**4.4 Conceptualizing and generalizing the method**

The estimation approach we developed is a real alternative for assessing AGB at communal and regional scales. It has a relatively low implementation costs when compared to traditional methods (i.e. inventories of land) and can be used to assess and monitor the stock of biomass over time on a country-wide scale. Although it requires field data, it is possible to estimate other areas using the already gained knowledge (in particular, the spatial correlation structure, via the direct and cross variogram models). The idea is based on two hypotheses:


Geostatistical Estimation of Biomass Stock in Chilean Native Forests and Plantations 289

i. Eliminate the presence of clouds and systematic errors. In the case of Landsat ETM +

c. Collection of field data: Field data can be collected especially for AGB estimation or can be available from previous inventories. Data should be expressed in biomass per unit area (i.e. hectare) and associated with a point in geographical formal space (i.e., UTM coordinates). Whatever the case, the field estimations of biomass can be made in any of

d. Data checking and cleaning: To perform the geostatistical analysis, it is necessary to create integrated data tables that incorporate both the biomass data, obtained from field data, plus all available covariates (satellite and topographical). The data should be reviewed, validated, and one should eliminate those displaying an aberrant behavior. e. Geostatistical modeling: Using the data structure obtained in step d, perform

i. Analysis of multivariate spatial correlation (variogram analysis) to obtain experimental direct and cross variograms and to fit variogram models required for interpolation. ii. Obtain local estimations of biomass and levels of uncertainty (estimation error

iii. Masking for estimating administrative units or stands of interest. To get accurate

In the previous paragraphs, one of the most important steps is the collection of field data. These data allow modeling the autocorrelation of biomass in space and its cross-correlation with satellite and topographic variables. When these data are available, the procedure for the estimation should be the one already described. However, sometimes no data are available or they cannot be collected on time or the costs are not reasonable. In this case, one may use known variogram models to make a "blind" estimation using a slightly alternative method: simple cokriging, which assumes that the average biomass is known. We propose that this average can be obtained using the forest growth simulators that are available in

The use of covariates extracted from SRTM elevation model and LANDSAT images allows for estimates with a greater level of detail than those obtained by using only data field. This can be corroborated by comparing the estimates by cokriging with those that would be obtained with univariate kriging. We hypothesize that the set of covariates accounts, through their spatial dependence, for two fundamental aspects to explain existing biomass: a. The quality of the forest site. Topographic variables (altitude, slope and orientation) inform in a synergistic way about the quality of the specific potential growing

b. Vegetation health and vigor are jointly captured by the vegetation indices and Tasseled

many countries, e.g., *Radiata* and *Eucasim* growing models and software in Chile.

i. If data are available at the individual tree, apply allometric equations by species.

error must be corrected SLC-off explicitly.

the following cases:

**5. Conclusions** 

conditions (forest site).

Cap components.

ii. Geometric correction for integration into GIS environment.

ii. If data are aggregated for each plot, apply biomass functions.

geostatistical analysis including the following sub-steps:

estimates, it is important to rule out areas with no vegetation.

variances) via cokriging throughout the study area.

f. Report results and prepare associated maps.

iii. Radiometric correction to reduce topographic and atmospheric errors. iv. Calculation of vegetation indexes and Tasseled Cap components.

Fig. 7. Cokriging results for Pantanillos area. Above: Original estimate (left) and processed (right) biomass (ton/ha). Bottom: Variance of estimation error for AGB (ton2/ha2), original (left) and processed (right).

In order to apply this approach in the estimation of AGB in any area the following steps should be done:


In the previous paragraphs, one of the most important steps is the collection of field data. These data allow modeling the autocorrelation of biomass in space and its cross-correlation with satellite and topographic variables. When these data are available, the procedure for the estimation should be the one already described. However, sometimes no data are available or they cannot be collected on time or the costs are not reasonable. In this case, one may use known variogram models to make a "blind" estimation using a slightly alternative method: simple cokriging, which assumes that the average biomass is known. We propose that this average can be obtained using the forest growth simulators that are available in many countries, e.g., *Radiata* and *Eucasim* growing models and software in Chile.
