Estimating Biomass Based on Remote Sensing Vegetation Index

Indices based on remote sensing data are a set of useful techniques to be used for estimating biomass. Some examples of indices are as follows: NDVI, or normalized difference vegetation index (Dong et al. 2003; Zheng et al. 2004; Fuentes et al. 2006; Tan et al. 2007); EVI, or enhanced vegetation index (Huete et al. 2002; Nagler et al. 2005; Ostwald and Chen 2006); LAI, or leaf area index (Fassnacht et al. 1997); PAR, or photosynthetically active radiation (Wylie et al. 2007) and FPC, or foliage projected cover and CPC, or canopy projected cover (Rosenqvist et al. 2003). Lu et al. (2002) gives examples of several more vegetation indices used in the Amazon. These indices can be used in combination with other techniques such as field measurements or other environmental data to get carbon content of a land-use system.

As the first illustration of steps involved in estimating biomass, NDVI is used here with forest vegetation. NDVI uses the ratio of the near-infrared and the red spectra (NIR - red/NIR + red) and can be used as a proxy for green leaf area (Myneni et al. 1998) since it signals photosynthesis and is often used as an index to reveal seasonal and/or inter-annual change in vegetation cover. NDVI has been widely used for biomass studies because of its availability and long history of vegetation measurements (Todd et al. 1998; Dong et al. 2003; Seaquist et al. 2003; Zheng et al. 2004; Fuentes et al. 2006; Myeong et al. 2006; Tan et al. 2007; Wylie et al. 2007).

Step 1: Collect inventory data for biomass of the forest

° Depending on spatial scale and accuracy needed, useful data can be collected from a plot size of 50 x 50 m (Lu et al. 2002; Lucas et al. 2006) to that from statistics for an entire province (Dong et al. 2003) ° Geoposition the data to make it correspond to the remote sensing data ° Convert the data into carbon (see Chapter 10 for methods) ° If there is large amount of data, split them into a model-development set and a validation set (Labrecque et al. 2004)

Step 2: Collect NDVI data for the area

° NDVI products do come ready-made from several databases but can also be obtained from data having bands covering the near-infrared and red spectra ° Resolution should, as far as possible, correspond or be smaller than the spatial coverage of the inventory data ° Techniques for stepwise nesting using different resolutions of remote sensing data are described by Muukkonen and Heiskanen (2006) ° One useful method is to collect NDVI data for several years preceding the date of biomass inventory covering the growing season of the forest (Dong et al. 2003). Dong et al. (2003) suggest cumulating the NDVI for the growing season

Step 3: Make sure the remote sensing data are calibrated to reduce satellite or technical noise or atmospheric issues (see Section 14.2.2)

° Most common are cloud effects (Lillesand et al. 2004) ° Data distributors usually have information on how this can be done

Step 4: Make sure the NDVI data are georeferenced so the inventory plots can be identified in the corresponding pixels Step 5: Produce a NDVI model

° Parameters to be considered in the NDVI model are composition of species (Labrecque et al. 2004), age of stand (Zheng et al. 2004), unsupervised classification of land-use classes (Labrecque et al. 2004), latitude if the area is covering a large area (Dong et al. 2003), vegetation texture (Lu et al. 2002) and logarithmic forest biomass (Tan et al. 2007)

Step 6: Find the relationship between carbon data and NDVI

° Correlate the two data sets to develop a statistical relationship. Examples of tests are Pearson's correlation coefficient and stepwise regression analysis (Lu et al. 2002).

° If using two sets of data, one for developing the model and one for validation, a simplified error matrix can be used (see Table 14.3).

Table 14.3 Error matrix table and evaluation of accuracy based on modelled and evaluated sets

Measured

Error matrix

Modelled

Low C

Medium C

High C

Total

Producer's accuracy (%)

Low C

180

50

38

268

180/268 = 67

Medium C

49

210

31

290

210/290 = 72

High C

40

30

200

270

200/270 = 74

Sum

269

290

269

828

Overall accuracy 590/828 = 71

Step 7: Evaluate the relationship

° Coefficient of determination (R2) describes the percent of variation explained in the regression model, with stronger correlation between variables with values close to 1. See Section 14.4 for accuracy of different remote sensing applications

° If using a model development set and an evaluation set, percent accuracy is given as stated in the error matrix (Table 14.3).

NDVI is used here to demonstrate the procedure, but other indices or single spectral bands can be used and tested depending on their availability and requirements of the project. These steps are fundamental, and irrespective of the type of sensor or type of remotely sensed data being used.

Project Management Made Easy

Project Management Made Easy

What you need to know about… Project Management Made Easy! Project management consists of more than just a large building project and can encompass small projects as well. No matter what the size of your project, you need to have some sort of project management. How you manage your project has everything to do with its outcome.

Get My Free Ebook


Post a comment