Transcript Document

Estimating aboveground biomass: remote
sensing approach
Daolan Zheng
Dept. of EEES, University of Toledo
Abstract
*** Aboveground biomass (AGB)—2.5 cm dbh for stands > 5 yrs and > 1.3 m for
stands < 5 yrs.
*** Bridges applications of remote sensing (RS) with forest practices in Chequamegon
National Forest, WI by producing high-resolution maps of age and AGB.
*** We coupled AGB values, calculated from tree dbh., with RS indices to produce
initial biomass map (IBM).
*** The IBM was overlaid with cover map to generate age map using biomass
threshold values for each age category (e.g. young, intermediate and mature) based on
field data and frequency analysis of IBM.
*** Hardwood forests--stand age and NIR (r2 = 0.95).
*** Pine forests were strongly related to NDVIc (r2 = 0.86)
*** The total AGB in 2001 = 3.3 million metric tons (dry weight), 76.5% of which was
in hardwood and mixed hardwood/pine forests. Ranged from 1 to 358 Mg/ha with an
average of 70 and standard deviation of 54 Mg/ha. The modal AGB class 81-100
Mg/ha (16.1%). 200 Mg/ha < 3% (mature HW).
*** Validation (R2 = 0.67, p < 0.001). The AGB and age maps can be used for
quantifying the regional carbon budget, fuel accumulation, or monitoring management
practices.
Introduction
AGB is a necessary component for studying productivity, carbon cycles, nutrient
allocation, and fuel accumulation;
RS techniques allow to examine properties and processes of ecosystems and their interannual variability at multiple scales because of ability to observe large areas/high revisitation frequencies; Success in estimating forest biomass/production using RS
reported worldwide. The models need local calibration with ground data;
Spectral Vegetation Index (SVI), Simple Ratio (SR), NDVI, and NDVIc are useful Xs
for estimating LAI/biomass/productivity;
Stand-level AGB is a function of species composition, tree height, basal area, and stand
structure, but dbh. is the most commonly used variable.
Species-based biomass models are accurate at tree/plot/stand but can’t be used across
the landscape for pattern analysis unless linked to RS data.
Combining carbon pool assessments from existing inventories with RS variables is
one of priorities identified in the NACP.
Research gaps--lack of a high-resolution stand-age map for ecological analyses in
the CNF. The existing USDA stand-age map developed for other purposes has coarse
spatial resolution, limited availability, and is infrequently updated.
Overall objectives: Combining field observations and RS data to 1)
produce a high-resolution age map; 2) generate a spatially explicit
AGB map; and 3) examine spatial patterns of AGB in CNF.
Three specific steps:
a) estimation of initial AGB by coupling field measurements with
solely RS data through stepwise regressions for HW, pine forests,
and both;
b) obtaining a landscape age map by overlaying the initial AGB map
with an existing land-cover map using biomass threshold values,
determined by frequency analysis and field observations, to
separate Y/INTER/MATURE hardwood and pine forests;
c) refining the initial landscape AGB estimates using a combination of
newly developed models incorporating age variable.
Methods
1) Study area CNF, WI ;
a)
Climate—Annual mean of 4.7 (0C) with a short/hot summer and cold winter,
Growing season (120-140 days), PPT 660-700mm/yr;
b) Soil--Wisconsin-age glaciated landscape with deep, coarse-textur;
c)
Topography--flat to rolling (Ele. 232-459 m);
d) Cover types—HW, JP, RP, MIX, RFS, and NFBG (Bresee et al. (2004). Forests
were harvested at an average age between 65 – 70 years (USDA, 1986), resulted
in more or less the even age forest structure.
2) Field design and measurements of tree dbh
a) Model development: 55 circular plots (2002). Continuous stands--2.6 km2 for
M/INTER and 1.3 km2 for Y and CC across cover types (i.e., RP, JP and HW).
In each type, 4 age classes were sampled (i.e., 3-8, 15-20, 32-40, and 65-75
years), total of 12 stands. In each stand, 4-5 plots were set around its center. All
plots (except for young hardwood, 0.01 ha) was approximately 0.05-ha. Within
each 0.05-ha plot the dbh. of all trees (> 2.5 cm dbh.) and the average stand age
of the plot was determined by tree-ring analysis and recorded. 1.3 m taller for
YHW.
Continued with METHODS
b)
Validation: 40 additional plots were selected in 2003 following the same criteria as in
2002. Once a suitable stand was found, a random number table was used to determine
plot location (i.e., compass bearing and distance). dbh. of the trees in each sub area (i.e.,
0.05 or 0.01 ha) were measured and adjusted
3) Biomass estimation
Models (species-based) developed in the area used first, if not available the ones from
the geographically closest regions were used.
4) Remotely sensed indices
a)
The image was geo-rectified to UTM and the raw satellite data in each ETM+ band
(except thermal and panchromatic) were converted to reflectance using an exoatmospheric model.
b)
Six individual bands (B, G, R, NIR, and 2 middle infrared), and 5 vegetation indices
including: 1) ratio of blue/red, 2) NDVI (NIR – red) / (NIR + red), (Rouse et al., 1973),
3) simple ratio (NIR/red), 4) Modified Soil Adjusted Vegetation Index (MSAVI) = (ρNIR
– ρred) / (ρNIR – ρred + L) * (1 + L), L is a soil-adjustment factor (Qi et al., 1994), and 5)
NDVIc calculated from NDVI * [1 – (mIR – mIRmin) / (mIRmax – mIRmin)] (Nemani et
al., 1993).
Continued with METHODS
5) Relating ground data with RS products to produce
maps of initial AGB, age, and final AGB
RESULTS
Remote-sensing derived variables including MSAVI, bands of red, near-infrared
(NIR), and middle-infrared (MIR), were useful predictors of AGB (Table 2). The
overall model explained 82% of variance (a=0.001). However, better models were
achieved by separating the plots into hardwood and pine forests. Hardwood AGB
was strongly related to stand age and NIR (r2 = 0.95) using a linear model, while
AGB for pine forests was strongly related to NDVIc using a sigmoidal model (r2 =
0.86).
Statistic models used for calculating aboveground biomass (AGB, Mg/ha).
__________________________________________________________________
Models
AGB = 48.8 * (NIR/red) + 2.3 * Age – 454 * MASVI - 38
AGB = 111 * (NDVIc10.3 / (NDVIc10.3 + 0.3510.3))
AGB = 232.5 * NIR + 2.7 * Age – 71
Description N
Overall
r2
55 0.82
Pine
35 0.86
Hardwood 20 0.95
140
AGB = 111 * (NDVIc
2
r = 0.86
120
10.3
/ (NDVIc
10.3
10.3
+ 0.35
))
AGB(Mg/ha)
100
80
60
40
20
0
0
0.1
0.2
0.3
NDVIc
0.4
0.5
The final predicted AGB values across the landscape ranged from 1 to 358
Mg/ha (mean = 70 Mg/ha and Std.== 54 Mg/ha, total = 3.3 million tons dry
weight). Spatially, low AGB occurred in RFS and CC areas while high
AGB occurred in mature hardwood forests.
The AGB class with the highest frequency (16.1%) was 80-100 Mg/ha. The
distribution was skewed toward lower values due to landscape structure. <
3% of the landscape had AGB > 200 Mg/ha.
18
16
Frequency (%)
14
12
10
8
6
4
2
0
1-20 2140
4160
61- 81- 101- 121- 141- 161- 181- 201- >
80 100 120 140 160 180 200 220 220
AGB classes (Mg/ha)
*** Hardwood and mixed forests contained approximately 77% of the total
AGB while PB stored less than 3%;
*** Pine forests comprised about 20% of the total AGB across the landscape.
*** Mean AGB value of red pine (57 Mg/ha) was about 33% higher than that
of jack pine (43 Mg/ha). Clearcuts had the lowest values in terms of both mean
AGB and proportion of total AGB (0.3%). Among the cover types, the AGB
estimates for hardwood had the largest variation (Std = 60 Mg/ha) while the
estimates for jack pine had the smallest variation (Std = 33 Mg/ha).
160
(46.7)
Mean AGB (Mg/ha)
140
(30.1)
120
(15.9)
100
(4.4)
80
(2.6)
60
40
20
(0.3)
0
PB
HW
MIX
CC
JP
RP
The final estimated AGB values compared reasonably with the independent field
observations in the forty validation plots (R2=0.67, p = 0.001).
Estimated AGB (Mg/ha)
200
Estimated = 0.73*Observed + 9.3
R2 = 0.67
160
120
80
40
0
0
40
80
120
160
Observed AGB (Mg/ha)
Others
Hardwood
Pine
200
Discussion
NDVIc proved to be a good predictor for pine forest. The majority of pine forests in the
study area were classified as young and intermediate ages with open canopy structures at
some degrees. Stand age is a strong predictor in estimating AGB of HW forests in the area.
(R2 = 0.67 vs. R2 = 0.56).
Two separate lines are needed for veg. Index if one or more infrared bands involved in the
GLR. Separating improved the AGB predictions (50% more in NIR for HW due to different
canopy structures).
The classification system for age map was defined to be meaningful for fuel loading. CC
were divided into pine-forest clearcuts (CCP) and CCH. Hardwood forests usually retained
more available fuel on the floor, pine much less.
Our models tended to underestimate the AGB at high and overestimate at low end. The
estimated AGB values corresponded well in general with previously studies lower/upper (60600 Mg/ha AGB for mature forests, Crow, 1978). 3 HW sites ranged from 94 to 119 Mg/ha
(ours 93 ± 60 Mg/ha). Our AGB estimates corresponded well with Brown et al. (1999), but
caution must be taken because: 1) Total vs AGB, 2) county (broader classes) vs. 30 m so can’t
be compared. It is likely that high spatial resolution inputs are more suitable for landscape
level analysis.
Potential errors could be from land-cover map, sampling errors, confounding effects of soil
moisture and soil color, and model utilization, 6:4 ratio for mixed forests. Biomass models-errors 3.2% for Red oak, 20% for Sugar maple, mean=12.5%. AGB estimates may be
improved by incorporating tree height using Lidar.
Conclusions
The AGB map may be used to refine the land cover classification by
differentiating young hardwood forests from the mature ones.
Our AGB map can be a useful source for estimating aboveground net primary
production (ANPP). A good relationship exists between AGB estimates and
ANPP before forest stands reach old stage (Euskirchen et al., 2002). Fuel
accumulation in forest ecosystems can be theoretically determined by ANPP
and the decomposition rate (Ryu et al., 2003).
This study provides needed baseline information for landscape level analyses
relating to regional carbon budget (i.e., monitoring changes of carbon pool
over time).
Acknowledgements
This research is supported by the USDA Joint Fire Science
Project. I thank the following individuals for field data
collection and preparation of the manuscript:
John Rademacher, Jiquan Chen, Thomas Crow, Mary Bresee,
James Le Moine, and Soung-Ryoul Ryu