Introduction

Due to rapid industrialization and increases in population pressure on forest-basedresources, the forest cover is shrinking at a rapid pace. Therefore, there’s anurgent need to periodically monitor these resources and develop appropriatelocation-specific conservation plans.

All of this wouldn’t be possible without a detailed resource inventory. But in mostunderdeveloped and developing countries, where large amounts of forestresources are available, there’s a dearth of detailed forest-inventory data andquality location-specific maps.

Rapid Inventory

Conventional field-based techniques of forest-resource inventory are typically based onsystematic random sampling, where an area is divided into square grids of equalsize. A survey team visits systematically selected random grids in the fieldand collects sample plots based information about the dominant type, speciescomposition, number of trees, height, diameter at breast height (DBH), etc.After the data from sample plots are collected, the information is representedon a respective grid.

The problem with such generic quantification is that it’s difficult to monitor, andall grids can’t be surveyed due to inaccessibility and geopolitical situations.Also, there’s no mechanism to validate field-based information collected orcompiled during the inventory process.

Planning

The most important aspect of resource-inventory planning is having a clearunderstanding of the information required and the type of satellite data thatcan provide the most desirable results. Presently, multiple satellite-imagesensors are available in the public domain, and they can acquire images ofEarth ranging from very coarse to very high spatial resolutions.

After satellite image selection is finalized and preliminary classification is completed,it’s important to plan the field work so training sets based on therepresentative signatures on satellite images can be identified and validatedin the field.

Pre-Processing Images

Pre-processing of satellite images is one of the most important aspects of data creation. Because the raw satellite images don’t relate to actual ground locations,they’re geo-referenced to real-world coordinates.

Geo-referencingrequires details of ground coordinates and their location on satellite images.Information about the geographical coordinates for geo-referencing satelliteimages can be collected from geo-referenced topographical maps.

In areas where either topographical maps are relatively old or on a smaller scale, it isvery important to collect GPS based geographical coordinates for easilyidentifiable control points visible on ground and satellite images.

This technique can be effectively used in areas where there’s a lack of existinginformation, as particularly seen in most of the developing countries.

Data Collection

The data collection activity should be planned to minimize cost, sowell-distributed heterogeneous patches are visited in the field. Forsatellite-based classification and quantification of stock and biomass,training sites should be identified on the basis of different forest types, anddensity maps should be prepared using satellite data.

The basic principal behind such stratification is to identify homogenous stratumwithin heterogeneous patches. Each stratum represents a unique set of foresttype.

In addition, proportional areas under each stratum can be assessed to estimate thenumber of sample plots based on probability proportional to area. Thestratification can be further refined using cost and time factors, allowableerror, and sampling intensity to estimate the number of sample plots for eachforest type.

These sites act as a base for “training” software with information collected from thefield during classification and analysis. The sample-based field informationabout forest type, species composition, and number of trees, height and DBH needto be collected for the training sites through ground measurement. For eachtraining site, 30- by 30-meter sample plots could be identified for each samplelocation in the field.

Data Interpretation and Analysis

After image pre-processing is completed, they’re interpreted, and pixels are groupedinto different classes based on variations in spectral properties and/or usingdifferent vegetation indices. Previously, satellite images were interpretedthrough visual-interpretation techniques in which visually perceptibledifferences in pixel clusters are grouped manually over hardcopy satelliteimages or through onscreen digitization.

With the advent of digital image-processing techniques, the classification processhas become much faster and now creates better-quality output, as minordifferences in spectral signatures that aren’t perceptible to human eyes can bedetected by computers.

In supervised classification techniques, human inputs are provided to the softwareas training sets. Numerical information in all spectral bands becomes a basisto “train” computers to recognize spectrally similar areas for each class. Inunsupervised classification, software groups the pixels into classes asprovided by users based on the spectral properties through clusteringalgorithms. A combination of supervised and un-supervised classification isknown as Hybrid classification where in iterative mode the output is generated.

After completing the preliminary classification, training sets need to be identifiedfrom well-distributed representative clusters of different vegetation classesusing stratified random-sampling techniques. A team of field experts isrequired to visit the training sites on the ground to collect GPS coordinatesand information related to forest type. The location based information thus utilizedfor further refinement of the classification results.

Forest type and stock are the most important inputs for assessing carbonsequestration. The field-based stock assessment for the sample sites shall be overlaidon forest-type and density maps and extrapolated to the entire forest to derivea forest-stock map.

Data Verification

For any GIS and remote-sensing study, it’s important to have stringent data qualitychecks at every stage of data collection, creation and analysis. It’s estimatedthat roughly 5 percent of the total effort should be spent on quality checks.Considering the data gaps in the study area, additional efforts need to beapplied for such quality checks.

Image geo-referencing is one of the biggest challenges during the data creation stage in developing countries due to a lack of large-scale topographical maps. The Landsat mosaic images and ground-based GPS points could be utilized as reference data for geo-referencing satellite images. If imaging geo-references aren’t perfect, an image point can be displaced from its true ground point, leading to misinterpretation and the wrong correlation of ground-basedinformation with respective signatures on satellite images.

After a final classification map is produced, an accuracy assessment should be made to measure a map’s reliability. Generally, several GPS-based random points orareas are selected on a map, and their class assignments are checked. Thenumber of points checked is determined by analysts, but more points yield a more robust conclusion.

Conclusion

To effectively use available remote-sensing and GIS tools, it’s always better to analyze the ultimate objectives and suitable options that provide desired results. Detailed planning about the effective use of input data as well asforming robust methodologies and effective data quality checks go a long way inmaking such projects successful, and they’ll help create logical outputs for planners.