Development of Mapping Design for Agricultural Features Extracted

from LiDAR Datasets

Nerissa B. Gatdula

, Mylene V. Jerez

, Therese Anne M. Rollan

, Ronalyn P.

Jose

Coleen Dorothy U. Caranza

, Joyce Anne Laurente

and Ariel C. Blanco

1,3

Phil-LiDAR 2 Agricultural Resources Extraction from LiDAR Surveys, Training Center for Applied Geodesy and

Photogrammetry, University of the Philippines, Diliman, Quezon City, Philippines

Phil-LiDAR 1 Data Archiving and Distribution Component, Training Center for Applied Geodesy and Photogrammetry,

University of the Philippines, Diliman, Quezon City, Philippines

Department of Geodetic Engineering, University of the Philippines, Diliman, Quezon City, Philippines

Keywords: Agriculture, LULC, Mapping Design, Post-classification, Geodatabase Schema, LiDAR, Minimum Mapping

Unit, Resource Maps, Models, Automated Workflows.

Abstract: Methods for agricultural feature extraction were developed to produce detailed (crop-level) agricultural land

use/land cover (LULC) maps from high resolution LiDAR datasets. As of February 2017, available LiDAR

data in the Philippines covers 125,200.00 sq.km. or 42.43% of the land area of the Philippines. As part of

product generation, definition of mapping design was considered. This includes algorithm for post-

classification, development of geodatabase schema, and map layouts. Output maps in custom and 1:10,000

scale JPEG maps, shapefiles and KMZ files are distributed to local government units, national government

agencies and other stakeholders for use in planning and other applications. Definition of LULC classes and

types is in accordance with the standard codes of Bureau of Soils and Water Management while 1:10,000 is

based on National Mapping and Resource Information Authority map indexes. Initial classified maps are

maintained in high resolution layers. Detailed objects are refined by determining the Minimum Mapping Unit

(MMU). The use of mapping design has standardized the output agricultural resource maps of implementing

universities involved in the Phil-LiDAR 2 Program. Models and automated workflows were developed to

improve the implementation of the map design.

1 INTRODUCTION

Agricultural land use/land cover (LULC) maps are

produced by utilizing Light Detection and Ranging

(LiDAR) Technology under the Phil-LiDAR 2

Program Project 1, funded by the Philippines’

Department of Science and Technology (DOST) and

monitored/co-managed by the Philippine Council for

Industry, Energy and Emerging Technology Research

and Development. This is to complement programs of

the Department of Agriculture and to provide

accurate, reliable, and detailed agricultural maps at

the crop level. The Project utilizes LiDAR data

acquired by the DREAM/Phil-LiDAR 1 Program,

other remote sensing systems, and field

measurements.

LiDAR, or 3D laser scanning, is an active remote

sensing which measures point cloud data at a rate of

100,000 to 500,000 points per second. These high

accuracy datasets and derivative layers enable

accurate and detailed classification of features on the

ground. As of February 22, 2017, available LiDAR

data covers 42.43 percent (125,200.00 sq.km.) of the

Philippines’ land area. LiDAR datasets, which

include Digital Surface Models (DSM), Digital

Terrain Models (DTM), Orthophotos, and Classified

LAZ, are available for distribution.

Agricultural resource maps produced by the Phil-

LiDAR 2 Program are disseminated to local

government units (LGUs), national government

agencies (NGAs) and other partners in order to

strengthen collaboration. These maps are used for

planning, decision making and development needs.

Combined with hazard maps and other thematic

maps, vulnerabilities of agricultural resources are

assessed.

The Program started in July 2014 with the

University of the Philippines Diliman, through the

276

Gatdula, N., Jerez, M., Rollan, T., Jose, R., Caranza, C., Laurente, J. and Blanco, A.

Development of Mapping Design for Agricultural Features Extracted from LiDAR Datasets.

DOI: 10.5220/0006365902760283

In Proceedings of the 3rd International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2017), pages 276-283

ISBN: 978-989-758-252-3

Training Center for Applied Geodesy and

Photogrammetry, leading fourteen universities in the

nationwide inventory of natural resources. Processes

implemented in relation to agricultural feature

extraction and resource mapping generate various

data layers and outputs in different forms. It was

deemed necessary to develop a map design in the

mapping of agricultural resources in the Philippines

for harmonization and standardization of maps

produced by agencies and universities.

2 RELATED LITERATURE

The development of mapping design intends to

standardize the output agricultural resource maps of

implementing institutions involved in the Program. It

is important that maps, through the map design, are

able to effectively show the results of analysis.

Developer must establish a ‘good design’ and

consider the objective/s and end use of maps. Maps

can represent and communicate the results of the

analysis to wide range of users. Maps interact with

users through the use of map products and how it is

represented (Longley et al, 2005). Theoretically, map

information is communicated to users through map

designs. In practical terms, however, this is not easily

achieved (ESRI, 1996).

In a map design process, considerations are

enumerated by Robinson et al (1995) as follows: (1)

purpose, determines what is to be mapped and how

message is to be represented; (2) reality, defines the

phenomena being mapped by limiting the design of

the map; (3) available data, the specific data type or

format affects the design; (4) map scale, controls how

data should appear; (5) audience, wide range of user

sees information differently; (6) conditions of use,

environment on which map is to be used will define

the design of the map; and (7) technical limits, digital

and printed formats are usually processed and

represented differently.

The standardization of output maps start with the

standardization of procedure for output generation. In

this light, development of algorithms for mapping

design are considered.

2.1 Minimum Mapping Unit

Post-processing of initial classified maps include the

determination of spatial grain or Minimum Mapping

Unit (MMU) (Rutchey et al, 2009). MMU is the

smallest entity size shown in a map. Several factors

are considered in determining the smallest map unit:

(1) data resolution, correponds to the ground

dimension of a single pixel; (2) map scale, refers to

the ratio between the map distance and ground

distance; (3) classification, refers to the specific class

type of an object; (4) print size, corresponds the

physical dimension of the map paper; (5) PPI, the

number of pixels within an inch of printed material;

and (6) viewing distance, considers the distance of a

person looking at a printed map, poster, signage, etc.

on display (Spangrud, 2015).

Identified MMU should provide information

without losing significant spatial information

(Rutchey et al, 2009). In Phil-LiDAR 2 Program,

MMU is applied to digital and printed formats, in

custom-scale and 1:10,000 scale based on NAMRIA

map index.

2.2 Agricultural LULC Schema and

Classes

The Department of Agriculture - Bureau of Soils and

Water Management (DA-BSWM) released in 2009

the standard codes for thematic mapping, including

the classes for LULC maps. Mapping codes are

grouped based on the most extensive dominated land

use, percent dominant land use, most extensive

associated land use, and percent associated land use.

Percent distribution ranges from 50% to 100% for the

dominant and below 5% to above 30% for the

associated land use.

Geodatabase schema stores the spatial attribute

data in table and polygon geometry which is

maintained through structured query language (SQL)

approach, a series of relational functions and

operators. Schema is documented in a data dictionary

wherein objects in a database, tables, fields in the

table, and the relationship between fields and tables

are well-defined. Attribute domains are applied to

enforce the integrity of the dataset. (ESRI, 2016).

Implementation of proposed map design entails

the use of Geographic Information System (GIS),

models and automated workflows in order to

standardize map production across universities.

3 METHODOLOGY

Mapping design for agricultural LULC maps,

including the algorithm for post-classification,

development of geodatabase schema, and map

layouts, are considered. For coastal municipalities

and cities, mangrove and aquaculture classes are

integrated into the agricultural maps.

Models and automation workflows were

developed to improve the implementation of the map

design.

Development of Mapping Design for Agricultural Features Extracted from LiDAR Datasets

277

The general workflow for agricultural map design

is shown in Figure 1, beginning from the initial

classified image to the maps that will handed over to

stakeholders. The classification image, showing the

agricultural features extracted from LiDAR dataset

using object based image analysis (OBIA), is

exported as vector file from eCognition and post-

classified in ArcGIS. Refinement and post-

classification are done by determining the MMU and

smoothening LULC polygons. Template geodatabase

schema are applied to the refined shapefile to produce

a series of LULC maps.

Figure 1: General workflow for the mapping design of

agricultural LULC maps.

Definition of LULC classes and types followed

the standard codes of Bureau of Soils and Water

Management. The standard 1:10,000 scale maps are

based on the National Mapping and Resource

Information Authority map indexes. These were used

so as not to deviate largely from the government’s

mapping standards.

Test data were used to show the performance of

the developed mapping design.

3.1 Refinement/Post-classification

Procedure

Initial LULC shapefiles were exported from

eCognition and added as layers in ArcGIS.

Determination of the MMU depends on the selection

of the significant feature with the smallest area. A

feature, or class, is considered significant when (1) a

class is surrounded by other classes but are near

clusters of the same class; (2) a class is abundant

alongside a river or road; (3) a class serves as

boundaries of a parcel; (4) a class is abundant even

within built-up areas; and other related scenarios.

Figure 2 shows the general workflow for LULC

post-classification and refinement. Implementation of

MMU was carried out through the Elimination and

Smoothing tools in ArcGIS.

Figure 2: Workflow for the refinement/post-classification

of LULC.

Objects are eliminated by determining the

smallest significant unit and merging the insignificant

objects with neighbouring polygons. Non-

agricultural features equal to the identified MMU and

features less than the identified MMU are removed.

For printed maps, MMU can be computed by

applying Equation 1.

  















(1)

As for the smoothing of LULC shapefiles, the

maximum smoothing tolerance is set at 3 meters as

higher tolerance can alter the shape and area of

significant objects. Percent changes in number of

objects and polygonal areas for the initial and post-

processed/refined LULCs were computed.

3.2 Geodatabase Schema

A geodatabase schema was developed to standardize

the output maps of agricultural feature extraction and

LULC mapping of fourteen (14) implementing

universities.

GISTAM 2017 - 3rd International Conference on Geographical Information Systems Theory, Applications and Management

278

3.2.1 LULC Classes

Definition of LULC classes is based on the mapping

standards developed by government agencies. Each

class has assigned code, defined in the table domain

and concatenated to come up with the unique

identifier.

3.2.2 LULC Schema and Domains

Schema refers to the structure of the agricultural

LULC dataset. Table structure consists of class, types

and corresponding unique identifier. Data values are

restricted by creating domains for ‘Classification

Types’, ‘LULC Classes’, ‘LULC Types’, ‘Crops’,

‘Farming System’, ‘Water body Types’, and ‘Road

Types’.

3.2.3 Automation of Schema

Implementation

Automation workflow for schema implementation

was developed in order to generate series of maps

more efficiently. Model is based on Python and

utilizes the ArcGIS geoprocessing toolbox.

Required parameters are the Excel file, where

LULC classes are specified, and template

geodatabase schema, where LULC data is loaded

(See Figure 3).

Figure 3: Automation workflow for schema

implementation.

3.3 Data Integration

Agricultural and aquaculture resource datasets are

combined for coastal areas. There are two cases in

data integration: (1) coastal classification was not

used as thematic layer in agricultural classification;

and (2) coastal classification was used as thematic

layer in agricultural LULC classification.

The integration aims to remove the overlapping

agricultural and coastal classes. The developed

workflow ensured that significant classes in coastal

areas are retained. Insignificant classes, mostly

‘Road’, ‘Bare/Fallow’ and ‘Building’, are reclassified

as auxiliary layer. Result of data integration is

appended to the geodatabase schema (See Figure 4).

Figure 4: Workflow for agricultural and coastal LULC

integration.

3.4 Map Layout

Final agricultural LULC files should be represented

in a map series. Mapping templates in custom-scale

and 1:10,000 were developed to standardize series of

layouts for agricultural LULC maps. Maps in

1:10,000 scale are produced through Data Driven

Pages (DDP) and ArcPy, while, maps in custom-scale

are generated through ArcMap layout page.

Development of Mapping Design for Agricultural Features Extracted from LiDAR Datasets

279

3.4.1 DDP and ArcPy for 1:10,000 Scale

Maps

DDP was used in the creation of series of layout pages

from a single map document. The capabilities of DDP

has been extended using ArcPy, a Python scripting

module that automates the exporting and printing of

maps. The iteration of map production is based on the

selected 1:10,000 indexes with available LULC files

(See Figure 5).

Figure 5: Automation workflow for the mapping of

1:10,000 scale maps.

3.4.2 Template Layout for Custom Scale

Maps

Maps are also represented in custom-scale layout.

Template was produced and distributed to

universities to harmonize map production.

3.5 Production of GIS Files

The final and completed LULC files are handed over

to LGUs, NGAs and other stakeholders. Vector files

are provided in 1:10,000 scale shapefiles and custom-

scale KML/KMZ file formats.

3.5.1 Clip per 1:10,000 Scale Model

Smaller regions of the final LULC files are produced

for easier data handling. For better visualization,

these shapefiles can be loaded in Google Earth Pro.

Figure 6 shows the model for the clipping of 1:10,000

LULC files.

Figure 6: Automation workflow for the clipping of LULC

in 1:10,000 scale grids.

3.5.2 LULC Shapefile to KML/KMZ

Conversion Model

Keyhole Markup Language is an XML-based format

used by Google Earth. Files can be in KML/KMZ file

formats. Model for the conversion of shapefiles to

KML/KMZ are developed (See Figure 7).

Figure 7: Automation workflow for the conversion of

shapefile to KMZ/KML.

GISTAM 2017 - 3rd International Conference on Geographical Information Systems Theory, Applications and Management

280

4 RESULTS AND DISCUSSION

The initial classified shapefile has been refined to

remove unnecessary objects and to set the minimum

spatial grain. In MMU identification, the first

agricultural class with the smallest area is observed.

Table 1 shows the list of agricultural and non-

agricultural classes commonly present in the

classification.

Table 1: Common LULC classes.

Non-agricultural

Class

Agricultural and other

Significant Classes

Water

Crop Fields

(i.e. Rice, Corn, Sugarcane)

Bare/Fallow

Crop Trees

(i.e. Coconut, Banana, Mango)

Building

Mangroves

Road

Shrubland

Grassland

The MMU should be the smallest significant

object in LULC map. Figures 8, 9, 10, and 11 show

the common scenarios where class is considered

important.

Figure 8: Banana object surrounded by other trees and

shrubs but cluster of banana are present nearby (left);

sugarcane within Barren parcel (right).

Figure 9: Abundant banana objects alongside rivers (left);

abundant non-agricultural trees along roads (right).

Figure 10: Coconut trees used as boundaries in a Mango

plantation.

Figure 11: Abundant coconut objects in built-up area (left);

abundant non-agricultural trees in built-up area (right).

The iterative application of the refinement process

has been tested. Tables 2 and 3 show the percent

change in number of objects and area. The percent

change may vary depending on how segmentation in

eCognition was done. Test Area A (Block 8H) has

lower values compared to Test Area B (Block 1C)

which may imply that segmentation is relatively good

and removal of the salt and pepper effect was done

prior to refinement.

Table 2: Percent change in number of objects after

refinement.

LULC Class

Test Area A

Test Area B

Bare/fallow

1.61%

-34.45%

Building

0.84%

-29.33%

Developed

2.09%

-29.66%

Grassland

1.22%

0.44%

Mango

1.16%

0.00%

Non-agri

0.81%

-44.28%

Rice

5.00%

-14.77%

Road

1.85%

-45.49%

Water

3.99%

-60.44%

Table 3: Percent change in area after refinement.

LULC Class

Test Area A

Test Area B

Bare/fallow

0.04%

-0.11%

Building

-0.01%

2.02%

Developed

-0.01%

1.09%

Grassland

0.03%

0.17%

Mango

-0.08%

-0.23%

Non-agri

-0.09%

-0.74%

Rice

0.01%

0.04%

Road

0.10%

1.32%

Water

0.00%

0.06%

Development of Mapping Design for Agricultural Features Extracted from LiDAR Datasets

281

The developed geodatabase schema is applied to

the post-classified/refined LULC shapefile. Table 4

shows the structure of the LULC datasets.

Table 4: Schema of the agricultural LULC map.

Name

Description

CLASSIFICATION

Classification types

RESOURCE_TYPE

Resource map types

ID_CLASS

ID of LULC class

MAIN_CLASS

Main LULC class

OTHER_CLASS1

Other LULC class

OTHER_CLASS2

Other LULC class

CLASS_DESCRIPTION

LULC class

description

ID_TYPE

ID of LULC type

MAIN_TYPE

Main LULC type

OTHER_TYPE1

Other LULC type

OTHER_TYPE2

Other LULC type

TYPE_DESCRIPTION

LULC type description

DATA_SOURCE

Source of dataset (e.g.

LiDAR, Landsat)

DATASET_

ACQUIRED

Acquisition date of

dataset

FARMING_SYSTEM

Farming system

CROP_PLANTING_PER

IOD

Farming period of

crops (e.g. Dec-Feb)

JAN

Crop planted in Jan

FEB

Crop planted in Feb

MAR

Crop planted in Mar

APR

Crop planted in Apr

MAY

Crop planted in May

JUN

Crop planted in Jun

JUL

Crop planted in Jul

AUG

Crop planted in Aug

SEP

Crop planted in Sep

OCT

Crop planted in Oct

NOV

Crop planted in Nov

DEC

Crop planted in Dec

AREA

Area of LULC

REGION

Region of the main

City/Muni

PROVINCE

Province of the main

City/Muni

CITYMUNI

Main City/Muni

BARANGAY

Main Barangay of the

main City/Muni

REMARKS

Remarks

Identification of the main class should be based on

the dominant crop in the area. Dominance is based on

height for intercropping systems and on hectarage for

mixed cropping systems.

The manual implementation of schema requires

doing some of the processes repeatedly. Thus,

automation workflow for the application of schema

were developed. Based on benchmark testing,

approximately 1,500 to 1,800 features were updated

per hour. This translates to 25 to 30 features per

minute. Processing time was observed to be highly

dependent on the size and number of features.

In agricultural and coastal data integration,

insignificant objects found in the coastal area are

reclassified as auxiliary layer (See Figure 12).

Figure 12: Roads observed in fishpond area (left) and

overlapping fishponds and bare objects (right).

Final LULC maps are produced in 1:10,000 scale

and custom-scale JPEG files. Figure 13 shows sample

agricultural and coastal LULC maps in custom-scale

layout.

Figure 13: Custom scale layout of agricultural and coastal

LULC map.

Vector files, in 1:10,000 scale shapefiles and

custom-scale KML/KMZ file formats, are also

generated (See Figures 14 and 15).

GISTAM 2017 - 3rd International Conference on Geographical Information Systems Theory, Applications and Management

282

Figure 14: KML/KMZ maps in custom-scale.

Figure 15: LULC shapefiles in 1:10,000 subset files.

5 SUMMARY AND

CONCLUSIONS

The development of mapping design was considered

in the production of agricultural LULC to ensure the

standardization of maps disseminated to various

stakeholders. The design has been useful in the

management of spatial information and maintenance

of LULC database. Model and scripts using ArcGIS,

ArcPy and Python are utilized in the production of

LULC maps, resulting in faster turn-around from data

to map products.

ACKNOWLEDGEMENTS

The research is made possible through the funding of

Philippines’ Department of Science and Technology

and monitoring of the Philippine Council for

Industry, Energy and Emerging Technology Research

and Development. Technical facilities and support

were provided by the UP College of Engineering, the

UP Department of Geodetic Engineering and the UP

Training Center for Applied Geodesy and

Photogrammetry. The research is also supported by

the Philippines’ Department of Agriculture through

the Information Technology Center for Agriculture

and Fisheries.

The authors acknowledge everyone who has

contributed to the research.

REFERENCES

Buckley, A. (2008). Guidelines for minimum size for text

and symbols on maps. Retrieved from

https://blogs.esri.com/esri/arcgis/2008/01/16/sizeforte

xtandsymbolsonmaps/

Bureau of Soils and Water Management (2009). Manual on

Map Standards & Symbols for Soil & Water GIS.

Carating, R., Manguerra, J., Samalca, I., Pascual, N.,

Albano, F. (2009). Manual on Map Standard &

Symbols for Soil & Water GIS. Bureau of Soils and

Water Management.

Conley, E. P. (2013). 2013 ESRI International User

Conference. Generalization for Multi-scale Mapping.

San Diego, California.

Environmental Systems Research Institute, Inc. (1996).

Introduction to Map Design. Retrieved from

http://healthcybermap.org/HGeo/res/intrcart.pdf

Environmental Systems Research Institute, Inc. (2007).

ArcGIS Desktop Help. Retrieved from

http://webhelp.esri.com/arcgisdesktop/9.2/index.cfm?

TopicName=Simplifying_and_smoothing_features.

Environmental Systems Research Institute, Inc. (2016). The

architecture of a geodatabase. Retrieved from

http://desktop.arcgis.com/en/arcmap/10.3/manage-

data/geodatabases/the-architecture-of-a-

geodatabase.htm

Environmental Systems Research Institute, Inc. (2016). An

overview of attribute domains. Retrieved from

https://pro.arcgis.com/en/pro-

app/help/data/geodatabases/an-overview-of-attribute-

domains.htm

Longley, P., Goodchild, M., Maguire, D., & Rhind, D.

(2005). Geographic Information Systems and Science.

ISBNs: 0-470-87000-1.

Philippe Rigaux, M. S. (2002). Spatial Database with

Application to GIS

Rutchey, K., & Godin, J. (2009). Determining an

appropriate minimum mapping unit in vegetation

mapping for ecosystem restoration: a case study from

the Everglades, USA. Landscape Ecol 24:1351–1362.

Springer Science+Business Media B.V

Spangrud, D. (2015). A Question of Scale, Resolution, and

MMU. Retrieved from https://blogs.esri.com/esri/esri-

insider/2015/05/21/a-question-of-scale-resolution-and-

mmu/ on 21 Jan 2016

Stohlgren, T.J. (2006). Measuring Plant Diversity: Lessons

from the Field. Oxford University Press, USA.

Stocksigns (2012). What size sign should I use? A viewing

distance guide. Retrieved from

http://blog.stocksigns.co.uk/size-sign-viewing-

distance-guide/

T. W. Lillesand and R. W. Kiefer (2004). Remote Sensing

and Image Interpretation. New York: Wiley, p. 157.

USGS Land Cover Institute. NLCD 92 Land Cover Class

Definitions. Retrieved from

http://landcover.usgs.gov/classes.php.

Development of Mapping Design for Agricultural Features Extracted from LiDAR Datasets

283