Remote Sensing Rough Draft.docxknightlab.org/rscc/projects/F14/cfc_map.docx · Web viewBy associating CFI plots to the stands, and the stands to cover type helps in identification

Updated Map of Cloquet Forestry CenterFor Continuous Forest Inventory

Anna HyytinenMaryada Shrestha

FNRM 5262December 8. 2014

Submitted to: Joe KnightUniversity of Minnesota

Introduction

The University of Minnesota’s Cloquet Forestry Center (CFC) was founded in 1909 with the purpose of serving as an experimental and demonstration forest for Minnesota students, researchers and foresters alike. In the fall of 1959 the Cloquet Forestry Center (CFC) Continuous Forest Inventory (CFI) began and has been conducted every five to ten years ever since. It was established for the purposes of 1) demonstrating systematic random sampling with random starts, 2) providing up-to-date inventory statistics for management purposes, 3) examining techniques for surveying mortality and related forest characteristics and 4) comparing the precision of estimates obtained from fixed and variable radius plots. This inventory is the most extensive of its kind in the Lake States (Reber & Ek, 1982). Through the CFC CFI we are able to gain long-term information about forest growth and change in Minnesota.

The Continuous Forest Inventory encompasses the CFC’s contiguous block of land in Carlton County approximates 3,506 acres with the legal description of all sections 29 and 32 in T49N R17W, section 36 of T49N R18W, and the N ½ of the NE ¼ of section 6 of T48N R17W (U of MN: 2004 Real Estate Inventory Report). Within this area 402 plots were established through a systematic sample with four random starts. At each plot center tree-level measurements (DBH, height, distance from plot center etc.) and observations (tree health, damage etc.) are conducted (Dieser & Ek, 2011). The predominant tree species found are red pine, jack pine, aspen and paper birch (Reber & Ek, 1982).

Continuous Forest Inventory is an invaluable tool for assessing forest management schemes over time. By taking field measurements through CFI, managers are able to see the direct correlation between management practices and forest composition, health etc. Under quality adaptive management, CFI can inform changes that will help maximize desired forest qualities into the future. Additionally, CFI allows researchers to travel back through time and see not only the effects of forest management, but also of climate and broad human activity. By relating CFI measurements to trends in population growth, weather and other factors, models can be created to predict how forests will fair in the future.

The objective of our study was to develop an updated map of Cloquet Forestry Center using eCognition for the enhancement and understanding of the Continuous Forest Inventory and aid in field measurement acquisition. The most recent set of reference maps used for field work on the CFC CFI were created in 2011. These maps are very basic and rely on stand boundaries and field measurements for the identification of cover type. By utilizing eCognition we are able to classify the CFC’s contiguous block based on satellite imagery. This allows for more precise identification of cover type. Satellite allows us to easily detect small difference in height, color, brightness etc. that lend themselves to large differences on the ground. Being able to better and more easily predict what field crews will encounter when collecting field data will allow for more efficient collection of data for the CFI.

2

Data Source

Detailed satellite imagery of Cloquet Forestry Center were obtained from several sources. ● Light Detection and Ranging (LiDAR) 2011

○ Minnesota Geospatial Information Office● National Agricultural Imagery Program (NAIP) 2013

○ National Association of Regional Councils’ Geospatial Data Gateway○ 1 meter resolution○ 4 band; Red,Green, Blue, Near Infrared (USDA, 2013)

● Color InfraRed (CIR)○ MnGEO NAIP Imagery WebService

● CFC Stand boundary layer○ 2005 vegetation mapping effort at CFC

● Roads layer and xy-coordinate of CFI plots

All data were projected using Transverse Mercator, NAD 1983, Zone 15N.

Methods

Color Infrared (CIR) was used as the base imagery for the cover type classification. It was also used as a reference source to verify that the rule set in eCognition produced the correct classifications. With plain sight it is easy to see that the red coloration is indicative of deciduous forest and green of coniferous. Bare soil, buildings and water can also be seen in the 1x1 meter CIR image.

Figure 1: Color Infrared (CIR) imagery of Cloquet Forestry Center

The boundary layer was clipped to LiDAR and NAIP imagery - using ArcMap - to extract CFC from Carlton County. The clipped imagery were then used in eCognition software to make a map by classifying the forest and setting the rules for classification.

3

Figure 3: NAIP Imagery of Cloquet Forestry Center

NAIP imagery has a near-infrared band which helps to determine the water bodies, vegetation vs. non-vegetation, young vs. old stand etc. NAIP 2013 imagery was used to derive the Normalized Difference Vegetation Index (NDVI) which was used to identify vegetated and non-vegetated areas. NDVI was calculated using the equation (NIR-Red)/(NIR+Red) in eCognition. NIR from NAIP imagery was used to differentiate evergreen vs. deciduous.

Figure 3: nDSM of Cloquet Forestry Center

4

The basic function of LiDAR is to determine the height of objects, which can be used as a way to differentiate features from one another. Applications of LiDAR beyond calculating the absolute heights of features include estimating volume, biomass etc. In this project, LiDAR was used to get the nDSM values for the objects in the image.

LiDAR point cloud was converted from .laz to .las and ArcMap 10.2 was used to create the Digital Elevation Model (DEM) and the Digital Surface Model (DSM). Finally, ‘Minus’ function was used to create the normalized Digital Surface Model (nDSM) from LiDAR data. nDSM helped to distinguish between tall and short objects.

Description of Rule Set

Classes:

Object Based Image Analysis was done to create an updated map of Cloquet Forestry Center.The forest was first classified into Tall, Short, Vegetation, Low/Non-Vegetation, Grass, Buildings, Roads/Parking lots, Bare Soil, Water/Shadows and Evergreen and Deciduous forest.

Process:

Multi-Resolution segmentation was done to do the object based image analysis. NDVI was calculated using the formula (NIR-Red)/(NIR+Red) to find the vegetative index.The forest was classified as Tall and Short.

Clicking around the image randomly and referring to the CIR image, nDSM greater than 3 was chosen to classify tall trees and buildings as level: Tall.

All the objects in Tall class were then differentiated into Vegetation and Buildings. Since vegetation has higher NDVI values than non-vegetation, and because NDVI values range from -1 to +1, NDVI>=-0.32 were referred as vegetation. Buildings have lower NDVI value and more blue color than vegetation, thus NDVI < -0.32 and Blue> = 135 were referred as buildings. Within vegetation, as deciduous trees reflect more NIR than coniferous, NIR value was used to set a threshold by which we were able to differentiate between tree type. Of the vegetation, that with a NIR > 150 was classified as Deciduous and Mean NIR <= 150 as Evergreen.

From class Short, Low Vegetation, Grass, Water/Shadow and Bare Soil were separated. Since water and shadow both appear dark in the NAIP imagery with IR band, water and shadows were classified as a single group. Brightness of water will be lower in NIR image and will have higher blue value, so brightness level and blue color were used to separate the stream and other water bodies. Again, within class ‘Short’, NDVI was used to differentiate low/non-vegetation and vegetation. From low/non-vegetation, nDSM was used to differentiate grasses from shrubs (low vegetation). Non-vegetation has much lower NDVI value than vegetation, so from class ‘low/non-vegetation’, NDVI <=-0.01 was used to separate non-vegetation from low vegetation as class ‘bare soil’. Since non-vegetation was called ‘bare soil’ to make it easier for us to process, roads/parking lots were separated from class ‘bare soil’ using NIR and brightness. Roads reflect lower NIR and have higher brightness values than bare soil.

5

Clean-up:Finally clean-ups were done merging different regions.

Results

Figure 4: Cloquet Land Classification Map

Figure 4 shows the land classification map that was produced using eCognition software. This map shows the evergreen and deciduous forest with possible clear cut and some low vegetations and some grasses or marshy areas that can be seen near the water bodies.

Light green color in the map represents evergreen forests and dark green color represents deciduous. We can see some water streams and grasses/marshes in some part of CFC. Low vegetation represents plants bigger than grass possibly shrubs. The bare soil which can be clear cut can be seen in some parts. Buildings are red which can be seen in upper part of the map. Parking lots and bare soil can bee seen by the side of buildings. The yellow lines represent roads.

This classification image can be used for both forest management and research purposes. In addition, the map can provide students a basis of understanding of the kind of forests types in CFC. It can also act as a reference maps for field work purposes.

6

Figure 5:Cloquet Forestry Center Stand MapStand ID legend Appendix B (Loeffelholz & Zimmerman, 2005)

Figure 5 is a map that depicts the basic cover types for all the stand in the Cloquet Forestry Center. Classification shows the main covertype (evergreen / deciduous) for the forested area in CFC. Overlayed are the stand boundaries with their associated stand ID number and roads. By associating CFI plots to the stands, and the stands to cover type helps in identification of the main covertype (evergreen or deciduous) for each stand to provides more precise understanding of cover type around cloquet.

In addition, it can help to figure out the stand with clear-cut plots, stands close to the roads and stand in marshy areas. With this map, it will be much easier for field measurement crews to anticipate the conditions/terrain they will encounter while doing the research.

7

Discussion

The maps we have produced provide inventory field-workers, students, researchers and managers of the Cloquet Forestry Center with an updated overview of what they can find within the CFC’s contiguous block. Understanding where deciduous and coniferous trees live, as well as small wet areas can inform field-workers of the conditions into which they will have to go to take measurements for inventory.

The eCognition software makes classification based on empirical, remotely sensed data quite easy - in theory. Rule sets are defined based on very specific values, however, and where there are discrepancies in sensor data, there are also discrepancies in classification. For example, roads were very difficult to isolate during classification. Because of their narrowness and close proximity to trees, they are often obscured in the satellite imagery. There is a way in eCognition to classify roads using a vector layer, like the road layer overlaid in the final maps. We had difficulty running the rule set for thematic road layers, as the use of vector roads as a thematic layer was beyond this course.

In addition, the shadows of the trees were also hindering the classification of road feature and road appeared as water in some places. Water and shadow both appear dark in the infrared band, which made it difficult to distinguish between the two features. A very little part of the image didn’t fall on either vegetation or low/non-vegetation and remained unclassified, those areas were near the side of the buildings and were named as the shadow of buildings. Overall, the map could be made better provided more knowledge of eCognition software, more resources and more time.

The ability to obtain the most up-to-date imagery with which to perform an updated classification of the CFC proved to be a challenge as well. The CFC boundary map that was used in this report does not contain new stands that were added to the CFI contiguous block in 2011. The maps could be improved upon by using the most recent extent of the CFC to define the CFI area.

The maps produced in this report are meant to be used as tools for Continuous Forest Inventory data collection for forest researchers. They do not attempt to replace the immense value of ground measurement at an individual tree level. It could be very useful to utilize eCognition to produce a cover-type map each year that the CFI is conducted. These maps are able to show large-scale change over time, which is one of the fundamental goals of continuous forest inventory. With more experience using eCognition software, more precise maps can be made to show change in more detail.

8

Division of Work

● Obtaining Imagery: Maryada and Anna● Processing Imagery: Maryada● Introduction: Anna● Methods: Maryada and Anna● Data: Anna and Maryada● Results: Anna and Maryada● Discussion: Anna and Maryada

References

Dieser, P. J., & Ek, A. R. (2011). Cloquet Forest Center Continuous Forest Inventory from 2000: Analysis and Integration with the Historical Database. Staff Paper Series No. 214. Department of Forest Resources, CFANS, University of Minnesota. St. Paul, MN.

Reber, C. A., & Ek, A. R. (1982). Cloquet Forestry Center Permanent Plot Records 1959-1976: Status Report and Plans for Remeasurement. Staff Paper Series No. 33. College of Forestry, University of Minnesota. St. Paul, MN.

United States Department of Agriculture (USDA). Four Band Digital Imagery. (2013). United States Department of Agriculture. Washington D.C.

9

Appendix

A. Rule Set Documentation:Classes:

BareSoilBuildingsDeciduousEvergreenGrassLow/NonVegetationRoads/ParkinglotsShortTallVegetationWater/Shadows

Process: Main:Segmentation

multiresolution segmentation: 30 [shape:0.1 compct.:0.5] creating 'New Level'Create_NDVI

layer arithmetics: layer arithmetics (val "(NIR-Red)\(NIR+Red)", layer NDVI[float])Classify_Tall

assign class: with Mean nDSM > 3 at New Level: Tall assign class: Tall with Mean NDVI < -0.32 and Mean Blue >= 135 at New Level: Buildings assign class: Tall with Mean NDVI >= -0.32 at New Level: Vegetation assign class: Vegetation with Mean NIR <= 150 at New Level: Evergreen assign class: Vegetation with Mean NIR > 150 at New Level: Deciduous

Classify_Short assign class: unclassified with Mean nDSM <= 3 at New Level: Short assign class: Short with Mean Blue < 106 and Brightness < 86 at New Level: Water\Shadows assign class: Short with Mean NDVI <= 0.52 at New Level: Low\NonVegetation assign class: Low\NonVegetation, Short with Mean nDSM < 1.6 at New Level: Grass assign class: Grass, Low\NonVegetation, Short with Mean NDVI <= -0.01 at New Level: BareSoil assign class: BareSoil, Short with Mean NIR < 170 and Brightness > 140 at New Level: Roads\Parkinglots assign class: Roads\Parkinglots with Length\Width <= 1.5 at New Level: BareSoil

Clean up merge region: Evergreen at New Level: merge region merge region: Deciduous at New Level: merge region merge region: Buildings at New Level: merge region merge region: Water\Shadows at New Level: merge region merge region: Low\NonVegetation at New Level: merge region merge region: Grass at New Level: merge region merge region: BareSoil at New Level: merge region merge region: Roads\Parkinglots at New Level: merge regio

10

B. Forest Cover Type Classification System:

Loeffelholz, B. C., & Zimmerman, G. (2005). Cloquet Forestry Center: Cover Type MappingProject – 2005. University of Minnesota, Department of Forest Resources. St. Paul, MN.

Cover Types: Commercial Tree Species

Symbol Cover Type Description

A Aspen More than 66% of total BA aspen

B Paper Birch More than 66% of total BA paper birch

C White Cedar More then 66% of total BA white cedar

D Scotch Pine More than 66% of total BA scotch pine

E Bottomland Hardwoods Lowland site with > 66% of total BA comprised of a mix of ash, cottonwood, willow and maple. No one species is greater than 66% of BA.

F Balsam Fir More than 66% of total BA balsam fir

H Black Ash More than 66% of total BA black ash

J Jack Pine More than 66% of total BA jack pine

Q Mixed Swamp Conifers Lowland site with > 66% of total BA comprised of a mix of spruce, cedar, balsam and tamarack. No one species is greater than 66% of BA.

UC Upland Conifers Upland site with > 66% of total BA comprised of a mix of red pine, white pine, jack pine, spruce and balsam.

MS Maple More than 66% of total BA is maple.

N Northern Hardwoods Upland site with > 66% of total BA comprised of a mix of birch, maple, aspen and ash. No one species is greater than 66% of BA.

NS Norway Spruce More than 66% of total BA Norway spruce.

O Oak More than 66% of total BA is oak.

HC Mix Hardwood/Conifer Upland site with a mix of hardwood and conifer species with no one species having > 66% of BA.

R Red Pine More than 66% of total BA red pine

S Black Spruce More than 66% of total BA black spruce

11

T Tamarack More than 66% of total BA tamarack

W White Pine More than 66% of total BA white pine

WS White Spruce More than 66% of total BA white spruce

12