the plantation forests stands with adjusted R2-values for volume, basal area (BA), stem per hectare (SPHA), and tree height of 0.51, 0.67, 0.65, and 0.52, respectively, but these results are not suitable for operational purpose in a forest company. However, the structural forest attribute predictions were markedly improved after incorporating age and site index as predictor variable. R2values for the stands increased by 42%, 20.2%, 16.8%, and 42.2% for volume, basal area, SPHA, and tree height, respectively. These results imply that ASTER satellite data alone are not applicable to forest structural attribute estimation; however, ASTER data can provide useful information if it is used in conjunction with age and site index data for forest structural attribute estimation in plantation forests. ß 2009 Elsevier B.V. All rights reserved.
Keywords: ASTER dataset Spectral vegetation indices Forest structural attributes
1. Introduction Forest structural attributes, such as volume, BA, SPHA, and tree height are important data needed for effective forest management. Currently, in South Africa, manual field surveys are used to gather information regarding the forest structural attributes, which will be referred as ‘‘plantations/plantation attributes’’ for the purpose of this study. Even though this method provides highly accurate measurements of plantation attributes, it is costly and time consuming (Trotter et al., 1997). Many researchers, e.g. Wulder (1998), Hyyppa¨ et al. (2000), Lu et al. (2004), and McRoberts and Tomppo (2007), have recommended that remotely sensed data be investigated as an alternative means of acquiring information about plantations resources. A large number of remote sensing studies has shown that prediction of forest structural attributes using optical remote sensing has been based on empirical relationships established between the field measured data and remote sensing data, such as wavelength bands and vegetation
* Corresponding author at: University of KwaZulu Natal, School of Environmental Sciences, King George V Avenue, Glenwood, Durban 4041, South Africa. Tel.: +27 72 744 9253; fax: +27 31 261 1216. E-mail address: [email protected] (M.T. Gebreslasie). 0303-2434/$ – see front matter ß 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jag.2009.11.006
indices. A variety of vegetation indices have been developed using broad-band remotely sensed data based on the spectral features of green vegetation (Birth and Mcvey, 1968; Rouse et al., 1973; Tucker, 1979; Spanner et al., 1990; Nemani et al., 1993; Qi et al., 1994; Brown et al., 2000). In addition, remote sensing data transformation methods like Principal Components Analysis appear to provide acceptable estimation results (Lu et al., 2004). Remote sensing studies, conducted in forested areas, have indicated that relationships between forest structural attributes and remote sensing data differ depending on the geographic settings of the study sites and level of management. In a managed forest plantation in northern Wisconsin, USA, Zheng et al. (2004) have found that diameter at breast height (DBH) estimates for hardwood forest were strongly related to stand age and near-infrared reflectance adjusted R2-values of 0.77, while for softwood forests the estimates were strongly related to the Normalised Difference Vegetation Index (NDVI) adjusted R2-values of 0.79. Another study in mountain birch forests northernmost Finland conducted by Heiskanen (2006) indicated that NDVI, MSAVI and simple ratio (SR) of ASTER data showed a significant relationship with biomass and LAI. In addition, a study in west-central Alberta, Canada by Hall et al. (2006) suggested that red and near-infrared bands returned the strongest relationships with above ground biomass and volume adjusted R2-values of 0.67 and 66, respectively. These studies in the
S24
M.T. Gebreslasie et al. / International Journal of Applied Earth Observation and Geoinformation 12S (2010) S23–S26
way indicated that satellite data in general are potentially valuable for characterizing forest structural variables in a variety of environments, differently. The objectives of this study was (i) to analyse the potential of ASTER data sets for estimating plantation attributes such as volume, BA, SPHA, and tree height and (ii) to assess the degree to which ancillary data i.e. age and site index impact estimations in plantation stands in the southern part of KwaZulu Natal, South Africa.
DBH and tree height. These variables were in turn used to derive plot-level basal area and volume. Site index, which was incorporated as an ancillary independent variable, is a measurement, commonly used by foresters, of the quality of an even-aged site for growing trees and is easily extracted from forestry records for even-aged, planted stands.
2. Materials and methods
An ASTER scene acquired on November 2006 and processed to level 1A product (Abrams, 2000) was used in this study. ASTER has three subsystems operating in different spectral regions, namely the visible and near-infrared (VNIR), shortwave infrared (SWIR), and thermal infrared (TIR) regions. The spatial resolution is 15 m, 30 m, and 90 m for VNIR, SWIR and TIR, respectively. The thermal data were not used in this study, given the focus on forest attribute assessment using relatively standard VNIR to SWIR wavelengths. These bands were normalised to absolute radiance. Spectral vegetation indices were calculated (refer to Table 1) in order to evaluate their potential for the prediction of plantation attributes.
2.1. Study area The study area is located in the Midlands southern KwaZulu Natal, South Africa. The sites chosen for this study are managed by MONDI-SA. The study area is bounded by 298460 31.4100 S and 30820 57.0700 E on the northwest and 298560 32.4300 S and 308190 18.0600 E on the southeast (refer to Fig. 1). The terrain in the study area ranges from gently undulating to highly dissected, strongly rolling, and hilly topography. Elevations range between 800 m and 1400 m above-mean-sea-level. Plantation forestry is a major land use in the study area due to the suitable climate and soils. Rainfall ranges from 820 mm to 1300 mm, but averages 1000 mm per annum mostly falling between October and April. Temperatures vary between 24 8C and 26 8C in summer, but drop to between 5 8C and 14 8C in winter. 2.2. Field measurements A Geographical Information System (GIS), compiled and provided by MONDI-SA, was consulted in order to select stands of interest. Attributes used in the selection procedure were the spatial location and extent of each forest compartment or stand, species type, age, site index, planting and felling dates, and coppice status. Site index is a quantitative measure of site in even-aged stands; it is calculated from tree height. Only Eucalyptus species such as Eucalyptus grandis and Eucalyptus nitens were considered from this dataset. These species were chosen given the need identified by the forestry sector to increase their productivity, which was in turn driven by the growing demands for its various end products. The field data collection was conducted in November 2006. Forest structural attributes measured during the field surveys were
2.3. Remote sensing data and processing
2.4. Plot-level spectral data extraction A plot-level average reflectance value was derived from the corresponding image for each plot subsequent to spectral bands and calculated of vegetation indices. ASTER band 2 was assigned to the ‘‘Red’’, band 3 to the ‘‘NIR’’, and band 4 to the ‘‘SWIR’’ variable in the index equations. The slope and intercept of the soil line required for the derivation of the PVI and TSAVI were determined from a scatter plot of the red and NIR ASTER reflectance values. 2.5. Statistical analysis For these cases the best predictors were selected from spectral bands and vegetation indices using stepwise regression analysis, while model selection was based on fit criteria. The results with the highest adjusted R2-values and p values equal to or less than 0.05 were selected from cases with both spectral bands and vegetation indices as independent variables. The final selection of the most appropriate models was based on the computation of adjusted R2, low RMSE, and the number of independent variables. This latter aspect relates to model robustness while avoiding over-fitting
Fig. 1. Map showing the location of the study sites.
M.T. Gebreslasie et al. / International Journal of Applied Earth Observation and Geoinformation 12S (2010) S23–S26
S25
Table 1 Spectral vegetation indices examined in this study. Spectral index NDVI MSAVI PVI TSAVI RSR
Equations
Reference
(NIR Red)/(NIR + Red) qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðNIR þ 0:5Þ2 2ðNIR RedÞ
Rouse et al. (1973)
NIR þ 0:5
Qi et al. (1994)
NIR a Red þ b pffiffiffiffiffiffiffiffiffiffiffiffiffiffi a2 þ 1 NIR a Red b a Red þ a2 Þ þ aðNIR bÞ þ 0:08ð1 NIR SWIR SWIRmin 1 Red SWIRmax SWIRmin
Baret and Guyot (1991)
Richardson and Wiegand (1977)
Brown et al. (2000)
SWIRmin and SWIRmax in the RSR are the minimum and maximum reflectances observed in the field plots; a (0.8452) and b (0.1932) in the PVI and TSAVI formulae represents the soil gradient and intercept, respectively.
concerns. Additional performance assessment was based on the evaluation of scatter plots in order to gauge field measured versus predicted residual properties. Its correlation coefficient (r) is reported. 3. Results Multiple regression model based on MSAVI and band 3 (red band) was selected as the best predictor of volume with the adjusted R2 = 0.51; RMSE = 0.14 m3 (refer to Table 2). However, with the addition of ancillary data i.e. age and site index as an independent variable, the adjusted R2-value increased to 0.88, and the RMSE decreased to 0.11 m3 as shown in Table 3. This model exhibited a negative relationship between MSAVI values and volume, whereas a direct relationship was observed between ASTER band 3 reflectance values and volume as shown in Table 3. Comparatively, a multiple regression model using spectral vegetation indices i.e. MSAVI and ASTER spectral band 3 explained more variability in basal area (i.e. adjusted R2 = 0.67; RMSE = 9.2 m2), as shown in Table 2, than using other predictors. The MSAVI and band 3, with the addition of age and site index as independent variables, have exhibited an improved basal area prediction (adjusted R2 = 0.84; RMSE = 7.39 m2) than the other predictor variables. A direct relationship was observed between the predictor values and basal area (refer to Tables 2 and 3). A multiple regression model consisting of ASTER spectral vegetation indices MSAVI and ASTER band 3 were shown as the best predictors of SPHA when compared to other predictor combinations. The incorporation of age and site index to the model improved the relationships from adjusted R2 = 0.65; RMSE 168 to adjusted R2 = 0.81 and RMSE = 124 (refer to Tables 2 and 3). The reflectance in band 3 increased with SPHA, while reflectance in MSVI decreased as shown in Table 3. Lastly, a multiple regression model, with MSAVI and band 3 as independent variables, proved to
be the model best suited to estimation of tree height. An improved relationship (adjusted R2 = 0.90, and RMSE = 1.98 m) was observed by adding age and site index as independent variable (refer to Tables 2 and 3). However, it can be justly argued that plantation attributes and site index are two related variables and with inclusion of age as independent variable, they can logically result in an improved prediction model. The ease with which current site index can be derived from plantation company database and the existence of age in plantation records were considered as valid reasons for their inclusion in the modeling effort. 4. Discussion The near-infrared spectral range (band 3) consistently returned a strong (positive) correlation with volume, basal area, SPHA, and tree height in the plantation forest. This was attributed to a homogenous full canopy cover, implying that as canopy cover increases the amount of bright smooth canopy surface increases sharply, resulting in an increased near-infrared reflectance. Nearinfrared reflectance has been found to be a significant predictor of forest structural variables in other studies that use correlation and regression analyses (Hall et al., 2006; Zheng et al., 2004). Vegetation indices MSAVI and NDVI showed the strongest correlation to the plantation attributes. MSAVI has been shown to be particularly useful in more closed canopy plantation forest stands (Heiskanen, 2006). MASVI and NDVI predictor also appeared to have a strong linear relationship with forest structural variables as suggested in studies by Ingram et al. (2005), Freitas et al. (2005), and Heiskanen (2006). The models using ASTER satellite data only resulted in satisfactory estimations of plantation structural attributes; however if used in conjunction with age and site index, data could be used for forest structural attribute estimation in the plantation forest context. The incorporation of site index as an ancillary data
Table 2 Models without ancillary data for the estimation of plantation attribute.
3
Volume (m ) Basal area (m2) SPHA Tree height (m) a
Modela
R2
RMSE
r
0.479 7.11(MSAVI) + 8.65(Band 3) 659 + 5066(MSAVI) + 5716(Band 3) 13.64 27.4(MSAVI) + 151(Band 3) 16.47 + 139.8(MSAVI) 278 (Band 3)
0.51 0.67 0.65 0.52
0.14 9.2 168 2.85
0.553 0.729 0.705 0.566
SI, site index; r, coefficient of correlation between field observed data and predicted.
Table 3 Models with ancillary data for the estimation of plantation attribute.
Volume (m3) Basal area (m2) SPHA Tree height (m) a
Modela
R2
RMSE
r
0.196 + 0.03497(Age) + 0.01568(SI) 3.9(MSAVI) + 5.38(Band 3) 229 + 26.58(Age) 9.02(SI) + 916(MSAVI) + 2981(Band 3) 13.64 + 1.049(Age) + 0.36(SI) 27.4(MSAVI) + 151.6(Band 3) 10.87 + 1.924(Age) + 0.555(SI) 183.9(MSAVI) 34.6(Band 3)
0.88 0.84 0.81 0.90
0.11 7.39 124 1.98
0.958 0.911 0.881 0.976
SI, site index; r, coefficient of correlation between field observed data and predicted.
S26
M.T. Gebreslasie et al. / International Journal of Applied Earth Observation and Geoinformation 12S (2010) S23–S26
source seemed to improve forest structural attribute prediction based on ASTER data for all models. This is partly due to the determination of site quality classes being based on tree height, which is an important structural-related variable. It could be argued that site index as an independent variable requires too much additional calculation that are independent of spectral data, thereby negating an approach based only on single-source reflectance data. However, the inclusion of especially site index in this research (i) has shown the extent to which the stand-alone spectral data are bolstered by ancillary variables for forest structural assessment and (ii) is based on the presumption that site index information is readily available for intensively managed plantations. 5. Conclusions The relationships between reflectance data recorded by the ASTER sensor and forest structural attributes of plantation forests were analysed through stepwise regression techniques in this study. A combination of ASTER data (i.e. MSAVI and band 3 with ancillary data stand age and site index) were the best model-based predictors of volume, BA, SPHA, and tree height. The combination of these two ASTER variables in conjunction with age and site index yielded adjusted R2-values of 0.88, 0.84, 0.81, and 0.90 for the respective dependent variables. Results from this analysis demonstrated that ASTER satellitebased reflectance values are relatively weak for prediction of plantation forest structural attributes, which requires high estimation accuracy. However, with the incorporation of age and site index the models have exhibited improved model fit metrics. When one considers the accuracy of the estimates and the spatial range of the input satellite data, the developed models are applicable to estimate forest structural attributes for plantation forests and therefore can be potentially employed to map these variables for similar areas in the future. However, further research is required to document the performance of the retrieval under different environmental conditions and topographical changes, as well as for other species. The associated model accuracy and precision could be considered trade-offs relative to the spatial extent and cost of such an approach. Factors affecting the reflectance of plantation forests also should be studied further through sensitivity analysis using a suitable canopy reflectance model. Acknowledgements This study was conducted as part of a remote sensing cooperative program between MONDI SA, the Council for Scientific and Industrial Research (CSIR), and the University of Kwa-Zulu Natal to investigate the potential of satellite remote sensing imagery for prediction of plantation forest structural attributes.
The authors would like to thank Wesley Roberts, Solomon Gebremariam, Thamsanqa Mzinyane, and Wesley Naidoo for their assistance during field work. Financial support for this study was provided by the CSIR and MONDI SA. References Abrams, M., 2000. The advanced spaceborne thermal emission and reflection radiometer (ASTER): data products for the high spatial resolution imager on NASA’s Terra platform. International Journal of Remote Sensing 21, 847–859. Baret, F., Guyot, G., 1991. Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sensing of Environment 35, 161–173. Birth, G.S., Mcvey, G.R., 1968. Measuring the colour of growing turf with a reflectance spectrophotometer. Agronomy Journal 60, 640–643. Brown, L., Chen, J.M., Leblanc, S.G., Cihlar, J., 2000. A shortwave infrared modification to the simple ratio for LAI retrieval in boreal forests: an image and model analysis. Remote Sensing of Environment 71, 16–25. Freitas, S.R., Mello, M.C.S., Cruz, C.B.M., 2005. Relationships between forest structure and vegetation indices in Atlantic rainforest. Forest Ecology and Management 218, 353–362. Hall, R.J., Skakun, R.S., Arsenault, E.J., Case, B.S., 2006. Modelling forest stand structure attributes using Landsat ETM+ data: application to mapping of aboveground biomass and stand volume. Forest Ecology and Management 225, 378– 390. Heiskanen, J., 2006. Estimating aboveground tree biomass and leaf area index in a mountain birch forest using ASTER satellite data. International Journal of Remote Sensing 27 (6), 1135–1158. Hyyppa¨, H., Inkinen, M., Engdahl, M., Linko, S., Zhu, Y.H., 2000. Accuracy comparison of various remote sensing data sources in the retrieval of forest stand attributes. Forest Ecology and Management 128 (1–2), 109–120. Ingram, J.C., Dawson, T.P., Whittaker, R.J., 2005. Mapping tropical forest structure in south-eastern Madagascar using remote sensing and artificial neural networks. Remote Sensing of Environment 94, 491–507. Lu, D., Mausel, P., Brondizio, E., Moran, E., 2004. Relationships between forest stand parameters and Landsat TM spectral response in Brazilian Amazon Basin. Forest Ecology and Management 198, 149–167. McRoberts, R.E., Tomppo, E.O., 2007. Remote sensing support for national forest inventories. Remote Sensing of Environment 110, 412–419. Nemani, R., Pierce, L., Running, S.W., Band, L., 1993. Forest ecosystem processes at the watershed scale: sensitivity to remotely sensed leaf area index estimates. International Journal of Remote sensing 14 (13), 2519–2534. Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H., Sorooshian, S., 1994. A modified soil adjusted vegetation index. Remote Sensing of Environment 48, 119–126. Richardson, A.J., Wiegand, C.L., 1977. Distinguishing vegetation from soil background information. Photogrammetry of Engendering and Remote Sensing 43, 1541–1552. Rouse, J., Haas, R., Schell, J., Deering, D., Harlan, J., 1973. Monitoring the vernal advancements and retrogradation (greenwave effect) of nature vegetation. NASA/GSFC Final Report, NASA, Greenbelt, MD. Spanner, M.A., Pierce, L.L., Peterson, D.L., Running, S.W., 1990. Remote sensing of temperate coniferous forest leaf area index: the influence of canopy closure, understory vegetation, and background reflectance. International Journal of Remote Sensing 11, 95–111. Trotter, C.M., Dymond, J.R., Goulding, C.J., 1997. Estimation of timber volume in a coniferous plantation forest using Landsat TM. International Journal of Remote Sensing 18, 2209–2223. Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment 8, 127–150. Wulder, M.A., 1998. Optical remote sensing techniques for the assessment of forest inventory and biophysical parameters. Progress in Physical Geography 22 (4), 449–476. Zheng, D., Rademacher, J., Chen, J., Crow, T., Bressee, M., Moine, J.L., Ryu, S., 2004. Estimating aboveground biomass using Landsat ETM+ data across a managed landscape in northern Wisconsin, USA. Remote Sensing of Environment.