evaluation of modis gross primary productivity of ... - Semantic Scholar
EVALUATION OF MODIS GROSS PRIMARY PRODUCTIVITY OF TROPICAL OIL PALM IN SOUTHERN PENINSULAR MALAYSIA Kian Pang Tana, Kasturi Devi Kanniaha, Ibrahim Bin Busua, Arthur Philip Cracknellb a
Department of Remote Sensing, Faculty of Geoinformation and Real Estate, Universiti Teknologi Malaysia, UTM Skudai, 81310 Johor, Malaysia.
b
Division of Electronic Engineering and Physics, University of Dundee, Dundee DDI 4HN, Scotland. ABSTRACT
Evaluation of the MODIS Gross Primary Productivity (GPP) or the MOD17A2 product has not been carried out for croplands, such as oil palm ecosystems in tropical regions. Thus, evaluation of the MOD17A2 is important in order to determine its usefulness for carbon related studies of oil palm. MOD17A2 and its inputs, such as MODIS land cover (MOD12Q1), meteorological data from Global NCEP/DOE II reanalysis dataset, LAI/fPAR (MOD15A2) and light use efficiency (LUE) were evaluated. Of the area of oil palm identified from a land use map, only 40.22% was correctly identified by croplands of MOD12Q1 as oil palm and the remaining 59.78% was misclassified as croplands and forest. From Global NCEP/DOE II reanalysis dataset, photosynthetic active radiation shows the lowest correlation (R2=0.02) whilst daily minimum temperature (Tmin) only agreed moderately (R2=0.26) compared to in-situ meteorological data. LAI and fPAR were underestimated by 10.00-16.79% and 9.66-10.80% respectively by MOD15A2. GPP was underestimated 19.10-29.48% by MOD17A2. LUE from site is higher than MODIS LUE of croplands for 0.19 g/m2/MJ/day. Index Terms— MODIS, gross primary productivity (GPP), evaluation, oil palm 1. INTRODUCTION Estimating primary productivity (PP) of various biomes has been conducted globally [1-2]. However, there are only a few studies have been conducted in tropical regions, i.e. tropical evergreen/deciduous forest [3-5] and tropical savanna [6-8]. Study on the PP of ecosystems is essential since it is a quantitative measurement of carbon intake by vegetation. In tropical regions, there is still a lack of studies estimating PP particularly for tree crops [9]. While the understanding of PP of tropical tree crops ecosystems remains scarce as well as tends to lag behind that of tropical
978-1-4577-1005-6/11/$26.00 ©2011 IEEE
756
forest ecosystems, estimating PP in tropical tree crops expressly for oil palm plantation (Elaeis guineensis) is critical. This is because oil palm is a prominent tropical tree crops that occupies over 14.70 Mha globally [10] and 4.49 Mha in Malaysia [11]. Presently, it has been expanded and has become one of the fastest developing agricultural crops in the tropical regions [12]. Because the demand for palm oil has increased, oil palm plantations in Malaysia as well as globally are increasing year by year. Consequently, there has been considerable interest to understand the PP and dynamics of oil palm estates in southern Peninsular Malaysia. Studies on PP of oil palm estates can provide a means for better land management to increase the productivity of oil palm that can benefit the economy of Malaysia. Moderate Resolution Imaging Spectroradiometer sensor (MODIS) onboard Terra and Aqua satellites are operationally estimating gross primary productivity (GPP) [13]. Near real-time MODIS GPP, so-called MOD17A2 at 1 km spatial and 8 day temporal resolution, have only been validated/calibrated mainly in forest ecosystems [14]. Various approaches were conducted to improve the product [15-16]. The improved product is disseminated in newer collections/versions from time to time. One of the assumptions of MOD17A2 is to treat all the croplands around the world by using the same MODIS Biome Properties Look-Up Table (BPLUT) and light use efficiency without taking into account the different species of crops in various climate regions of the world [13]. This can cause errors in GPP estimation of crops. Since validation of the MOD17A2 product has not been carried out for oil palm in tropical regions, this study evaluates MOD17A2 in order to determine its usefulness for carbon related studies. Besides, various studies have shown that the product can be further improved. Hence, this study will provide insights to further improve the MOD17A2 algorithm as well as to suggest possible ways to estimate the PP of oil palm more accurately.
IGARSS 2011
2. DATA AND METHODS 2.1. Site observations The study site was located at 103o18’55’’E, 2o3’40’’N. It is an area of homogenously planting of matured oil palm. In addition, it is classified as oil palm based on the land use map of Johor state (southern Peninsular Malaysia), year 2006, with scale of 1:250,000 as well as classified as croplands by the MODIS land cover (MOD12Q1). Site observations such as temperature, relative humidity, shortwave radiation and precipitation were obtained from the Kluang station (103o19’E, 2o1’N). Southern Peninsular Malaysia (Fig.1) has provided a suitable opportunity to be a test site because of the large coverage areas of oil palm estates which is comparable with the size of a pixel of MODIS or even multiple pixels [17]. Moreover, oil palm estates in Peninsular Malaysia contain a homogeneous and single species plantation planted on a regular pattern, which could reduce uncertainties using the light use efficiency (LUE) estimation. Some variables such as maximum light use efficiency were obtained from literature and ancillary data [18-19] when the ground measurements were absent or non-reliable.
Southeast Asia Study site Peninsular Malaysia
constraints such as VPD under dry conditions and Tmin under extremely cold conditions using linear ramp functions [16]. MOD17A2 product at MODIS location tile (h28v08) was downloaded from the Numerical Terradynamic Simulation Group (NTSG) website, available at ftp://ftp.ntsg.umt.edu/pub/MODIS/, for two years (20082009). MOD17A2 imageries were reprojected and subset, corresponding to the geo-location of the site observations. Multi-temporal GPP values were extracted and the unit was converted to g C/m2/day. MOD12Q1 for year 2004 and MOD15A2 (LAI/fPAR) collection 5 (2008-2009) were downloaded from the Land Processes Distributed Active Archive Center. Meteorological inputs were obtained from Global NCEP/DOE II reanalysis dataset. Only single pixel of meteorological, MOD15A2 and MOD17A2 that corresponding to the site observations were chosen for the evaluation. 2.4. Statistical analysis A confusion matrix was used to validate the MOD12Q1 (land cover). Mean, correlation coefficient (R2) and root mean square error (RMSE) were calculated to evaluate the discrepancy of MOD17A2 product and site observations. Relative predictive error (RPE) was used to measure the percentage of underestimation or overestimation of remotely sensed data. 3. RESULTS AND ANALYSIS
103 o 18’55’’E, 2o 3’40’’N
3.1. MODIS Land cover (MOD12Q1) Fig.1 Study site in southern Peninsular Malaysia 2.2. MOD17A2 and its related inputs MODIS utilises a modified version of the LUE algorithm of Monteith [20] to estimate GPP. The MODIS algorithm used to estimate GPP is: GPP = APAR x LUE APAR is then calculated as fPAR x PAR LUE = maximum LUE x Tmin (daily minimum temperature) x VPD (daytime vapour pressure deficit) Where fractional photosynthetic active radiation (fPAR) data is from MOD15A2 product in which the value utilized is based on maximum fPAR across the 8 days [16]. Daily climate data (shortwave radiation, VPD and Tmin) used in GPP estimations by the MOD17A2 product are provided by the Global NCEP/DOE II reanalysis dataset. Photosynthetically active radiation (PAR) is equal to shortwave radiation (SWrad) multiplied by 0.45 [16]. The estimated maximum potential LUE, which can be obtained from MODIS BPLUT is further reduced by environmental
757
a
b
Fig.2.a) Croplands of MOD12Q1 b) Oil palm of land use map
Croplands (class 12) from University of Maryland (UMD) land cover scheme of MOD12Q1, was extracted and subset to the southern part of Peninsular Malaysia. The croplands of MOD12Q1 were compared with the land use map of Johor state. Of the area of oil palm identified from the land use map, only 40.22% was correctly identified by croplands of MOD12Q1 as oil palm and the remaining 59.78% was misclassified as croplands and forest. Misclassification of oil palm occurred mainly because MODIS cannot sense the specific species of croplands (Zhao, personal communication).
Table 3 Comparison of varriables of MOD15A2 with site observations for years 2008 and 2009
Table 1 Comparison between variables of Global NCEP/DOE II reanalysis dataset with site observations for years y 2008 and 2009 Year
Variable
2008
Tmin (oC) Tmin (oC) VPD (kPa) VPD (kPa) PAR (MJ/m2/d) PAR (MJ/m2/d)
2009 2008 2009 2008 2009
Global NCEP/DOE II reanalysis (mean) 24.28
Site observaations (mean))
RMSE
RPE(%)
23.10
1.27
5.13
24.33
23.24
1.18
4.66
0.59
0.58
0.08
2.16
0.63
0.51
0.17
22.95
5.53
6.05
1.68
-8.57
5.84
6.05
0.91
-3.45
Year
Variable
MOD15A2 (mean)
RMSE
RPE(%)
5.41
Site observatioons (mean) 6.50
2008
LAI (m2 m-2)
1.81
-16.79
2009
LAI
5.85
6.50
1.54
-10.0
fPAR fPAR
0.85 0.86
0.95 0.95
0.13 0.14
-10.80 -9.66
2008
GPP (gC/m2/day) GPP (gC/m2/day)
RPE(%)
3.09
-29.48
6.53
8.07
2.15
-19.10
ϱϬϬ ϰϬϬ ϯϬϬ ϮϬϬ ϭϬϬ Ϭ J F M A M J J A S O N D
'WW;ŐͬŵϮͬĚĂLJͿ
ϭϬ͘ϬϬ ϴ͘ϬϬ ϲ͘ϬϬ ϰ͘ϬϬ Ϯ͘ϬϬ Ϭ͘ϬϬ
Ϯ ϮϬϬϴ MOD17
GPP site observvations
Precipitation
Fig.4. Comparison between MOD17A2 (g C/m2/day), GPP calculated using site observatioons as input (g C/m2/day), and the variation of precipitation (mm/m month) for year 2008.
ϭϬ͘ϬϬ ϴ͘ϬϬ ϲ͘ϬϬ ϰ͘ϬϬ Ϯ͘ϬϬ Ϭ͘ϬϬ
3.5. MODIS GPP (MOD17A2) In general, MOD17A2 product underestimates GPP of oil palm most of the time. GPP off MOD17A2 was underestimated 29.48 % and 19.10% for years 2008 and 2009 respectively (Table 3). The RM MSE was 3.09 g C
RMSE
5.59
Site observations (mean) 7.93
Southern Peninsular Malaysia received high precipitation (mean=196.80 mm/month) m for years 2008 and 2009. Generally, the variatiion of MODIS GPP does not agree well with the GPP callculated using site observations as input. Low MODIS GPP P could be relatively related to high precipitation (Fig.4-5). This is because during the high precipitation, high cloud covers take place and this could cause low values of LAI/fP PAR, low signal reaching the sensor and etc. However, hiigh precipitation does not much affect the variation of GPP ussing site observations as input.
(m2 m-2) 2008 2009
MOD17A2 (mean)
'WW;ŐͬŵϮͬĚĂLJͿ
Table 2 Comparison of variables of MOD15A2 M with site observations for years 2008 and 2009
Variable
2009
3.3. LAI/fPAR (MOD15A2) The annual mean LAI from MOD15A22 ranges from 5.415.85 m2/m2 whereas LAI from site obserrvations was uniform (6.50 m2/m2) (Table 2). Generally, LA AI from MOD15A2 was underestimated by 16.79% and 10.00% compared with the LAI from site observations for yeear 2008 and 2009 respectively. The annual mean fPAR value froom MOD15A2 was 0.85-0.86 for years 2008 and 2009 resspectively (Table 2). Meanwhile, the mean fPAR from site observations (calculated using a simple Beer’s Law)) for both years was 0.95, fPAR of MOD15A2 was underesstimated by 10.80% and 9.66% for years 2008 and 2009 resppectively.
Year
WƌĞĐŝƉŝƚĂƚŝŽŶ;ŵŵͬŵŽŶƚŚͿ
dataset were overestimated meteorolog gical data from site observations by 4.66-5.13% and 2.16-222.95% respectively. Whilst, PAR was underestimated by 3.445-8.57 % (Table 1). From Global NCEP/DOE II reanalysis dataset, PAR shows the lowest correlation (R2=0.02), followed f by VPD (R2=0.07) whereas Tmin agreed moderaately (R2=0.26) with site observations.
ϰϬϬ ϯϬϬ ϮϬϬ ϭϬϬ Ϭ
ϮϬϬϵ MOD17
GPP site obserrvations
ƉƌĞĐŝƉŝƚĂƚŝŽŶ;ŵŵͬŵŽŶƚŚͿ
Tmin and VPD from Global NCEP//DOE II reanalysis
/m2/day and 2.15 g C /m2/day / for years 2008 and 2009 respectively. LUE calculateed from site observations was 1.40 g /m2/MJ/day [18]. Th his value is 0.19 g/m2/MJ/day higher than MODIS LUE off croplands (1.21 g/m2/MJ/day). (BPLUT provided by Zhao, personal communication, dated 25 April, 2010).
J F M A M J J A S O N D
3.2. Global NCEP/DOE II reanalysis dataset d
Precipitation
Fig.5. Comparison between MOD17A2 (g C/m2/day), GPP calculated using site observatioons as input (g C/m2/day), and the variation of precipitation (mm/m month) for year 2009.
758
4. CONCLUSION In this study, we evaluated the MODIS GPP (MOD17A2) and its inputs. We relied heavily on in-situ data obtained from the literature and meteorological data provided by the Meteorological Department of Malaysia as well as the land use map of Johor state, year 2006 from the Agriculture Department of Malaysia. In general, MODIS GPP for oil palm does not agree well with the GPP of oil palm calculated using site observations as input. Therefore, we suggest that further studies must be carried out to model GPP of oil palm over regional scale using remote sensing data. Acknowledgements This study is financially supported by Universiti Teknologi Malaysia Research Management Centre, UTM Zamalah Scholarship and IGARSS 2011 travel grant. The authors would like to thank the Land Processes Distributed Active Archive Center for providing near real-time and free of charge MODIS dataset. Special thanks to Dr. Maosheng Zhao from the NTSG for providing datasets for this research. Literature from Dr. Ian E. Henson and Dr. Haniff Harun as well as site observations data from Malaysia Palm Oil Board, Meteorological Department of Malaysia and Agriculture Department of Malaysia are also acknowledged. 5. REFERENCES [1] S.W. Running, D.D. Baldocchi, D.P Turner, S.T. Gower, P.S. Bakwin and K.A. Hibbard, “A global terrestrial monitoring network integrating tower fluxes, flask sampling , ecosystem modeling and EOS satellite data,” Remote Sensing of Environment, vol. 70, pp 108-127, 1999. [2] S.W. Running,, R.R. Nemani, F.A. Heinsch,, M. Zhao, M. Reeves and H. Hashimoto, “A continuous satellite-derived measure of global terrestrial primary production,” Bioscience, vol.54, no.6, pp 547-560, 2004. [3] X. Xiao, Q. Zhang, S. Saleska, L. Hutyra, P.D. Camargo, S. Wofsy, S. Frolking, S. Boles, M. Keller and B. Moore III, “Satellite- based modeling of gross primary production in a seasonally moist tropical evergreen forest,” Remote Sensing of Environment, vol. 94, pp 105-122, 2005. [4] M. Gebremichael and A.P. Barros, “Evaluations of MODIS gross primary productivity (GPP) in tropical monsoon regions,” Remote Sensing of Environment, vol.100, pp 150166, 2006. [5] D.A. Clark, S. Brown, D.W. Kicklighter, J.Q. Chambers, J.R. Thomlinson, J. Ni and E.A. Holland , “Net primary production in tropical forest: An evaluation and synthesis of existing field data,” Ecological Applications, vol. 11, pp 371-384, 2001. [6] K.D. Kanniah, J. Beringer, L.B. Hutley, N.J. Tapper and X. Zhu, “Evaluation of collection 4 and 5 of the MODIS gross primary productivity product and algorithm improvement at a tropical savanna site in northern Australia,” Remote Sensing of Environment,vol. 113, pp 1808-1822, 2009.
759
[7] L.B. Hutley, R. Leauning, J. Beringer and H.A. Cleugh, “The utility of the eddy covariance techniques as a tool in carbon accounting: tropical savanna as a case study,” Australian Journal of Botany,vol. 53, pp 663-675, 2005. [8] R. Leauning, H.A. Cleugh, S.J. Zegelin and D. Hughes, “Carbon and water fluxes over a temperate Eucalyptus forest and a tropical wet/dry savanna in Australia: Measurements and comparison with MODIS remote sensing estimates,” Agricultural and Forest Meteorology, vol. 129, pp 151-173, 2005. [9] M.N.V. Navarro, C. Jourdan, T. Sileye, S. Braconnier, I. Mialet-serra, L. Saint-andre, J. Dauzat, Y. Nouvellon, D. Epron, J.M. Bonnefond, P. Berbigier, A. Rouziere, J.P. Bouillet and O. Roupsard, “Fruit development, not GPP, drives seasonal variation in NPP in a tropical palm plantation,” Tree Physiology, vol. 28, pp 1661-1674, 2008. [10] FAO, “FAOSTAT agriculture data”. Available at: http://apps/fao.org/, 2009. [11] Malaysian Palm Oil Board, “Malaysia Palm Oil Industry Performance 2009,” Global Oils & Fats Business Magazine, vol. 7, no. 1, pp 1-11, 2010. [12] L.P. Koh and D.S. Wilcove, “Is oil palm agriculture really destroying tropical biodiversity?” Conservation Letters, vol. 1, pp 60-64, 2008. [13] F.A. Heinsch, M. Reeves, P. Votava, S. Kang, C. Milesi and M. Zhao, “User’s Guide: GPP and NPP (MOD17A2/A3) Products NASA MODIS Land Algorithm” Version 2.0, December 2, 2003. [14] S. Plummer, “On Validation of the MODIS Gross Primary Production Product” IEEE Transactions on Geoscience and Remote Sensing, vol. 44, no.7, pp 1936-1938, 2006. [15] M. Zhao, F. A. Heinsch, R.R. Nemani and S. W. Running , “Improvements of the MODIS terrestrial gross and net primary production global data set,” Remote Sensing of Environment, vol. 95, pp164-176, 2005. [16] M. Zhao, S. W. Running and R.R. Nemani, “Sensitivity of Moderate Resolution Imaging Spectroradiometer (MODIS) terrestrial primary production to the accuracy of metrological reanalyses,”.Journal of Geophysical Research, vol. 111: doi : 10.1029/2004JG000004, pp 1-13, 2006. [17] A.P. Cracknell, “The MODIS NPP (net primary productivity) product- concept, definition and validation,” Proceedings of the Remote Sensing and Photogrammetric Society Conference, Burlington House, London, 2010. [18] Mohd Roslan Md Noor and Mohd Haniff Harun, “The Role of Leaf Area Index (LAI) in Oil Palm,” Oil Palm Bulletin 48, pp 11-16, 2004. [19] I.E. Henson and Mohd Haniff Harun, “The influence of climatic conditions on gas and energy exchanges above a young oil palm stand in north Kedah, Malaysia,” Journal of Oil Palm Research, Vol. 17, pp73-91, 2005. [20] J.L. Monteith, “Solar radiation and productivity in tropical ecosystem,” Journal of Applied Ecology, vol. 9, pp 747-766, 1972.
Department of Remote Sensing, Faculty of Geoinformation and Real Estate, Universiti Teknologi Malaysia, UTM Skudai, 81310 Johor, Malaysia.
b
Division of Electronic Engineering and Physics, University of Dundee, Dundee DDI 4HN, Scotland. ABSTRACT
Evaluation of the MODIS Gross Primary Productivity (GPP) or the MOD17A2 product has not been carried out for croplands, such as oil palm ecosystems in tropical regions. Thus, evaluation of the MOD17A2 is important in order to determine its usefulness for carbon related studies of oil palm. MOD17A2 and its inputs, such as MODIS land cover (MOD12Q1), meteorological data from Global NCEP/DOE II reanalysis dataset, LAI/fPAR (MOD15A2) and light use efficiency (LUE) were evaluated. Of the area of oil palm identified from a land use map, only 40.22% was correctly identified by croplands of MOD12Q1 as oil palm and the remaining 59.78% was misclassified as croplands and forest. From Global NCEP/DOE II reanalysis dataset, photosynthetic active radiation shows the lowest correlation (R2=0.02) whilst daily minimum temperature (Tmin) only agreed moderately (R2=0.26) compared to in-situ meteorological data. LAI and fPAR were underestimated by 10.00-16.79% and 9.66-10.80% respectively by MOD15A2. GPP was underestimated 19.10-29.48% by MOD17A2. LUE from site is higher than MODIS LUE of croplands for 0.19 g/m2/MJ/day. Index Terms— MODIS, gross primary productivity (GPP), evaluation, oil palm 1. INTRODUCTION Estimating primary productivity (PP) of various biomes has been conducted globally [1-2]. However, there are only a few studies have been conducted in tropical regions, i.e. tropical evergreen/deciduous forest [3-5] and tropical savanna [6-8]. Study on the PP of ecosystems is essential since it is a quantitative measurement of carbon intake by vegetation. In tropical regions, there is still a lack of studies estimating PP particularly for tree crops [9]. While the understanding of PP of tropical tree crops ecosystems remains scarce as well as tends to lag behind that of tropical
978-1-4577-1005-6/11/$26.00 ©2011 IEEE
756
forest ecosystems, estimating PP in tropical tree crops expressly for oil palm plantation (Elaeis guineensis) is critical. This is because oil palm is a prominent tropical tree crops that occupies over 14.70 Mha globally [10] and 4.49 Mha in Malaysia [11]. Presently, it has been expanded and has become one of the fastest developing agricultural crops in the tropical regions [12]. Because the demand for palm oil has increased, oil palm plantations in Malaysia as well as globally are increasing year by year. Consequently, there has been considerable interest to understand the PP and dynamics of oil palm estates in southern Peninsular Malaysia. Studies on PP of oil palm estates can provide a means for better land management to increase the productivity of oil palm that can benefit the economy of Malaysia. Moderate Resolution Imaging Spectroradiometer sensor (MODIS) onboard Terra and Aqua satellites are operationally estimating gross primary productivity (GPP) [13]. Near real-time MODIS GPP, so-called MOD17A2 at 1 km spatial and 8 day temporal resolution, have only been validated/calibrated mainly in forest ecosystems [14]. Various approaches were conducted to improve the product [15-16]. The improved product is disseminated in newer collections/versions from time to time. One of the assumptions of MOD17A2 is to treat all the croplands around the world by using the same MODIS Biome Properties Look-Up Table (BPLUT) and light use efficiency without taking into account the different species of crops in various climate regions of the world [13]. This can cause errors in GPP estimation of crops. Since validation of the MOD17A2 product has not been carried out for oil palm in tropical regions, this study evaluates MOD17A2 in order to determine its usefulness for carbon related studies. Besides, various studies have shown that the product can be further improved. Hence, this study will provide insights to further improve the MOD17A2 algorithm as well as to suggest possible ways to estimate the PP of oil palm more accurately.
IGARSS 2011
2. DATA AND METHODS 2.1. Site observations The study site was located at 103o18’55’’E, 2o3’40’’N. It is an area of homogenously planting of matured oil palm. In addition, it is classified as oil palm based on the land use map of Johor state (southern Peninsular Malaysia), year 2006, with scale of 1:250,000 as well as classified as croplands by the MODIS land cover (MOD12Q1). Site observations such as temperature, relative humidity, shortwave radiation and precipitation were obtained from the Kluang station (103o19’E, 2o1’N). Southern Peninsular Malaysia (Fig.1) has provided a suitable opportunity to be a test site because of the large coverage areas of oil palm estates which is comparable with the size of a pixel of MODIS or even multiple pixels [17]. Moreover, oil palm estates in Peninsular Malaysia contain a homogeneous and single species plantation planted on a regular pattern, which could reduce uncertainties using the light use efficiency (LUE) estimation. Some variables such as maximum light use efficiency were obtained from literature and ancillary data [18-19] when the ground measurements were absent or non-reliable.
Southeast Asia Study site Peninsular Malaysia
constraints such as VPD under dry conditions and Tmin under extremely cold conditions using linear ramp functions [16]. MOD17A2 product at MODIS location tile (h28v08) was downloaded from the Numerical Terradynamic Simulation Group (NTSG) website, available at ftp://ftp.ntsg.umt.edu/pub/MODIS/, for two years (20082009). MOD17A2 imageries were reprojected and subset, corresponding to the geo-location of the site observations. Multi-temporal GPP values were extracted and the unit was converted to g C/m2/day. MOD12Q1 for year 2004 and MOD15A2 (LAI/fPAR) collection 5 (2008-2009) were downloaded from the Land Processes Distributed Active Archive Center. Meteorological inputs were obtained from Global NCEP/DOE II reanalysis dataset. Only single pixel of meteorological, MOD15A2 and MOD17A2 that corresponding to the site observations were chosen for the evaluation. 2.4. Statistical analysis A confusion matrix was used to validate the MOD12Q1 (land cover). Mean, correlation coefficient (R2) and root mean square error (RMSE) were calculated to evaluate the discrepancy of MOD17A2 product and site observations. Relative predictive error (RPE) was used to measure the percentage of underestimation or overestimation of remotely sensed data. 3. RESULTS AND ANALYSIS
103 o 18’55’’E, 2o 3’40’’N
3.1. MODIS Land cover (MOD12Q1) Fig.1 Study site in southern Peninsular Malaysia 2.2. MOD17A2 and its related inputs MODIS utilises a modified version of the LUE algorithm of Monteith [20] to estimate GPP. The MODIS algorithm used to estimate GPP is: GPP = APAR x LUE APAR is then calculated as fPAR x PAR LUE = maximum LUE x Tmin (daily minimum temperature) x VPD (daytime vapour pressure deficit) Where fractional photosynthetic active radiation (fPAR) data is from MOD15A2 product in which the value utilized is based on maximum fPAR across the 8 days [16]. Daily climate data (shortwave radiation, VPD and Tmin) used in GPP estimations by the MOD17A2 product are provided by the Global NCEP/DOE II reanalysis dataset. Photosynthetically active radiation (PAR) is equal to shortwave radiation (SWrad) multiplied by 0.45 [16]. The estimated maximum potential LUE, which can be obtained from MODIS BPLUT is further reduced by environmental
757
a
b
Fig.2.a) Croplands of MOD12Q1 b) Oil palm of land use map
Croplands (class 12) from University of Maryland (UMD) land cover scheme of MOD12Q1, was extracted and subset to the southern part of Peninsular Malaysia. The croplands of MOD12Q1 were compared with the land use map of Johor state. Of the area of oil palm identified from the land use map, only 40.22% was correctly identified by croplands of MOD12Q1 as oil palm and the remaining 59.78% was misclassified as croplands and forest. Misclassification of oil palm occurred mainly because MODIS cannot sense the specific species of croplands (Zhao, personal communication).
Table 3 Comparison of varriables of MOD15A2 with site observations for years 2008 and 2009
Table 1 Comparison between variables of Global NCEP/DOE II reanalysis dataset with site observations for years y 2008 and 2009 Year
Variable
2008
Tmin (oC) Tmin (oC) VPD (kPa) VPD (kPa) PAR (MJ/m2/d) PAR (MJ/m2/d)
2009 2008 2009 2008 2009
Global NCEP/DOE II reanalysis (mean) 24.28
Site observaations (mean))
RMSE
RPE(%)
23.10
1.27
5.13
24.33
23.24
1.18
4.66
0.59
0.58
0.08
2.16
0.63
0.51
0.17
22.95
5.53
6.05
1.68
-8.57
5.84
6.05
0.91
-3.45
Year
Variable
MOD15A2 (mean)
RMSE
RPE(%)
5.41
Site observatioons (mean) 6.50
2008
LAI (m2 m-2)
1.81
-16.79
2009
LAI
5.85
6.50
1.54
-10.0
fPAR fPAR
0.85 0.86
0.95 0.95
0.13 0.14
-10.80 -9.66
2008
GPP (gC/m2/day) GPP (gC/m2/day)
RPE(%)
3.09
-29.48
6.53
8.07
2.15
-19.10
ϱϬϬ ϰϬϬ ϯϬϬ ϮϬϬ ϭϬϬ Ϭ J F M A M J J A S O N D
'WW;ŐͬŵϮͬĚĂLJͿ
ϭϬ͘ϬϬ ϴ͘ϬϬ ϲ͘ϬϬ ϰ͘ϬϬ Ϯ͘ϬϬ Ϭ͘ϬϬ
Ϯ ϮϬϬϴ MOD17
GPP site observvations
Precipitation
Fig.4. Comparison between MOD17A2 (g C/m2/day), GPP calculated using site observatioons as input (g C/m2/day), and the variation of precipitation (mm/m month) for year 2008.
ϭϬ͘ϬϬ ϴ͘ϬϬ ϲ͘ϬϬ ϰ͘ϬϬ Ϯ͘ϬϬ Ϭ͘ϬϬ
3.5. MODIS GPP (MOD17A2) In general, MOD17A2 product underestimates GPP of oil palm most of the time. GPP off MOD17A2 was underestimated 29.48 % and 19.10% for years 2008 and 2009 respectively (Table 3). The RM MSE was 3.09 g C
RMSE
5.59
Site observations (mean) 7.93
Southern Peninsular Malaysia received high precipitation (mean=196.80 mm/month) m for years 2008 and 2009. Generally, the variatiion of MODIS GPP does not agree well with the GPP callculated using site observations as input. Low MODIS GPP P could be relatively related to high precipitation (Fig.4-5). This is because during the high precipitation, high cloud covers take place and this could cause low values of LAI/fP PAR, low signal reaching the sensor and etc. However, hiigh precipitation does not much affect the variation of GPP ussing site observations as input.
(m2 m-2) 2008 2009
MOD17A2 (mean)
'WW;ŐͬŵϮͬĚĂLJͿ
Table 2 Comparison of variables of MOD15A2 M with site observations for years 2008 and 2009
Variable
2009
3.3. LAI/fPAR (MOD15A2) The annual mean LAI from MOD15A22 ranges from 5.415.85 m2/m2 whereas LAI from site obserrvations was uniform (6.50 m2/m2) (Table 2). Generally, LA AI from MOD15A2 was underestimated by 16.79% and 10.00% compared with the LAI from site observations for yeear 2008 and 2009 respectively. The annual mean fPAR value froom MOD15A2 was 0.85-0.86 for years 2008 and 2009 resspectively (Table 2). Meanwhile, the mean fPAR from site observations (calculated using a simple Beer’s Law)) for both years was 0.95, fPAR of MOD15A2 was underesstimated by 10.80% and 9.66% for years 2008 and 2009 resppectively.
Year
WƌĞĐŝƉŝƚĂƚŝŽŶ;ŵŵͬŵŽŶƚŚͿ
dataset were overestimated meteorolog gical data from site observations by 4.66-5.13% and 2.16-222.95% respectively. Whilst, PAR was underestimated by 3.445-8.57 % (Table 1). From Global NCEP/DOE II reanalysis dataset, PAR shows the lowest correlation (R2=0.02), followed f by VPD (R2=0.07) whereas Tmin agreed moderaately (R2=0.26) with site observations.
ϰϬϬ ϯϬϬ ϮϬϬ ϭϬϬ Ϭ
ϮϬϬϵ MOD17
GPP site obserrvations
ƉƌĞĐŝƉŝƚĂƚŝŽŶ;ŵŵͬŵŽŶƚŚͿ
Tmin and VPD from Global NCEP//DOE II reanalysis
/m2/day and 2.15 g C /m2/day / for years 2008 and 2009 respectively. LUE calculateed from site observations was 1.40 g /m2/MJ/day [18]. Th his value is 0.19 g/m2/MJ/day higher than MODIS LUE off croplands (1.21 g/m2/MJ/day). (BPLUT provided by Zhao, personal communication, dated 25 April, 2010).
J F M A M J J A S O N D
3.2. Global NCEP/DOE II reanalysis dataset d
Precipitation
Fig.5. Comparison between MOD17A2 (g C/m2/day), GPP calculated using site observatioons as input (g C/m2/day), and the variation of precipitation (mm/m month) for year 2009.
758
4. CONCLUSION In this study, we evaluated the MODIS GPP (MOD17A2) and its inputs. We relied heavily on in-situ data obtained from the literature and meteorological data provided by the Meteorological Department of Malaysia as well as the land use map of Johor state, year 2006 from the Agriculture Department of Malaysia. In general, MODIS GPP for oil palm does not agree well with the GPP of oil palm calculated using site observations as input. Therefore, we suggest that further studies must be carried out to model GPP of oil palm over regional scale using remote sensing data. Acknowledgements This study is financially supported by Universiti Teknologi Malaysia Research Management Centre, UTM Zamalah Scholarship and IGARSS 2011 travel grant. The authors would like to thank the Land Processes Distributed Active Archive Center for providing near real-time and free of charge MODIS dataset. Special thanks to Dr. Maosheng Zhao from the NTSG for providing datasets for this research. Literature from Dr. Ian E. Henson and Dr. Haniff Harun as well as site observations data from Malaysia Palm Oil Board, Meteorological Department of Malaysia and Agriculture Department of Malaysia are also acknowledged. 5. REFERENCES [1] S.W. Running, D.D. Baldocchi, D.P Turner, S.T. Gower, P.S. Bakwin and K.A. Hibbard, “A global terrestrial monitoring network integrating tower fluxes, flask sampling , ecosystem modeling and EOS satellite data,” Remote Sensing of Environment, vol. 70, pp 108-127, 1999. [2] S.W. Running,, R.R. Nemani, F.A. Heinsch,, M. Zhao, M. Reeves and H. Hashimoto, “A continuous satellite-derived measure of global terrestrial primary production,” Bioscience, vol.54, no.6, pp 547-560, 2004. [3] X. Xiao, Q. Zhang, S. Saleska, L. Hutyra, P.D. Camargo, S. Wofsy, S. Frolking, S. Boles, M. Keller and B. Moore III, “Satellite- based modeling of gross primary production in a seasonally moist tropical evergreen forest,” Remote Sensing of Environment, vol. 94, pp 105-122, 2005. [4] M. Gebremichael and A.P. Barros, “Evaluations of MODIS gross primary productivity (GPP) in tropical monsoon regions,” Remote Sensing of Environment, vol.100, pp 150166, 2006. [5] D.A. Clark, S. Brown, D.W. Kicklighter, J.Q. Chambers, J.R. Thomlinson, J. Ni and E.A. Holland , “Net primary production in tropical forest: An evaluation and synthesis of existing field data,” Ecological Applications, vol. 11, pp 371-384, 2001. [6] K.D. Kanniah, J. Beringer, L.B. Hutley, N.J. Tapper and X. Zhu, “Evaluation of collection 4 and 5 of the MODIS gross primary productivity product and algorithm improvement at a tropical savanna site in northern Australia,” Remote Sensing of Environment,vol. 113, pp 1808-1822, 2009.
759
[7] L.B. Hutley, R. Leauning, J. Beringer and H.A. Cleugh, “The utility of the eddy covariance techniques as a tool in carbon accounting: tropical savanna as a case study,” Australian Journal of Botany,vol. 53, pp 663-675, 2005. [8] R. Leauning, H.A. Cleugh, S.J. Zegelin and D. Hughes, “Carbon and water fluxes over a temperate Eucalyptus forest and a tropical wet/dry savanna in Australia: Measurements and comparison with MODIS remote sensing estimates,” Agricultural and Forest Meteorology, vol. 129, pp 151-173, 2005. [9] M.N.V. Navarro, C. Jourdan, T. Sileye, S. Braconnier, I. Mialet-serra, L. Saint-andre, J. Dauzat, Y. Nouvellon, D. Epron, J.M. Bonnefond, P. Berbigier, A. Rouziere, J.P. Bouillet and O. Roupsard, “Fruit development, not GPP, drives seasonal variation in NPP in a tropical palm plantation,” Tree Physiology, vol. 28, pp 1661-1674, 2008. [10] FAO, “FAOSTAT agriculture data”. Available at: http://apps/fao.org/, 2009. [11] Malaysian Palm Oil Board, “Malaysia Palm Oil Industry Performance 2009,” Global Oils & Fats Business Magazine, vol. 7, no. 1, pp 1-11, 2010. [12] L.P. Koh and D.S. Wilcove, “Is oil palm agriculture really destroying tropical biodiversity?” Conservation Letters, vol. 1, pp 60-64, 2008. [13] F.A. Heinsch, M. Reeves, P. Votava, S. Kang, C. Milesi and M. Zhao, “User’s Guide: GPP and NPP (MOD17A2/A3) Products NASA MODIS Land Algorithm” Version 2.0, December 2, 2003. [14] S. Plummer, “On Validation of the MODIS Gross Primary Production Product” IEEE Transactions on Geoscience and Remote Sensing, vol. 44, no.7, pp 1936-1938, 2006. [15] M. Zhao, F. A. Heinsch, R.R. Nemani and S. W. Running , “Improvements of the MODIS terrestrial gross and net primary production global data set,” Remote Sensing of Environment, vol. 95, pp164-176, 2005. [16] M. Zhao, S. W. Running and R.R. Nemani, “Sensitivity of Moderate Resolution Imaging Spectroradiometer (MODIS) terrestrial primary production to the accuracy of metrological reanalyses,”.Journal of Geophysical Research, vol. 111: doi : 10.1029/2004JG000004, pp 1-13, 2006. [17] A.P. Cracknell, “The MODIS NPP (net primary productivity) product- concept, definition and validation,” Proceedings of the Remote Sensing and Photogrammetric Society Conference, Burlington House, London, 2010. [18] Mohd Roslan Md Noor and Mohd Haniff Harun, “The Role of Leaf Area Index (LAI) in Oil Palm,” Oil Palm Bulletin 48, pp 11-16, 2004. [19] I.E. Henson and Mohd Haniff Harun, “The influence of climatic conditions on gas and energy exchanges above a young oil palm stand in north Kedah, Malaysia,” Journal of Oil Palm Research, Vol. 17, pp73-91, 2005. [20] J.L. Monteith, “Solar radiation and productivity in tropical ecosystem,” Journal of Applied Ecology, vol. 9, pp 747-766, 1972.
