Evaluation of MODIS GPP product and scaling up GPP over Northern Australian savannas

1
Evaluation of MODIS GPP product and scaling up
GPP over Northern Australian savannas
Kasturi Devi KanniahJason Beringer Lindsay
Hutley Nigel TapperXuan Zhu
2
Objectives

• To validate different versions/collections of
  MODIS GPP (MOD17) -Collections 4.5, 4.8 and 5
• To validate input parameters used to estimate
  MODIS GPP - LAI/fPAR (MOD 15A2), Light Use
  Efficiency and meteorological variables (VPD,
  PAR)
• To estimate GPP using MOD17 algorithm with site
  specific values

3
Howard Springs

• Open woodland savannas forest 50-60 canopy
  cover.
• Over storey - evergreen trees
• Under storey - by C4 grasses
• Wet season GPP 7-8 g C m-2 day-1
• Dry season GPP 0.3 to 1.6 g C m-2 day-1

Wet season (Dec-Mac)
Dry season (May-Sept)
4
MODIS GPP
fPAR
MOD15A2
BCG model
Max. LUE
Light Use Efficiency
APAR
PAR
Tmin VPD scalar
GPP MOD17A
NASA DAO/GMAO
NASA DAO/GMAO
Global product, 1 km, 8 day Only useful if its
relative accuracy can be determined
5
MODIS Collections
6
Seasonal GPP pattern

• Correct seasonal pattern
• GPP Col. 4.5 5 lt 4.8
• Col 4.8- good agreement with tower in the wet
  (RPE 1, IOA 0.72, RMSE 1 g C m-2 day-1
  explained 75 variation in tower GPP.
• Poor performance in the dry (RPE 31, RMSE 1.4,
  IOA 0.59, R2 0.33)

• Col. 4.5 good in the dry (RPE 4, RMSE 1, IOA
  0.72 R2 0.35), but poor
• in the wet (RPE -14, RMSE 1.53, IOA 0.63 and
  R2 0.46)
• Col. 5 underestimated by 40 in the wet and 10
  in the dry

7
LAI/fPAR

• Wet season MODIS 3.8 vs. site 2.2
• Dry season LAI -MODIS 1.3 vs. site 0.9
• Wet- MODIS fPAR 0.90 vs. site fPAR 0.67
• Dry- MODIS 0.67 vs. site 0.35
• correct LAI fPAR in Col. 5
• Rapid increase in fPAR from September

8
Meteorology
Underestimation of PAR in the wet season-RPE 9
in DAO 11 in GMAO In the dry- underestimation
of 5-6 Underestimation of VPD scalar in DAO- 4,
but negligible in GMAO In the dry-
underestimation 11 in DAO 17 in GMAO
9
Maximum LUE

• LUE GPP/APAR
• Site specific ?max 1.26 g C MJ-1
• 17 higher than standard MODIS algorithm value of
  1.03 gCMJ-1 in col. 4.8
• 35 higher than col. 4.5 (0.80 gCMJ-1)

10
Source of error
Test 1- MODIS LUE Test 2- MODIS meteorology Test
3- MODIS fPAR
11
Algorithm improvements

• ? GPP was recalculated using MOD17 algorithm but
  with site specific values
• ? GPP was recalculated using MOD17 algorithm with
  VPD scalar was replaced with soil moisture index.
• Evaporative Fraction LE/(LEH) from flux tower
• EF - indicator of soil or vegetation moisture
  conditions because decreasing amounts of energy
  partitioned into latent heat flux suggests a
  stronger moisture limitation

12
Improved methods

• Improved methods captured the start of the wet
  season correctly- used correct fPAR
• Method with VPD still overestimates GPP in the
  dry season
• Method with EF accurately reduce GPP in the dry
  season captured the beginning of the wet
  season.

Overall these methods reduced RPE by 50, RMSE
by 42, increased IOA by 6 compared to Col. 4.8
and explained gt90 variation in tower GPP
13
Conclusion

• MODIS - reasonable estimation of GPP (lt?12) -
  annual basis and perfect in the wet in Col. 4.8
  (1) and in the dry in Col. 4.5 (4).
• Main source of error in MODIS- fPAR, and use of
  VPD as a surrogate for soil water deficit in the
  dry season
• Overestimation in fPAR was compensated by
  relatively low PAR, VPD scalar and LUE in the wet
  season.
• In the dry season, VPD scalar PAR was
  underestimated, but high fPAR resulted in the
  overestimation of GPP
• Col. 5 fPAR accurate but low PAR LUE-
  underestimated GPP- LUT need to be readjusted
• Use of VPD in MOD17 has limitation-arid semi
  arid areas

14
Future work

• Validation at other locations
• Analyse the spatial temporal patterns of GPP
  over NT using MOD 4.5 4.8
• Estimate GPP using fPAR from collection 5
  other site specific values