ALGORITHMS FOR ESTIMATING FOREST INVENTORY

PARAMETERS FROM DATA ACQUIRED BY REMOTE

SENSING METHODS

Ingus Smits and Salvis Dagis

Faculty of Information Technologies, Latvia University of Agriculture, Liela Street 2, Jelgava, Latvia

Keywords: Image processing, data mining, segmentation, forest inventory, LiDAR, aero photography.

Abstract: Two technologies- LiDAR (Light Detection and Ranging) and aero photography- have a huge potential in

forest taxation, which is a process of gathering different parameters of specific region. Both technologies

can be used for finding different parameters of interest, such as number of trees, tree height and other. This

paper presents the results of usage analysis for LiDAR and aero photography, and describes their

possibilities. Also it contains an analysis of tree identifying algorithms and describes ways of their usage in

different processes.

1 INTRODUCTION

Taxation process is very important for performing

forest management planning activities, especially

because its performance frequency and prerequisites

are defined by law. Unfortunately, methods that are

available today require a lot of time and funds.

To reduce the usage of resources, new methods must

be developed or some existing ones must be adopted

from different fields. Two appropriate candidates,

which already have application for data collection in

other fields, are described in this article. Those

technologies are aero photography and Light

Detection and Ranging (LiDAR)

(Lim et al. 2003).

Using them to acquire taxation parameters is not a

simple task which options must be analyzed.

In both cases the main task is not to deal with the

problems of technical type (technology is already

being used in other fields and is also adaptable for

this one). The main problem is to interpret the result

data appropriately – to find out taxation parameters

of different regions using information which is

gathered by aero photography or LiDAR

technologies. In this article use cases and potentials

of both methods are analyzed and algorithms for tree

identification are described.

2 PROCESS OF TAXATION

The aim of forest taxation and accounting

corresponding data is to gather geodetic materials

and overall information about woods that is under

jurisdiction of some managers or owners. Data about

territories is collected using geodetic methods and

aero photography materials (

Sile 2006). There are

two types of taxation: precise (every tree is

measured using special tools – a very expensive and

time consuming process) and group evaluation (in

defined region average parameters are measured for

each kind of tree)

(Dagis 2006). In practice precise

taxation is used only to calculate value of property

before performing the main timber cutting process.

Different situation is with group evaluation which

must be performed once in every ten years as it is

defined by law. This taxation type is used mostly for

informative reasons - to find out exact situation on

given property and to decide what economical

activities must be performed there (Dagis, 2007).

Precise evolution of forest region from group data is

almost impossible. Methods that allow to execute

this task exist (Ozolins, 2004; Ozolins, 2006; Dagis

et al. 2006), but they only give approximate results

(Smits, Dagis, 2007). One of the biggest needs for

forest owners and some republic services is to

improve taxation process in the way that it becomes

faster and cheaper so the management planning and

all economical activities would consume less

resources (Oss, 2006).

583

Smits I. and Dagis S. (2008).

ALGORITHMS FOR ESTIMATING FOREST INVENTORY PARAMETERS FROM DATA ACQUIRED BY REMOTE SENSING METHODS.

In Proceedings of the Tenth International Conference on Enterprise Information Systems - AIDSS, pages 583-587

DOI: 10.5220/0001720005830587

 SciTePress

3 PROCESS OF MANAGEMENT

Planning is the most complicated task that is

performed in enterprise and it requires very precise

data.

Information that is used for planning must describe

not only existing situation, but also be with high

accuracy and degree of detailed elaboration to serve

as a basis for forecasting evolution possibilities in

defined terms. Every company must have this kind

of data to solve everyday tasks and all their

decisions need to be data based. This statement

applies to all kind of enterprises and especially in the

field of forestry where data acquisition is called

taxation. Of course company can have a lot of

different information sources which must be taken

into account, but taxation data that describes real

time quantitative parameters of forest is the most

important one.

Figure 1: Data acquisition in management process.

Figure 1 shows the process of planning economical

activities that must be performed in forest. The main

part of the diagram is a loop in which potential result

is evaluated by considering how it will change after

certain economical tasks are performed. These

activities are executed as many times as needed to

find some politically correct result. Before the

beginning of searching process real taxation are

preformed and data that describes the existing

situation are gathered. Planning tasks are being done

by using this information and some peripheral data -

normative documents, laws, and descriptions about

availability of resources.

4 REMOTE SENSING METHODS

Data gathering is the main task of taxation process

and to complete it specially educated people must go

to forest and find out information that describes

some regions. Performance procedure and time

schedules of this process are described by law.

Unfortunately, traditional way of doing it is

expensive and time consuming – a specialist with

adequate knowledge to create sampling plot and

measure its characteristics must visit every territory

of interest. Costs of this work scenario can be found

as sum of: specialist education, time and travel price

and expenses of measuring process. Traditional

taxation is not only ineffective in terms of time, but

it also gives only very local information that must be

generalized by specific mathematical methods.

These calculations are related with certain growth of

data error which can be statistically evaluated.

Methods that are not so local and describe much

larger territories consuming less time (and also

reducing the main source of mistakes – human

factor) must be created. In the next chapters of this

article two technologies (aero photography and

LiDAR) for data acquisition that can describe very

large regions will be overlooked. Both methods have

big potential and are being examined by specialists

in different countries and science fields. The task

that should be solved in nearest times to enable full

scale usage of these methods is data processing to

convert them into form which is understandable to

information systems. Transforming information that

is gathered either by aero photography or LiDAR is

not a trivial task, because complicated algorithms

that help to find informative relationships in data are

needed. Also to use both of mentioned technologies

and to find out specific characteristics of different

regions precise GIS coordinates of each point must

be available.

4.1 Use Cases of Aero Photography

Aero photography is an image that is taken from

airplane flying at certain height. It allows acquiring

data that describes earth surface and all objects that

is located on it. The quality of pictures and possible

use cases are closely related to the height and

stability of a flight. An image acquiring by using

pilotless technologies has become very popular in

the most recent periods of time, because of its small

expense and simplicity of usage.

An aero photograph that describes different

territories has been used for many years and the

ICEIS 2008 - International Conference on Enterprise Information Systems

584

main field of interest for it was cartography. Many

well known geographic information systems are

equipped with algorithms that allow creating a photo

layer to display them on a map.

The most important feature of such photography is

that the objects shown on it are attached to their

location by coordinates of the real world. With the

improvement of technologies and rise of calculation

powers of computer systems new usage possibilities

for these images have appeared, for example

evaluation of specific objects.

Using algorithms for object recognition in images

can be performed in three modes – automatic, half

automatics and manual. Each of the mentioned

modes has well known precondition – photo quality

which can be described by many components such

as color, contrast, graininess, the amount of objects

and others. The relationship between quality and

automation level can be stated – for higher quality

higher automation level can be used.

Acquiring useful image is not a trivial task, because

many factors such as weather, time of the year,

quality of equipment must be taken into account.

Also height of flight is very important and can be

altered depending on the goals of photography. For

example, if the main task is to get the density of

trees in some region then the height can be

comparatively big, but if it is necessary to find the

size of leafage or even a kind of a tree, the height

must be small. In the first situation the main benefit

can be found in the fact that for describing some

region a smaller number of photos can be taken then

in the second situation. A specific task can be solved

only when all needed data is acquired for the

territory of interest.

Aero photography has many use cases, but for

taxation the most important ones are:

1. Finding the number and coordinates of a

tree;

2. Finding the size of a leafage ;

3. Finding the kind of a tree;

4. Finding the borderlines of a territory;

5. Finding forest vistas and roads;

6. Evaluating territories gutted by fire;

7. Evaluating windfalls.

Image processing algorithms are needed for solving

all of the mentioned tasks. The first, the fourth, and

the fifth tasks need photos with the smallest

resolution. All other tasks need qualitative pictures

which contain plenty of data to use methods that can

separate tree leafage and measure its parameters. For

all of the pictures widely known algorithms or

simple each pixel overlooking loops can be used. In

this situation a very popular group of segmentation

an algorithm that needs to know a number of clusters

cannot be utilized because it is the parameter that the

system is looking for.

If the system finds specific segments by using these

methods, then by knowing the height of a flight and

the angle of a camera, the size of leafage can be

calculated.

For tree kind determination it is necessary to

recognize the structure of an object, were at a certain

scale contours of a leaf which is taken from the

image and searched in the previously defined

collection. Another way of solving this task is by

using colored recognition (each tree kind has its

specific color). The main drawback of this method is

that the color depends on weather and photo filters

used in the picture making. It means that automatic

use of this method is almost impossible.

4.2 Tree Identification using Aero

Photography

Crating automatic methods is a very difficult task

and the first step in it is to understand half automatic

or even manual solutions. We will describe an

algorithm for identifying a tree that works in a half

automatic mode. For this method images attached to

GIS coordinates are needed and also a full photo

cannot be used, but only a part of it where the angle

of photography and surface is close to 90

, because

in other case it is a side-view. So the images must

overlap and cover some part of the same territory.

Tops of a tree on the images usually can be well

separated from the background and their color

depends on the time of the year and weather.

Therefore to realize half automatic way of

recognition it is necessary to select a few pixels from

tree tops (2 to 5). Every shown point gives us

information of the possible color and by adding

some dispersion to it (recommended 5-15%) we

define a pattern to look for in the rest of a picture.

Dispersion and location of points are parameters that

user can change depending on results. Minimal

min

) and maximal (R

max

) radius is given in

numbers and by using them in combination with tree

top color dispersion, searching for a particular tree

can be preformed.

Algorithm for finding tree center works with one

correction (Fig. 2) – searching is performed on X

and Y scales by using colors. First of all, tree top

start point x

and end point x

have to be found, then

an average value x

is calculated x

= (x

)/2. From

the point x

on Y scale minimal y

and maximal y

values are found. In the same way as x

the value of

is calculated y

= (y

)/2.

ALGORITHMS FOR ESTIMATING FOREST INVENTORY PARAMETERS FROM DATA ACQUIRED BY REMOTE

SENSING METHODS

585

The center of a tree top is found by correcting x

values corresponding to the newly found y

value.

Using center coordinates every point that is located

at a distance of radius R

tree

=min(R

max

lefage

) is

considered to belong to a single tree. After writing

necessary data into result collection newly found

tree top is removed from the searching region. After

that new searching iterations are being performed

while group of points where R

lefage

min

can be

found.

Figure 2: Activities of finding tree top center in aero

photography.

By utilizing described method every tree center in a

photo is found. Using GIS coordinates’ a real

location of a tree on 2D plane can be calculated and

later utilized for taxation data concretization with

different methods. From discovered points the

number of trees per hectare can be found. And also

by using leafage diameter and tree height the

relationships between an approximate diameter and

the density of a tree trunk can be calculated. For the

precise taxation these results must be improved with

a data from different methods.

Figure 3: Photo before and after using tree recognition

algorithm.

Identifying a tree on a photography using automatic

mode is a very complicated task. It has a lot of

preconditions for data gathering technologies and

processing algorithms. Two tasks that must be

solved to realize fully automatic tree recognition are:

1. Acquisition of qualitative picture where all

necessary attributive data is saved together with

the image (for example previously mentioned

tree top point colors);

2. Use of universal algorithms that regardless on

different picture parameters can find number of

segments and filter needless data.

5 USAGE OF LiDAR IN

TAXATION

LiDAR is a new theology that is being used to

acquire information about earth relief and objects

located on it in three dimensional plans

(Lim et al.

2003)

. It is done by using special laser and light

registering device. The main principle of this

method is the same as in the case of radar – the time

in which emitted signal returns to the source is being

measured. The only difference from simple radar is

that instead of radio signal laser light is used.

LiDAR technology is used in many fields and some

of them are: archeology, geography, geology and

remote sensing.

Figure 4: LiDAR data acquisition.

LiDAR measurements are performed by using

airplane that is equipped with a device of global

positioning and laser scanner device that during a

flight perpendicularly to its direction (Fig. 4)

measure light reflection points. Mistakes of data that

rise depending on scanner angle and flight stability

are eliminated in the process of measurement.

By visualization of gathered points three

dimensional model of a tree (or any other object),

that describes its characteristics can be created.

These models can be analyzed with aim of finding

some special parameters of an individual tree or

even of all sampling plot. Usage of this method in

taxation process still is an object of research and it is

necessary to clarify its qualitative and quantitative

parameters – costs, preciseness of measurements and

how it works in different kinds of forests.

ICEIS 2008 - International Conference on Enterprise Information Systems

586

Tree counting using LiDAR data can be done

similarly to aero photo. the difference is that in

photography the color was analyzed for each pixel,

but in LiDAR data 3D coordinates of each point

must be examined.

Figure 5: Tree identification by using LiDAR data.

For identifying a single tree or finding an overall

number simple algorithm that analyze picture by

dividing it into small regions can be used. The

highest point in each field is calculated and it is

assumed to be the top point of the highest tree (Fig.

5). Around this point using defined radius

segmentation can be performed to define all points

that belong only to a single tree. Afterwards all

located segments are removed from data collection

and new search for the highest point started. Tests

show that this simple algorithm works well in a

forest with not very high density and only one floor

of trees.

6 CONCLUSIONS

Aero photography and LiDAR data can be used for

finding a number of trees and its coordinates.

Identifying a kind of a tree is a task that doesn’t have

one universal solution and research of its

possibilities must be continued.

The main benefit from using LiDAR technology is

that from its data tree height can easy be found,

because every point is described in three

dimensions.

The biggest drawback for both methods is that there

is no direct way of finding the most important

taxation parameter – diameter of a tree.

Using both of described technologies together in the

same region can raise their efficiency.

Data processing can be performed in three modes –

automatic, half automatic or manual. For realization

of automatic mode it is necessary to create not only

adequate data processing algorithms, but also it

requires certain methodology of data gathering.

ACKNOWLEDGEMENTS

Authors wish to thank EU funded project “Support

for realization of doctoral program and after doctoral

researches” and “Forest development fund” for their

support of this synthesis.

REFERENCES

Dagis S., Arhipovs S., 2006, “The Static Model of Latvian

Forest Management Planning and Capital Value

Estimating”, Proceedings, “8th International

Conference on Enterprise Information Systems”,

Cyprus, Paphos.

Dagis S., Arhipovs S., Dubrovskis D., 2006, “The Growth

of Trees Motion Mathematical Models and Their

Adaptation the Latvia Circumstances”, Lithuania,

Kauno.

Dagis S., 2007, „Information System Requirement

Analysis And Specification in Forest Management

Planning Process”, 9th International Conference on

Enterprise Information Systems, Funchal, Madeira –

Portugal.

Ozolins R. 2004, “Tree Trunk Distribution by the

Diameter Class”. Proceedings of the 5 international

conference – Latvian academy of pedagogy.

Ozolins R. 2002, “Forest Stand Assortment Structure

Analysis Using Mathematical Modelling”. Forest

structure and growth – Tartu, – (Forestry studies

XXXVII) p. 33-42.

Inita Sile, Sergejs Arhipovs, 2006 ,“The Development of

The Precedent Model For the Latvia Forest

Management Planning Processes”, Proceedings, “8th

International Conference on Enterprise Information

Systems”, Cyprus, Paphos.

Janis Oss, 2006, "The Model of Wood Resource Flow",

Proceedings, "The 2006 European Simulation and

Modeling conference", Toulouse, France.

Lim, K., Treitz, P., Wulder, M., St-Onge, B. and Flood,

M. 2003. LiDAR remote sensing of forest structure.

Prog. Phys. Geog. 27:88-106.

Smits Ingus, Arhipovs Sergejs, 2006, “The Usage of Palm

Top Computer in Forest Inventory”. International

scientific conference Biometrics and Information

Technologies in Agriculture: Research and

Development, Kaunas, Lithuania.

Smits Ingus, Salvis Dagis, 2007, “Tasks and Data

Precision Problem in Forest Management Planning

Information System”,"The 2007 European Simulation

and Modeling conference", Porto, Portugal.

ALGORITHMS FOR ESTIMATING FOREST INVENTORY PARAMETERS FROM DATA ACQUIRED BY REMOTE

SENSING METHODS

587