Technical ReportPDF Available

A fine scale approach to riparian habitat classification and spatial analysis in the St. Mary's River, Nova Scotia, Canada

Authors:

Abstract

Human land-use activities that occur near and around watercourses and waterbodies play a critical role in the overall health, function and biodiversity of aquatic habitats. Changes to riparian zones, areas between a waterbody’s high-water mark and the upland area, can cause cascading effects to aquatic environments. Many provinces throughout Canada have implemented fixed-width riparian buffer zones around watercourses to help mitigate impacts from land use change to aquatic ecosystems. There is a complex set of regulations and guidelines that apply to different land use activities in Nova Scotia's riparian zones due, in part, to overlapping jurisdictions. However, approaches to investigate riparian ecosystem health at the watershed scale have been underutilized in this region. The St. Mary’s River, located along the Eastern Shore of Nova Scotia, is a relatively natural watershed with dynamic aquatic and riparian habitats. This study provides robust methods to classify land cover dynamics and land use activities using Sentinel-2A satellite imagery and four fixed-width riparian buffer zones (30 m, 100 m, 150 m, and 300 m) within the St. Mary’s River watershed, Nova Scotia, Canada. Riparian zones were classified to identify fine-scale patterns in disturbance, identifying areas that may benefit from conservation or restoration across various land ownership types (i.e., Crown versus private land).
1
A fine scale approach to riparian habitat classification and
spatial analysis in the St. Mary's River, Nova Scotia,
Canada
Caelin A.E. Murray, Ben R. Collison, Aimee G. Gromack, Madeline M.
Lawler, Sarah M. Tuziak, and Sean M.M. Butler
Aquatic Ecosystems Branch
Maritimes Region
Fisheries and Oceans Canada
Bedfored Institute of Oceanography
1 Challenger Drive
Dartmouth, NS B2Y 4A2
2024
Canadian Technical Report of
Fisheries and Aquatic Sciences 3610
Canadian Technical Report of Fisheries and Aquatic Sciences
Technical reports contain scientific and technical information that contributes to existing knowledge but
which is not normally appropriate for primary literature. Technical reports are directed primarily toward a
worldwide audience and have an international distribution. No restriction is placed on subject matter and the series
reflects the broad interests and policies of Fisheries and Oceans Canada, namely, fisheries and aquatic sciences.
Technical reports may be cited as full publications. The correct citation appears above the abstract of each
report. Each report is abstracted in the data base Aquatic Sciences and Fisheries Abstracts.
Technical reports are produced regionally but are numbered nationally. Requests for individual reports will
be filled by the issuing establishment listed on the front cover and title page.
Numbers 1-456 in this series were issued as Technical Reports of the Fisheries Research Board of Canada.
Numbers 457-714 were issued as Department of the Environment, Fisheries and Marine Service, Research and
Development Directorate Technical Reports. Numbers 715-924 were issued as Department of Fisheries and
Environment, Fisheries and Marine Service Technical Reports. The current series name was changed with report
number 925.
Rapport technique canadien des sciences halieutiques et aquatiques
Les rapports techniques contiennent des renseignements scientifiques et techniques qui constituent une
contribution aux connaissances actuelles, mais qui ne sont pas normalement appropriés pour la publication dans un
journal scientifique. Les rapports techniques sont destinés essentiellement à un public international et ils sont
distribués à cet échelon. II n'y a aucune restriction quant au sujet; de fait, la série reflète la vaste gamme des
intérêts et des politiques de Pêches et Océans Canada, c'est-à-dire les sciences halieutiques et aquatiques.
Les rapports techniques peuvent être cités comme des publications à part entière. Le titre exact figure au-
dessus du résumé de chaque rapport. Les rapports techniques sont résumés dans la base de données Résumés des
sciences aquatiques et halieutiques.
Les rapports techniques sont produits à l'échelon régional, mais numérotés à l'échelon national. Les demandes
de rapports seront satisfaites par l'établissement auteur dont le nom figure sur la couverture et la page du titre.
Les numéros 1 à 456 de cette série ont été publiés à titre de Rapports techniques de l'Office des recherches sur
les pêcheries du Canada. Les numéros 457 à 714 sont parus à titre de Rapports techniques de la Direction générale
de la recherche et du développement, Service des pêches et de la mer, ministère de l'Environnement. Les numéros
715 à 924 ont été publiés à titre de Rapports techniques du Service des pêches et de la mer, ministère des Pêches et
de l'Environnement. Le nom actuel de la série a été établi lors de la parution du numéro 925.
i
Canadian Technical Report of Fisheries and Aquatic Sciences 3610
2024
A fine scale approach to riparian habitat classification and spatial
analysis in the St. Mary's River, Nova Scotia, Canada
Caelin A.E. Murray1, Ben R. Collison1,2, Aimee G. Gromack1, Madeline M. Lawler1,
Sarah M. Tuziak1, and Sean M.M. Butler1
1 Aquatic Ecosystems Branch
Maritimes Region
Fisheries and Oceans Canada
Bedford Institute of Oceanography
1 Challenger Drive
Dartmouth, NS B2Y 4A2
2 School for Resource and Environmental Studies
Dalhousie University
Halifax, NS B3H 4R2
ii
© His Majesty the King in Right of Canada, as represented by the Minister of the
Department of Fisheries and Oceans, 2024.
Cat. No. Fs97-6/3610E-PDF ISBN 978-0-660-71476-9 ISSN 1488-5379
Correct citation for this publication:
Murray, C.A.E., Collison, B.R., Gromack, A.G., Lawler, M.M., Tuziak, S.M., and Butler,
S.M.M. 2024. A fine scale approach to riparian habitat classification and spatial analysis
in the St. Mary's River, Nova Scotia, Canada. Can. Tech. Rep. Fish. Aquat. Sci. 3610: viii
+61 p.
iii
Table of Contents
List of Figures .............................................................................................................. iv
List of Tables ................................................................................................................ iv
List of Abbreviations .................................................................................................... vi
Abstract ........................................................................................................................ vii
Résumé ....................................................................................................................... viii
1. Introduction ............................................................................................................. 1
1.1. Study area .......................................................................................................... 3
1.1.1. Geographical scope ..................................................................................... 3
1.1.2. Biodiversity .................................................................................................. 3
1.2. Research objectives ........................................................................................... 4
2. Methods ................................................................................................................... 5
2.1. Acquisition and preparation of satellite imagery ................................................. 5
2.2. Validation and identification of land cover land use classes ............................... 6
2.3. Supervised Classification ................................................................................. 10
2.4. Reclassification ................................................................................................ 11
2.5. Data Analysis: Riparian Zone Fixed-width Buffers ........................................... 12
2.6. Data Analysis: Weighting and Ranking ............................................................ 14
3. Results ................................................................................................................... 16
3.1 Land Cover and Land Use Supervised SVM Classification .............................. 16
3.2 Riparian Fixed-width Buffer Summaries ........................................................... 21
3.3 Weighting and Ranking .................................................................................... 25
4. Discussion & Conclusion .................................................................................... 31
Acknowledgements ..................................................................................................... 36
References ................................................................................................................... 37
Appendix A Validation and Class Codes ............................................................... 44
Appendix B Data cleaning and pre-processing ..................................................... 47
Appendix C Fixed-width Riparian Buffers (estuary coastline) ............................. 48
Appendix D Proportion of LCLU Class per sub watershed ID (100 m riparian
buffer) ........................................................................................................................... 49
iv
List of Figures
Figure 1. Pathway of effects to fish habitat from riparian disturbances. ....................................................... 1
Figure 2. The 12-band Sentinel 2A satellite imagery of the St. Mary’s River watershed ............................. 4
Figure 3. Methodological workflow for generating the LCLU map for the St. Mary’s River watershed. ....... 5
Figure 4. An example of data summarized for private lands within each of the four fixed-width riparian
buffers.. ....................................................................................................................................................... 13
Figure 5. Sub watershed identifiers (ID) within the St. Mary’s River watershed ......................................... 14
Figure 6. Results of the classified raster (LCLU raster 1) supervised Support Vector Machine (SVM) land
cover land use classification. ...................................................................................................................... 17
Figure 7. Results of the reclassified raster (LCLU raster 2) ........................................................................ 19
Figure 8. LCLU reclassified raster clipped to each of the fixed-width riparian buffers................................ 22
Figure 9. Land ownership as well as conservation and protected areas within the St. Mary’s River
watershed. ................................................................................................................................................... 25
Figure 10. Results of the weighted sum and ranking 1-9 using the classified land cover land use classified
raster ........................................................................................................................................................... 28
Figure 11. Results of the weighted sum and ranking 1-8 using the classified land cover land use classified
raster. .......................................................................................................................................................... 29
Figure 12. Results of the weighted sum and ranking 1-3 using the classified land cover land use classified
raster ........................................................................................................................................................... 30
List of Tables
Table 1. Description and data sources used in the validation of land cover and land use classification of the
St. Mary’s River watershed. .......................................................................................................................... 7
Table 2. LCLU class name and associated descriptions as defined in the assembled validation datasets. 8
Table 3. Validation datasets used to classify the Sentinel 2A image. .......................................................... 9
Table 4. Training data used to classify the Sentinel 2A satellite image.. .................................................... 10
Table 5. Testing data used to classify the Sentinel 2A satellite image. ...................................................... 10
Table 6. Re-grouped LCLU used in the reclassified land cover map. ........................................................ 11
Table 7. Testing data (20%) used to validate the reclassified raster. ......................................................... 12
Table 8. Ranking of each LCLU from 1-9. .................................................................................................. 15
Table 9. Ranking of each LCLU from 1-8 ................................................................................................... 15
Table 10. Ranking of each Land Cover Land Use (LCLU) from 1-3 ........................................................... 16
Table 11. Distribution of LCLU classes within the St. Mary’s River watershed from the Sentinel 2 Level 2A
satellite image ............................................................................................................................................. 17
Table 12. Accuracy assessment of Raster 1 supervised Support Vector Machine Land cover land use
classification ................................................................................................................................................ 18
v
Table 13. Distribution of LCLU classes within the St. Mary’s River watershed .......................................... 20
Table 14. Accuracy assessment of the reclassified land cover land use map ........................................... 20
Table 15. Pixels reclassified using LCLU Raster 1 to generate LCLU Raster 2. ........................................ 21
Table 16. Proportion of LCLU that is natural and disturbed for each fixed-width riparian buffer. ............... 23
Table 17. Distribution of land ownership within each of the three main branches of the St. Mary’s Watershed
(Figure 2). .................................................................................................................................................... 23
Table 18. Distribution of LCLU activities on crown lands (“C”) compared to private lands (“P”) within each
fixed-width riparian buffers. ......................................................................................................................... 24
Table 19. Total area (in hectares) of OGF in the St. Mary’s River watershed that is protected under the
Nova Scotia Old Growth Forest Policy ........................................................................................................ 25
vi
List of Abbreviations
AAFC Agriculture and Agri-food Canada
ARA Active River Area
ASCII American Standard Code for Information Interchange
CLDFHD Canada Landsat Derived Forest Harvest Disturbance
DFO Fisheries and Oceans Canada
GIS Geographic Information Systems
ID Identifier
LCLU Land cover and land use
LiDAR Light detection and ranging system
NCC Nature Conservancy of Canada
NRCan Natural Resources Canada
NSFI Nova Scotia Forest Inventory
NSTD Nova Scotia Topographic Database
OBIA Object-based Image Analysis
OGF Nova Scotia Old Growth Forest Policy
SMR St. Mary’s River
SVM Support Vector Machine
US United States
vii
Abstract
Murray, C.A.E., Collison, B.R., Gromack, A.G., Lawler, M.M., Tuziak, S.M., and Butler,
S.M.M. 2024. A fine scale approach to riparian habitat classification and spatial analysis
in the St. Mary's River, Nova Scotia, Canada. Can. Tech. Rep. Fish. Aquat. Sci. 3610: viii
+61 p.
Human land-use activities that occur near and around watercourses and waterbodies play
a critical role in the overall health, function and biodiversity of aquatic habitats. Changes
to riparian zones, areas between a waterbody’s high-water mark and the upland area,
can cause cascading effects to aquatic environments. Many provinces throughout
Canada have implemented fixed-width riparian buffer zones around watercourses to help
mitigate impacts from land use change to aquatic ecosystems. There is a complex set of
regulations and guidelines that apply to different land use activities in Nova Scotia's
riparian zones due, in part, to overlapping jurisdictions. However, approaches to
investigate riparian ecosystem health at the watershed scale have been underutilized in
this region. The St. Mary’s River, located along the Eastern Shore of Nova Scotia, is a
relatively natural watershed with dynamic aquatic and riparian habitats. This study
provides robust methods to classify land cover dynamics and land use activities using
Sentinel-2A satellite imagery and four fixed-width riparian buffer zones (30 m, 100 m, 150
m, and 300 m) within the St. Mary’s River watershed, Nova Scotia, Canada. Riparian
zones were classified to identify fine-scale patterns in disturbance, identifying areas that
may benefit from conservation or restoration across various land ownership types (i.e.,
Crown versus private land).
viii
Résumé
Murray, C.A.E., Collison, B.R., Gromack, A.G., Lawler, M.M., Tuziak, S.M., and Butler,
S.M.M. 2024. A fine scale approach to riparian habitat classification and spatial analysis
in the St. Mary's River, Nova Scotia, Canada. Can. Tech. Rep. Fish. Aquat. Sci. 3610: viii
+61 p.
Les activités d’utilisation humaine des terres qui se déroulent à proximité et autour des
cours d’eau et des plans d’eau jouent un rôle essentiel dans la santé, la fonction et la
biodiversité globales des habitats aquatiques. Les modifications apportées aux zones
riveraines, c’est-à-dire les zones situées entre la ligne des hautes eaux d’un plan d’eau
et la zone sèche, peuvent avoir des effets en cascade sur les milieux aquatiques. De
nombreuses provinces canadiennes ont mis en place des zones tampons riveraines de
largeur fixe autour des cours d’eau pour aider à atténuer les répercussions des
changements d’utilisation des terres sur les écosystèmes aquatiques. Il existe un
ensemble complexe de règlements et de lignes directrices qui s’appliquent aux
différentes activités d’utilisation des terres dans les zones riveraines de la Nouvelle-
Écosse, en partie en raison de compétences qui se chevauchent. Toutefois, les
approches visant à étudier la santé des écosystèmes riverains à l’échelle des bassins
versants ont été sous-utilisées dans cette région. La rivière St. Marys, située le long de
la côte est de la Nouvelle-Écosse, est un bassin versant relativement naturel l’on
trouve des habitats aquatiques et riverains dynamiques. La présente étude fournit des
méthodes robustes pour classifier la dynamique de la couverture terrestre et les activités
d’utilisation des terres en utilisant le système d’imagerie satellite Sentinel-2A et quatre
zones tampons riveraines de largeur fixe (30 m, 100 m, 150 m et 300 m) dans le bassin
versant de la rivière St. Marys, en Nouvelle-Écosse, au Canada. On a classé les zones
riveraines afin de déterminer les schémas de perturbation à petite échelle et les zones
qui pourraient bénéficier d’une conservation ou d’une restauration dans les différents
types de propriétés (c’est-à-dire les terres de la Couronne et les terres privées).
1
1. Introduction
The riparian zone is defined as the area located between a waterbody’s high-water mark
and the upland area (DFO 2020a). Riparian habitats are unique and complex ecological
systems that sustain considerable amounts of biodiversity and provide essential
ecosystem services to both terrestrial and aquatic species and their habitats (Caskenette
et al. 2020; Riis et al. 2020). Streams, floodplains, wetlands, and the adjacent surrounding
land are dynamic and interlinked, representing a sensitive two-way connection that may
transition across spatial and temporal hydrological shifts (e.g., river meandering;
Tolkkinen et al. 2020). Migratory and resident fish depend on riparian ecosystems to
maintain appropriate habitat attributes (e.g., temperature and water quality) for survival
and reproduction. Healthy riparian zones function as a buffer to protect fish habitat from
point-source threats, like sediment runoff, and chronic, longer-term threats, such as
climate change (Albertson et al., 2018). When riparian zones are destroyed or degraded,
negative cascading effects can often occur in aquatic environments (Figure 1). To
mitigate land use impacts on aquatic ecosystems many provinces have implemented
fixed-width riparian buffer zones around watercourses (Collison & Gromack 2022; Tiwari
et al. 2016; Richardson et al. 2012; De Sosa et al. 2017).
Figure 1. Pathway of effects to fish habitat from riparian disturbances (Collison & Gromack,
2022).
Advancements in the accessibility of satellite imagery have increased researchers' ability
to remotely map and classify land cover and land use (LCLU) activities, including those
within riparian ecosystems. Land cover refers to the physical morphology and biology of
the landscape (such as forests, wetlands, impervious surfaces), while land use is how
2
humans are modifying the land cover (Lambin et al. 2001; Collison & Gromack 2022).
Several spatial monitoring techniques using geographic information systems (GIS) and
satellite imagery have been applied globally to increase our understanding of land use
activities, disturbance, and monitor changes to landscapes (Aresnault & O’Sullivan 2021;
Daryaei et al. 2020; Eskandari & Pourghasemi 2022; Mary-Lauyé et al. 2022; Furuya et
al. 2020; Piedelobo et al. 2019; Phiri et al. 2020; Rusnák et al. 2022). In Canada, the use
of remotely sensed data (satellite imagery and light detection and ranging system
(LiDAR)) has been used to characterize fish habitat to support management decisions for
riparian conservation and restoration planning (Bachiller-Jareno et al. 2019; Budlong
2004; Coleshill & Watt 2017; Jones et al. 2006; Kupier et al. 2022; France & Pardy 2018;
Tompinski et al. 2017; Roth et al. 2020). The use of satellite imagery to identify LCLU and
disturbances are essential components in understanding what activities have altered
riparian zones which have the potential to negatively impact aquatic ecosystems.
In Atlantic Canada, Nussey and Noseworthy (2020) adapted the United States (US)
Nature Conservancy’s Active River Area (ARA) model and applied it to the Northern
Appalachian Acadian Region. Land cover intactness was calculated within the ARA
across different watersheds using a 30-metre resolution Landsat-derived product, which
was classified as natural, clear cut, agriculture, or developed land (Nussey & Noseworthy
2020). In this context, “intactness” was defined as landcover types including forest,
wetland, barren, and water (Nussey & Noseworthy 2020). While this product serves as a
valuable resource at the regional level, there remains a need for higher resolution GIS
and remote sensing analyses of the riparian zone at the watershed scale. This can be
achieved by exploring the use of higher resolution satellite imagery, topographic datasets
(i.e., LiDAR), and more detailed classifications of different land use activities. High
resolution imagery (i.e., Sentinal-2A) can offer greater quantities of information to
increase the accuracy of riparian zone classification compared to coarser resolution
images such as those obtained by Landsat (Fauvel et al. 2012; Lacelle & Shi 2021;
Kamenova & Dimitrov 2021; Parker & Lee 2016).
Fisheries and Oceans Canada (DFO) (2020) has indicated that “the goal of protecting
riparian habitat is to ensure that there is sufficient area to provide the ecosystem services
(i.e., processes) that the aquatic habitat requires; which also means maintaining a large
enough riparian zone to allow for proper function and resilience of riparian features to
natural variation and to extreme events”. A literature review by Collison & Gromack (2022)
evaluated recommended fixed-width riparian buffer sizes in Nova Scotia, Atlantic Canada,
and the USA that are likely adequate for protecting fish and fish habitat. Several studies
have examined the impact of different riparian buffer sizes on parameters important to
fish and fish habitat, including protection against contamination, riparian corridor
microclimate, stream temperature, invertebrate prey, leaf litter input, input of fine
sediments, stream temperature, maintaining benthic communities, contamination,
maintaining water quality, shading, and cumulative effects (Collison & Gromack 2022).
Smaller buffer sizes (20-30 m) were adequate for parameters such as coarse woody
debris recruitment and bank stability, while buffers of 15-20 m were adequate to maintain
water quality. Consequently, increased buffer sizes, typically ≥ 30 m, are associated with
greater ecological and environmental benefits including the protection of fish and fish
habitat (Collison & Gromack 2022; Smokorowski & Pratt 2007; Stoffyn-Egli & Duinker
3
2013; Lahey 2018; Albertson et al. 2018; Cole et al. 2020; Lind et al. 2019). However,
while wider riparian buffers would provide greater protection to fish and fish habitat,
enforcing these buffers may be more difficult to implement in practice (Collison &
Gromack 2022; Richardson et al. 2012; Tiwari et al. 2016).
DFO is now exploring the inclusion of riparian zones in Species at Risk Act (SARA) critical
habitat (Species at Risk Act, s. 58(1)(b); Caskenette et al., 2021) and Fisheries Act
Ecologically Significant Area (ESA; Fisheries Act, s. 35.2; Collison & Gromack 2022)
designations and how it may be managed given the cross-jurisdictional responsibilities
and potential challenges. A baseline understanding of riparian zone intactness is
important to identify potential locations to apply conservation actions. Here, we highlight
the application of fine-scale riparian zone spatial analysis to fish habitat, using the St.
Mary’s River (Mik’maw: Napu’saqnuk) watershed in Nova Scotia, Canada, with a diversity
of LCLU dynamics.
1.1. Study area
1.1.1. Geographical scope
The St. Mary’s River (Figure 2) is one of the longest rivers (250 km) in Nova Scotia,
Canada, flowing through five counties (Guysborough, Antigonish, Colchester, Pictou and
Halifax) and drains into the Atlantic Ocean at the Sonora estuary (Government of Canada
2021; St. Mary’s River Association 2019). The watershed covers approximately 1,350
km2 and is comprised of three main branches; East, West, and North, which combine to
form the main branch that drains into the Atlantic Ocean (St. Mary’s River Association
2019).
1.1.2. Biodiversity
The St. Mary’s River contains a wide diversity of fish species, including Atlantic Salmon
(Salmo salar; Nova Scotia Southern Upland designatable unit), Brook Trout (Salvelinus
fontinalis), American Eel (Anguilla rostrata), White Sucker (Catostomus commersonii),
Sea Lamprey (Petromyzon marinus), and Rainbow Smelt (Osmerus mordax) (St. Mary’s
River Association, 2019). Both Atlantic Salmon (Southern Upland population) and
American Eel have been assessed as by the Committee on the Status of Endangered
Wildlife in Canada (COSEWIC) and are under consideration for Species at Risk Act
(SARA; Atlantic Salmon as Endangered and American Eel as Threatened) listing
(COSEWIC 2010; COSEWIC 2012). Brook Floater (Alasmidonta varicosa), a freshwater
mussel listed as Special Concern under the SARA, is also present in the East and North
branches of the watershed (Government of Canada 2018; St. Mary’s River Association
2019). The species is also listed as Threatened under Nova Scotia’s Endangered Species
Act and has ‘core habitat’ identified along the entire East and North branches of the
watershed, including a 30 m riparian buffer extending out from the high-water mark in
areas where the species is found (Government of Canada 2023). According to the
Endangered Species Act, core habitat refers to “specific areas of habitat essential for the
long-term survival and recovery of endangered or threatened species and that are
designated as core habitat pursuant to s. 16 or identified in an order made pursuant to s.
18 (Government of Canada 2023).
4
Figure 2. The 12-band Sentinel 2A satellite imagery of the St. Mary’s River watershed (June 10,
2019). The 12-band image is displayed in true colour composite (bands 4-R 3-G 2-B). Images
were derived from tile identifier named ‘TNR’ and ‘TNQ’ footprints. Bands in each image were
resampled to 10 m and mosaicked in ArcGIS Pro v.2.8 to obtain complete coverage of the
watershed. The watershed boundary shapefile was obtained from Open Data Nova Scotia
(1:10,000 Primary Watersheds layer).
1.2. Research objectives
Objectives of this project were to:
1. Identify LCLU within the riparian zone of the St. Mary’s River and all connecting
surface waterbodies (including lakes, rivers, streams, wetlands, and estuaries);
2. Determine the proportion and types of land use activities occurring within fixed-
width riparian buffers;
3. Identify how LCLU activities vary between Crown and private land; and
4. Rank sub-watersheds in terms of “intactness” to determine which areas of the
watershed may benefit most from proactive riparian conservation and/or
protection.
5
2. Methods
Terrestrial LCLU classes within the St. Mary’s River watershed were characterized using
a supervised support vector machine (SVM) classification of Sentinel-2 satellite imagery.
Previous classified land cover maps as well as field-validated data sources were used to
train the classification and to validate the LCLU classified image (Figure 3).
Figure 3. Methodological workflow for generating the LCLU map for the St. Mary’s River
watershed.
2.1. Acquisition and preparation of satellite imagery
Atmospherically corrected satellite images were downloaded from the Copernicus Open
Access Hub (https://scihub.copernicus.eu/dhus/ #/home). Sentinel 2 Level 2A top of
atmosphere imagery was acquired, as this satellite offered open-source data with a high
temporal (5-day re-visit time) and spatial (10-60 m) resolution compared to Landsat (16-
day revisit time and a 30 m resolution; (European Space Agency 2022a; 2022b). Images
were filtered by the following criteria: (1) less than 5% cloud cover, and (2) whether
6
appropriate to mosaic together for full watershed coverage. Two images from June 10,
2019 were used, demonstrating full coverage of the watershed. All 12 bands of the
Sentinel-2A imagery were downloaded and resampled to 10 m in ArcGIS Pro v. 2.8 using
the Resample tool and the ‘nearest’ sampling technique.
2.2. Validation and identification of land cover land use classes
Datasets considered for validating the LCLU classification map of the St. Mary’s River
watershed classification came from five sources: 1) Nova Scotia Forest Inventory (NSFI);
2) Nova Scotia Topographic Database (NSTD) roads, trails and rails line layer and the
break line layers (merged and dissolved together); 3) Canada Landsat Derived Forest
Harvest Disturbance (CLDFHD); 4) Natural Resources Canada (NRCan) Classified Land
Cover; and 5) Agriculture and Agri-food Canada (AAFC) Annual Crop Inventory (Table
1).
7
Table 1. Description and data sources used in the validation of land cover and land use
classification of the St. Mary’s River watershed. Note: data source webpages are hyperlinked to
the names in the ‘Dataset name’ column.
Dataset name
Description
Organization
Data type,
Coordinate
System,
Resolution
Year(s)
Nova Scotia
Forest Inventory
(NSFI)
This dataset includes 23 forested classes
and 26 non-forested classes (e.g.,
agriculture, freshwater wetlands and
coastal habitats). Forested stands include
attributes for tree species, percentage,
height and crown closure, plus calculated
values such as volume. Inventory is
maintained through aerial photo
interpretation and is supplemented with
field data at select locations.
Province of
Nova Scotia
Polygon vector,
NAD83 UTM
(1:10,000)
2004-
2018
Nova Scotia
Topographic
Database (NSTD)
roads, trails, and
rails
This layer contains information on roads,
trails and rails and was used to identify
roads as urban areas.
Province of
Nova Scotia
Line vector,
WGS 1984
(1:10,000)
2002-
present
Canada Landsat
Derived Forest
Harvest
Disturbance
(CLDFHD)
This raster contains 16 classes that reflect
the year forest loss occurred. The raster
does not contain information on the type of
loss, only the year loss occurred. This
includes any conversion of natural forests,
be it plantations, selective logging or
shifting cultivation practiced by local
communities.
Canadian
Forest Service
Raster, NAD83
UTM (30 m
resolution)
1985-
2020
Natural Resources
Canada Classified
Land Cover Map
2020 (NRCan)
This raster contains information on land
cover with a total of 15 classes.
Natural
Resources
Canada
Raster, WGS
1984 (30 m
resolution)
2020
Agriculture and
Agri-food Canada
Annual Crop
Inventory
Classified Land
Cover Map
(AAFC)
This raster contains 72 classes including
information on forest cover, wetlands,
agriculture, etc.
Agriculture and
Agri-food
Canada
Raster, WGS
1984 (30 m
resolution)
2021
Forest composition and relative age can have a major influence on riparian zone service
provisioning to the aquatic environment. Many old growth hardwood and Acadian forest
stands have been lost due to timber harvesting throughout Nova Scotia, shifting the
composition of tree species over time (Noseworthy & Beckley 2020). The long history of
forest management in the St. Mary's River warranted an investigation into the general
composition of forest stands to better understand potential interactions with freshwater
8
fish habitat in the watershed. Therefore, the first classification separated forest into
coniferous, deciduous, and mixedwood classes.
Validation points were grouped into nine general land cover classes: 1) coniferous, 2)
deciduous, 3) mixedwood, 4) shrubland, 5) grassland, 6) wetland, 7) agriculture, 8)
barren, and 9) urban based on what was observed to be consistent among all five
datasets (Table 2). Only the first three datasets (NSTD, NRCan, and AAFC) were used
to guide (i.e., train) the classification. The remaining two datasets (NSFI and CLDFHD)
were then used for reclassification and validation of the reclassified raster.
Table 2. LCLU class name and associated descriptions as defined in the assembled validation
datasets. Only the first three validation datasets (NRCan, AAFC, and NSTD) were used for the
first image classification.
LCLU Class
Class description
Coniferous
Coniferous forested stand which has not been treated silviculturally and does not
qualify as clear cut, partial cut, burn, old field, windthrow, alders, brush, or dead
categories. Those identified as natural stand under NS forest inventory layer.
Deciduous
Deciduous forested stand which has not been treated silviculturally and does not
qualify as clear cut, partial cut, burn, old field, windthrow, alders, brush, or dead
categories. Those identified as natural stand under NS forest inventory layer.
Mixed wood
Mixed forest stand which has not been treated silviculturally and does not qualify as
clear cut, partial cut, burn, old field, windthrow, alders, brush, or dead categories.
Those identified as natural stand under NS forest inventory layer.
Shrubland
Predominantly woody vegetation of relatively low height (generally +/-2 metres). May
include grass or wetlands with woody vegetation, regenerating forest. This is classified
in the AAFC layer and NR Can land cover data as shrubland.
Grassland
This class includes native grasses and other herbaceous vegetation and may include
some shrubland cover.
Wetland
Any wet area, not identified as a lake, river or stream, excluding open and treed bogs,
and beaver flowage.
Urban
Any area used primarily as residential, industrial and related structures such as
streets, sidewalks, parking lots, railway surfaces, industrial sites, mine structures, etc.
This includes NSDNR roads and railways.
Agriculture
Periodically cultivated areas. These include tame grasses and other perennial crops
such as alfalfa and clover grown alone or as a mixture for hay, pasture, or seed. It also
includes any hay field, pasture, tilled crop, or orchard including annual and perennial
crops which contain no merchantable tree species.
Barren
Land that is predominately non-vegetated and non-developed. Includes: glacier, rock,
sediments, burned areas, rubble, mines, other naturally occurring non-vegetated
surfaces. Excludes fallow agriculture.
To prepare validation points for classification, the NSTD vector layer was converted to
raster and the NRCan and AAFC rasters were resampled to match the pixel resolution of
the Sentinel 2A satellite imagery. Classes from each dataset were reclassified based on
their class code (Appendix A). Water classes were removed from all validation datasets.
The NRCan validation data used all classes except urban, as this class was better
represented by the NSTD roads layer. The AAFC dataset was reduced to include only
agriculture, grassland, urban, and barren classes while the NSTD layer only used for the
9
urban class (roads). A 10 m x 10 m grid generated for the watershed was used to extract
class values from all resampled rasters and cleaned to remove duplicates. The points
were exported from ArcGIS Pro (v.2.8) and imported into R Studio (v.4.2.2) to derive a
random sample of points and to split the validation data into training and testing data for
classification and evaluating classification accuracy. A total of 15,644,309 validation
points were extracted and reduced to train and validate the supervised classification of
the LCLU map for the St. Mary’s River watershed (Table 3).
Table 3. Validation datasets used to classify the Sentinel 2A image.
Dataset
Land Cover Land Use Class
NRCan
AAFC
NSTD
Total Validation
Points
Coniferous
4,295,188
0
0
4,295,188
Deciduous
4,832,617
0
0
4,832,617
Mixed wood
4,191,192
0
0
4,191,192
Shrubland
1,184
0
0
1,184
Grassland
1,013,054
60,082
0
1,073,136
Wetland
23,931
0
0
23,931
Urban
280,117
175,792
374,066
829,975
Agriculture
135,829
130,609
0
266,438
Barren
661
129,987
0
130,648
Total
14,773,773
496,470
374,066
15,644,309
An initial point reduction was done in R Studio (subset function) through filtering each
dataset (Table 1) individually for each corresponding LCLU class (e.g. filtering the NRCan
dataset for the ‘urban’ class). A random sample of 1000 validation points per class was
then obtained for each dataset using the sample function in R Studio. Validation points
from all datasets were then merged into a single dataset, in which all duplicates were
removed using the unique grid ID. This ensured only a single point (and validation class)
fell within each 10 m pixel for validation. The sample size for the NRCan barren class was
limited to 661 points, therefore all validation points were used.
The final validation dataset was split into testing (80%) and training (20%) validation
datasets using the caret package in R. The final reduced and resampled validation
dataset resulted in a total of 20,665 validation points (n = 16,533 training data and n =
4,132 testing data) to guide and validate the classification (Table 4, Table 5). All validation
data was verified and reclassified on an as-needed basis against the Sentinel 2A satellite
imagery before this data was used to train and validate the classification. Contrasting the
previously assembled datasets (Table 2) against the satellite image was essential, as
validation data did not represent field ground-truthing. Validation points obtained from
NRCan and AAFC were comprised of classified data derived from classification outputs
(i.e., NRCan accuracy of 86.9% and AAFC accuracy of 64.4%).
10
Table 4. Training data used to classify the Sentinel 2A satellite image. A random sample of 1000
points per class was used to reduce the number of validation points.
Land Cover Land Use Class
Dataset
Conif.
Decid.
Mixed
wood
Shrub
land
Grass
land
Wet
land
Urban
Agri.
Barren
Total
NRCan
2,438
1,730
2,247
163
268
506
2,182
1,483
817
11,834
AAFC
6
33
39
1,940
34
21
70
1
193
2,337
NSTD
256
442
284
60
1104
15
76
5
120
2,362
Total
2,700
2,205
2,570
2,163
1,406
542
2,328
1,489
1,130
16,533
Table 5. Testing data used to classify the Sentinel 2A satellite image. A random sample of 1000
points per class was used to reduce the number of validation points.
Land Cover Land Use Class
Dataset
Conif.
Decid.
Mixed
wood
Shrub
land
Grass
land
Wet
land
Urban
Agri.
Barren
Total
NRCan
565
478
579
19
62
128
571
368
185
2,955
AAFC
1
17
7
555
3
1
0
0
0
584
NSTD
44
85
72
16
343
2
19
0
12
593
Total
610
580
658
590
408
131
590
368
197
4,132
2.3. Supervised Classification
A supervised SVM classification was conducted using the 12-band original Sentinel-2A
satellite image to classify LCLU activities within the St. Mary’s River watershed. The SVM
was conducted using the 80% training data in ArcGIS Pro using the Train Support Vector
Machine tool (Spatial Analyst) with a maximum of 1000 samples per class, the 12 band
Sentinel 2A image, and a nine class schema (as described in Table 2). Training areas
were examined against the classified SVM raster using the Inspect training samples tool
which assigned a score from 0 (inaccurate) to 1 (accurately classified) to help identify
misclassified training areas. The reduced LCLU class validation points (Table 4; Table 5)
were manually inspected to validate the class assigned to a given point against the raw
satellite image. True classification errors were manually assigned to a new class, but
areas misclassified by the SVM and identified by the Inspect trailing samples tool as
incorrect were not re-assigned as they were technically correct, but the SVM predictor
was not. Similar spectral signatures prevented the SVM from distinguishing between
classes in some locations (e.g., abandoned roads, barren vs. urban), and factors such as
misaligned image capture timing or sensor resolution may have contributed to initial
classification errors. The classified raster was imported to R Studio to further investigate
accuracy and generate error matrices. A script was developed to generate an error matrix
and calculate an overall accuracy and kappa statistic. The results of this script were
exported as an American Standard Code for Information Interchange (ASCII; .csv) file.
Once the overall accuracy and kappa statistic of the classified raster was sufficient (kappa
between 0.4-0.8), the classified raster was reclassified into nine broader classes using all
five datasets (Table 3; Appendix B).
11
2.4. Reclassification
The loss of mature riparian forest through natural (e.g., wildfire) or anthropogenic (e.g.,
timber harvesting) disturbance can significantly influence freshwater ecosystems (Fuller
et al. 2022; Cunningham et al. 2023). Further study into the relative age of the riparian
forests can offer insights at the spatial and temporal scale that broad tree species
composition cannot. The classified LCLU raster was reclassified to incorporate areas of
forest loss and secondary growth and re-group forest classes such as coniferous,
deciduous, and mixedwood into one category named natural stand (Appendix A). Forest
loss and secondary growth classes were reclassified using the NSFI and CAFDHD
datasets (Appendix A). The reclassified raster was comprised of nine broader classes: 1)
natural stand, 2) secondary growth, 3) forest loss, 4) agriculture, 5) barren, 6) shrubland,
7) grassland, 8) wetland, and 9) urban (Table 6). Creation of the final reduced and
resampled validation dataset followed the same approach described in Section 2.2. and
2.3., which resulted in 4,732 testing validation points to validate the reclassified raster
(Table 6).
Table 6. Re-grouped LCLU used in the reclassified land cover map. The reclassified map used
all five validation datasets. For class codes see Appendix A.
Class #
LCLU Class
Class description
1
Natural Stand
Any forested stand which has not been treated silviculturally and does not qualify
under clear cut, partial cut, burn, old field, windthrow, alders, brush, or dead
categories. Those identified as natural stand under NS forest inventory layer.
2
Secondary
Growth
Forest or woodland area which has re-grown after a timber harvest or clear-cut for
agriculture, wind-throw, or wildfire. Areas identified in NSDNR Forest inventory as of
2009 as burn, old field, windthrow, clear cut, partial depletion, partial cut are
presumed to now be Secondary Growth (2019 satellite image).
3
Forest Loss
Canadian Landsat Forest Harvest Derived Forest Harvest Disturbance 1985-2021
dataset represents 36 years of harvest change over Canada’s forests. These data
represent annual stand replacing forest changes by wildfire and harvest labelled by
year of disturbance.
4
Shrubland
Predominantly woody vegetation of relatively low height (+/-2 meters). May include
grass or wetlands with woody vegetation, regenerating forest. This is classified in the
AAFC layer and NRCan land cover data as Shrubland.
5
Grassland
This class includes native grasses and other herbaceous vegetation and may include
some Shrubland cover.
6
Wetland
Any wet area, not identified as a lake, river, or stream, excluding open and treed
bogs, and beaver flowage.
7
Urban
Any area used primarily as residential, industrial, and related structures such as
streets, sidewalks, parking lots, railway surfaces, industrial sites, mine structures.
This includes NSDNR roads and railways.
8
Agriculture
Periodically cultivated areas. These include tame grasses and other perennial crops
such as alfalfa and clover grown alone or as a mixture for hay, pasture, or seed. It
also includes any hay field, pasture, tilled crop, or orchard consisting of annual and
perennial crops (i.e., this does not contain merchantable tree species).
9
Barren
Land that is predominately non-vegetated and non-developed. These includes
glacier, rock, sediments, burned areas, rubble, mines, other naturally occurring non-
vegetated surfaces. Excludes fallow agriculture.
12
Table 7. Testing data (20%) used to validate the reclassified raster.
Land Cover Land Use Class
Dataset
Nat.
Stand
Second.
Growth
Forest
Loss
Agri.
Urb.
Barren
Grass
land
Wet
land
Shrub
land
Total
NRCan
1,305
0
0
14
57
120
343
366
144
2,349
AAFC
21
0
0
548
3
1
0
0
0
573
NSTD
166
0
0
13
298
2
9
0
9
497
NSFI
0
343
0
0
10
0
0
0
0
353
CLDFHD
2
0
351
0
0
0
0
0
0
353
Total
1,494
343
351
575
368
123
352
366
153
4,125
The CLDFHD raster was resampled from 25 m to 10 m to match the pixel resolution of
the satellite imagery. The 2004-2018 NSFI polygon layer was converted to a raster with
a resolution of 10 m. Both the CLDFHD and NSFI data were clipped to the St. Mary’s
River watershed and had water classes removed. For this study, the NSFI data layer is
considered older validation data and therefore disturbances listed in NSFI layer were
considered re-growth in the 2019 image. The CLDFHD layer was also used to compare
areas of loss that may have demonstrated regrowth in the 2019 image. Similarly, other
classes such as agriculture, barren, wetland, urban, etc., were not used from this layer as
changes to the land may have occurred within the data gap years. Finally, the CLDFHD
2000-2021 data layer was used to obtain the Forest Loss class for years ≥ 2011 and
2019. The CLDFHD layer records what year loss occurred yet does not indicate the type
of land-based activity or disturbance that caused Forest Loss as found in the NSFI layer.
The Reclassify tool was used to reclassify raster values and the Mosaic to new raster tool
was used to mosaic the reclassified pixels and non-reclassified areas back together. The
final reclassified raster consisted of nine broader classes: natural stand, secondary forest,
forest loss, agriculture, barren, shrubland, grassland, wetland, and urban.
2.5. Data Analysis: Riparian Zone Fixed-width Buffers
Baseline riparian buffers begin at the ordinary high-water mark or landward edge of a
floodplain and extend outward into the upland habitat around all connecting watercourses
and waterbodies including lakes, rivers, streams, and wetlands (Collison & Gromack
2022). In this study, four fixed-width riparian buffers (30 m, 100 m, 150 m, and 300 m)
were generated within the St. Mary’s River watershed. The 30 m riparian buffer was
chosen as this size is consistent with current buffer regulations in Nova Scotia and is
sufficient in capturing changes to LCLU (Collison & Gromack 2022). The three additional
fixed-width buffers of 100 m, 150 m, and 300 m offered a larger sample size of classified
LCLU types to assess and analyzed riparian LCLU at broader spatial scales.
Riparian fixed-width buffers (30 m, 100 m, 100 m estuarine coastline, and 300 m) were
derived from the 1:50,000 CanVec hydrographic features of water courses and water
bodies layers (Appendix C, estuarine Coastline). A 150 m riparian fixed-width buffer was
generated using the NCC water line and water bodies layer. Buffers of varying widths
were used to assess the potential differences in LCLU distribution ranging from near-
shore riparian habitat and extending outward from the high-water mark. The NCC water
layer was compared with the CanVec layer to examine differences in LCLU distribution
13
(using a middle width buffer between our maximum and minimum) based on the predictive
hydrological model base layer used to build the buffer zone.
LCLU classes within each of the four fixed-width buffers were examined using the
reclassified raster clipped to each riparian buffer. The proportion (%) of LCLU was
calculated for each buffer throughout the watershed. To examine how land classes were
distributed among various land ownerships, the Nova Scotia Crown Land layer was
clipped to each of the four riparian fixed-width buffer zones.
The Nova Scotia Old Growth Forest Policy (OGF) data was used to examine natural old
growth areas that are currently protected (Natural Resources and Renewables 2022).
The current provincial OGF Policy layer represents old growth areas protected from
timber harvesting that occur inside and outside of protected areas. Using the OGF layer,
old growth and old growth restoration opportunities that are located on crown land outside
of protected areas (select method = 1) and OGF located inside protected areas (crown or
private) (select method = 2) were examined (Natural Resources and Renewables 2022).
Total areas that were intact (natural) and disturbed were summarized for each riparian
fixed-width buffer size. The amount of natural and disturbed areas was calculated for
crown land, private lands, and OGF areas. Tools were placed in ArcGIS ModelBuilder for
each calculation and run separately to extract and summarize information for each LCLU
class, by ownership layer, and OGF Policy, for each of the various riparian buffers (Figure
4). Batch clips and other relevant tools were placed in model builder were frequently used
for all repetitive analyses. Private land in this analysis did not include privately owned
conservation lands and protected areas.
Figure 4. An example of data summarized for private lands within each of the four fixed-width
riparian buffers. The 100 m riparian buffer was done for the entire watershed as well as just the
estuarine coastline. Tools and layers were placed in model builder to summarize and calculate
proportions of LCLU within each buffer layer and this process was performed for all summary
data. Outside of the model, each buffer and land ownership data were clipped using a batch clip.
14
2.6. Data Analysis: Weighting and Ranking
To identify which sub watersheds would benefit from proactive riparian conservation
/protection (i.e., riparian habitat is highly intact), using a 100 m buffer, land cover classes
were ranked from 1 to 9 where 1 was most intact or highly natural (1 = natural stand) and
9 was least intact or less natural (9 = urban). Three versions of ranking were generated
to examine results (Tables 7, 8 and 9; Figure 5).
Figure 5. Sub watershed identifiers (ID) within the St. Mary’s River watershed derived from the
Nova Scotia 1:10,000 tertiary and sub-tertiary watershed boundary data layers.
In model builder, the reclassified land cover land use raster was clipped to the 100 m
fixed-width riparian buffer. The resulting clipped 100 m riparian land cover land use buffer
was then joined with the Nova Scotia sub watershed layers via the Spatial join tool. Each
land cover class was extracted using the Select tool. Once all classes were extracted,
each LCLU layer was re-joined using the Add join tool to determine the percent of each
land class grouping per sub watershed.
For example, in version 1, the sum (area m2) of the ‘natural standclass and wetland class
within a given sub watershed identifier (ID) were added together as natural stand, which
15
was divided by the total sum of all classes to obtain the percent of the ‘natural stand’ class
(Appendix D.). In version 2, no classes were combined and therefore each layer
represented the proportion of each LCLU class, and in version 3, select LCLU classes
were grouped together into highly natural, regenerative, and regeneration unlikely
categories (Table 9).
Once the proportion of area of each LCLU class grouping per sub watershed was
determined for all three versions, a single raster representing the proportion of a given
LCLU grouping were generated (i.e., total of 9 raster’s version 1, 8 raster’s version 2, and
3 raster’s version 3) (Tables 7,8 and 9). A weighting for each class was assigned
according to the importance (rank) of each land cover class. The Weighted Sum tool was
used to summarize the results of the three different ranking versions and was used to
identify areas based on sub watershed that may benefit from proactive riparian
conservation /protection using a 100 m buffer.
Table 8. Ranking of each LCLU from 1-9. Ranking is assigned from most intact (natural) to least
intact (less natural). Assigned weights sum to 100.
Rank
Overall Category
Version 1 LCLU Class
Grouping
Weight (%)
Weight (%) by
Category
1
Highly Natural
Natural Stand
0.28
0.64
2
Wetland
0.16
4
Shrubland
0.12
6
Grassland
0.08
3
Regenerative
Secondary Growth
0.14
0.24
5
Forest Loss
0.10
7
Regeneration Unlikely
Agriculture
0.06
0.12
8
Barren
0.04
9
Urban
0.02
Table 9. Ranking of each LCLU from 1-8 where natural stand and wetland are grouped as a single
class. Ranking is assigned from most to least intact (natural). Weights sum to 100. Overall
Category column represents a means to compare maps and versions of weighting.
Rank
Overall Category
Version 2 LCLU
Class Grouping
Weight (%)
Weight (%) by
Category
1
Highly Natural
Natural Stand + Wetland
0.44
0.64
3
Shrubland
0.12
5
Grassland
0.08
2
Regenerative
Secondary Growth
0.14
0.24
4
Forest Loss
0.10
6
Regeneration Unlikely
Agriculture
0.06
0.12
7
Barren
0.04
8
Urban
0.02
16
Table 10. Ranking of each Land Cover Land Use (LCLU) from 1-3 where 1) highly natural is
comprised of natural stand, wetland, shrubland, and grassland 2) regenerative forest is comprised
of secondary growth and forest loss, and 3) regeneration unlikely which is comprised of urban,
agriculture, and barren. Ranking is assigned from most to least intact (natural). Weights sum to
100.
Rank
Overall Category
Version 3 LCLU Class
Grouping
Weight (%)
Weight (%)
by Category
1
Highly Natural
Natural Stand +
Wetland +
Shrubland +
Grassland
-
-
-
-
0.64
2
Regenerative
Secondary Growth +
Forest Loss
-
0.24
3
Regeneration
Unlikely
Agriculture +
Barren +
Urban
-
-
-
0.44
3. Results
3.1 Land Cover and Land Use Supervised SVM Classification
The supervised SVM LCLU classification classified approximately 1,487 km2 of terrestrial
habitat in the St. Mary’s River watershed. The St. Mary’s River watershed (SMRW) is
comprised primarily of coniferous forest followed by deciduous and mixedwood forest,
grassland, shrubland, urban, agriculture, wetland, and barren (Table 11). The classified
map did not account for waterbodies, rivers or streams. The SVM LCLU classification
achieved an overall accuracy of 84.4% and a kappa statistic of 0.82, which is considered
substantial agreement (Sim & Wright 2005) (Figure 6; Table 12). The classification
demonstrated the wetland (92.6%) class to have the greatest number of correctly
identified pixels (producer’s accuracy) followed by agriculture (91.3%), barren (89.1%),
coniferous forest (88.7%,) urban and grassland (both at 86.8%), mixedwood forest
(86.6%), deciduous forest (84.9%), and shrubland (44.8%) class (Table 12). Shrubland
had the least amount of correctly identified pixels, and was most misclassified with
agriculture and deciduous areas (Table 12).
17
Figure 6. Results of the classified raster (LCLU raster 1) supervised Support Vector Machine
(SVM) land cover land use classification of the 12-band Sentinel-2A satellite imagery June 10,
2019.
Table 11. Distribution of LCLU classes within the St. Mary’s River watershed from the Sentinel 2
Level 2A satellite image, June 10, 2019.
LCLU class
Count of pixels
Area (km2)
% Area
Coniferous
4,088,139
408.81
27.49
Deciduous
3,587,684
358.77
24.13
Mixedwood
3,591,716
359.17
24.16
Shrubland
634,760
63.48
4.27
Grassland
1,490,452
149.05
10.02
Wetland
398,218
39.82
2.68
Urban
573,783
57.38
3.86
Agriculture
466,142
46.61
3.13
Barren
38,176
3.82
0.26
Total
14,869,070
1,486.91
100.00
18
Table 12. Accuracy assessment of Raster 1 supervised Support Vector Machine Land cover land
use classification of the 12-band Sentinel-2A satellite imagery June 10, 2019 using 20% testing
and 80% training data. LCLU class 1) coniferous, 2) deciduous, 3) mixedwood, 4) shrubland, 5)
grassland, 6) wetland, 7) urban, 8) agriculture, and 9) barren. Descriptions can be found in Table
2.
LCLU Class Number
Class
Name
1
2
3
4
5
6
7
8
9
Total
User
Acc
Omission
Error
Conif.
551
5
36
3
3
6
6
0
0
610
90.3
9.7
Decid.
1
501
25
16
16
0
8
13
0
580
86.4
13.6
Mixed
57
39
517
13
13
6
10
3
0
658
78.6
21.4
Shrub
2
9
5
154
7
1
5
14
0
197
78.2
21.8
Grass
0
5
0
83
474
11
8
8
1
590
80.3
19.7
Wetland
5
2
0
4
17
338
0
1
1
368
91.8
8.2
Urban
5
13
12
1
0
1
375
0
1
408
91.9
8.1
Agri.
0
16
1
66
15
2
7
473
10
590
80.2
19.8
Barren
0
0
1
4
1
0
13
6
106
131
80.9
19.1
Total
621
590
597
344
546
365
432
518
119
4,132
Kappa: 0.82
Prod.
Accuracy
88.7
84.9
86.6
44.8
86.8
92.6
86.8
91.3
89.1
Overall accuracy: 84.4%
The reclassified raster was comprised primarily of natural stand followed by secondary
growth, grassland, forest loss, urban, shrubland, wetland, agriculture, and barren (Table
13; Figure 7). Approximately 916 km2 (61%) of the watershed is natural stand (Table 13).
The reclassified raster resulted in a higher overall accuracy of 90.0% and a kappa statistic
of 0.89 demonstrating a substantial agreement (Table 14; Figure 7). Grouping all forested
classes (coniferous, mixedwood, and deciduous forest) as natural stand and validating
against the NSFI natural stand likely increased accuracy as error in differentiating
between forest type was removed. For example, the classified raster (LCLU 1)
demonstrated areas where mixedwood was misclassified with coniferous and deciduous
forest and contributing to the second highest omission error (Table 12). In the reclassified
raster no error was found between secondary growth and other classes as these areas
were carefully masked (clipped and erased) and assessed to match the satellite imagery
using the NSFI layer and the CLDFHD layers together. Secondary growth was the most
accurately classified LCLU class as this class was comprised of areas that were
previously classified as forest loss (NSFI and CLDFHD) and did not occur in areas
classified as natural stand (Table 14). It is possible that some areas of secondary growth
may be misclassified as natural stand and/or forest loss due to varying stages of
secondary growth occurring throughout the watershed and this is apparent in natural
stand being misclassified as forest loss (Table 14).
While the results of the error matrix for the reclassified raster (LCLU 2) (Table 13)
demonstrate a high overall accuracy, there remains a need for careful attention in
distinguishing natural stand (intact areas), secondary growth, and forest loss LCLU
classes using the NSFI and CLDFHD layers and when considering automation
processes. Natural stand remained the LCLU class that had the greatest mis-classified
validation points (Table 14). The natural stand class was the most complex LCLU class
19
due to the nature of assembling multiple forest types and based on the spectral signatures
may easily be confused with shrubland or agricultural classes. Urban held the highest
number of misclassified points within natural stand likely due to pixels placed too closely
to the road, on the shoulder of the road, or in regenerating areas. Similar trends in LCLU
raster 2 were found in LCLU raster 1 where shrubland class was misclassified with
agriculture as the shrubland class often occurred near crops for berries or edges of crops
and fallow fields. The reclassified LCLU raster 2 revealed the watershed has experienced
~6.08% forest loss between 2011 and 2019 (Table 14). In contrast, ~13.11% of the
watershed has regenerated.
Figure 7. Results of the reclassified raster (LCLU raster 2) using remaining validation datasets
(NSFI and CLDFHD).
20
Table 13. Distribution of LCLU classes within the St. Mary’s River watershed from the reclassified
raster.
LCLU class
Count of pixels
Area (km2)
% Area
Natural Stand
9,156,081
915.61
61.57
Secondary Growth
1,949,342
194.93
13.11
Forest Loss
903,829
90.38
6.08
Shrubland
454,329
45.43
3.06
Grassland
1,055,988
105.60
7.10
Wetland
394,531
39.45
2.65
Urban
532,005
53.20
3.58
Agriculture
390,891
39.09
2.63
Barren
33,515
3.35
0.23
Total
14,870,511
1,487.05
100.00
Table 14. Accuracy assessment of the reclassified land cover land use map. LCLU class 1)
natural stand, 2) secondary growth, 3) forest loss, 4) shrubland, 5) grassland, 6) wetland, 7) urban,
8) agriculture, and 9) barren. Descriptions can be found in Table 6.
LCLU Class Number
1
2
3
4
5
6
7
8
9
Total
User
Acc
Omission
Error
Natural
Stand
1380
0
12
28
25
12
24
13
0
1494
92.4
7.6
Second.
Growth
0
343
0
0
0
0
0
0
0
343
100.0
0.0
Forest Loss
1
0
345
1
3
0
1
0
0
351
98.3
1.7
Shrub-land
11
0
0
120
4
1
5
12
0
153
78.4
21.6
Grass-land
2
0
2
39
289
11
5
4
0
352
82.1
17.9
Wetland
5
0
0
4
17
338
0
1
1
366
92.3
7.7
Urban
29
0
0
1
1
2
333
1
1
368
90.5
9.5
Agriculture
16
0
0
62
12
2
7
466
10
575
81.0
19.0
Barren
1
0
1
4
1
0
13
6
97
123
78.9
21.1
Total
1445
343
360
259
352
366
388
503
109
4125
Kappa: 0.89
Prod.
Accuracy
95.5
100
95.8
46.3
82.1
92.3
85.8
92.6
89.0
Overall accuracy: 90.0%
This study reclassified all LCLU classes that occurred within the secondary growth and
forest loss areas and therefore captured areas where changes have occurred throughout
the watershed between 2011 and 2019. Conducting reclassification in the LCLU raster 2
demonstrated results that are useful in monitoring and measuring changes to landscapes
(Table 15). For example, within the SMRW 47.90% of forest loss occurred in areas
previously classified as grassland in LCLU raster 1. Other previously classified LCLU
classes demonstrated forest loss between 2011-2019 for example shrubland (19.88%),
followed by deciduous (12.31%), and agriculture (8.22%). Similarly, areas where forest
loss occurred between 2011 to 2019 demonstrated varying signs of regeneration where
secondary growth is regenerating as deciduous forest (87.18%), mixedwood forest
(70.23%), and coniferous forest (37.21%). The comparison in the classified raster (LCLU
1) and reclassified raster (LCLU 2) demonstrate methods in which changes to landscape
21
may be monitored and measured moving forward whether it is due to anthropogenic
impacts or natural events.
Table 15. Pixels reclassified using LCLU Raster 1 to generate LCLU Raster 2. Coniferous,
deciduous, and mixedwood forests were reclassified based on the NSFI and CLDFHD layers for
years 2011-2021.
Reclassified
Class
LCLU Raster 1
Class
Reclassified
pixels (Count)
Reclassified area
(km2)
Reclassified area
(%)
Secondary
Growth
Coniferous
372100
37.21
19.09
Deciduous
871800
87.18
44.72
Mixed wood
702300
70.23
36.03
Shrubland
390
0.039
0.02
Grassland
920
0.092
0.05
Wetland
46
0.0046
0.002
Urban
680
0.068
0.03
Agriculture
980
0.098
0.05
Barren
49
0.0049
0.003
Total
1,949,265
194.93
100.00
Forest Loss
Coniferous
19900
1.99
2.21
Deciduous
111000
11.1
12.31
Mixed wood
36400
3.64
4.04
Shrubland
179300
17.93
19.88
Grassland
431900
43.19
47.90
Wetland
3700
0.37
0.41
Urban
40800
4.08
4.52
Agriculture
74100
7.41
8.22
Barren
4600
0.46
0.51
Total
901,700
90.17
100.00
Natural Stand
Coniferous
3695200
369.52
40.38
Deciduous
2603900
260.39
28.45
Mixed wood
2852200
285.22
31.17
Total
9,151,300
915.13
100.00
3.2 Riparian Fixed-width Buffer Summaries
Using the four fixed-width riparian buffers (30 m, 100 m, 150 m, 300 m) the proportion
(%) of land cover that is in a natural state across the watershed was determined (Figure
8). Results indicate that the SMRW is fairly natural across all riparian buffer sizes. Across
the various buffers, 76.2 to 85.5% of LCLU classes are in a natural state (Table 16).
Disturbance within all fixed-width buffers demonstrated secondary growth as the greatest
contributor to disturbance, followed by urban, forest loss, and agriculture classes (Table
16). The estuary coastline revealed a higher proportion of urban as expected due to roads
along the coastline.
22
Figure 8. LCLU reclassified raster clipped to each of the fixed-width riparian buffers and zoomed
to Glenelg, St. Mary’s River watershed, Nova Scotia. For the estuarine coastline fixed-width buffer
see Appendix D.
23
Table 16. Proportion of LCLU that is natural and disturbed for each fixed-width riparian buffer.
Buffer type and size (m)
Type
LCLU Class
Name
CanVec
NCC
ARA
CanVec estuary
coastline
30
100
300
150
100
Natural
Natural Stand
64.95
65.06
62.64
64.32
45.11
Wetland
6.76
3.79
2.86
3.08
3.86
Barren
0.76
0.54
0.34
0.36
9.15
Shrubland
5.26
3.71
3.16
3.4
9.83
Grassland
7.77
7.56
7.23
7.38
12.67
Total (%)
85.50
80.65
76.23
78.55
80.62
Disturbed
Secondary Growth
5.73
8.82
11.92
10.29
4.48
Forest Loss
1.9
3.51
5.2
4.32
0.23
Urban
4.66
4.15
3.71
3.89
10.38
Agriculture
2.21
2.87
2.93
2.95
4.28
Total (%)
14.50
19.35
23.77
21.45
19.38
Land ownership is a key factor in developing policy recommendations and making
informed decisions regarding habitat. Within 100 m of the SMR and connected
waterbodies, 49.0% of terrestrial habitat occurs on private land and 51.0% on Crown land
(Table 17). Private land within the watershed is concentrated along the main tributaries
of the SMR.
Examining the distribution of LCLU classes across the various fixed-width buffers and
among land ownership demonstrated ~ 75.5 88.9% of the watershed is intact and ~
11.08 26.7% of it is disturbed within the four main riparian buffer zones (Table 18).
However, the estuarine coastline appeared more natural with ~7.2% disturbed and ~
92.8% intact, yet this is due to the presence of coastal barrens. Among all riparian buffers
analyzed, secondary growth is the greatest contributor to disturbance followed by forest
loss, urban, and agriculture classes (Table 17; Figure 8).
Table 17. Distribution of land ownership within each of the three main branches of the St. Mary’s
Watershed (Figure 2).
Sub watershed
boundaries
Crown land
(hectares)
Crown land (%)
Private land
(hectares)
Private land (%)
West St. Mary’s River
12,313
64.0
6,910
36.0
North St. Mary’s River
4,397
38.0
7,242
62.0
East St. Mary’s River
604
19.0
2,590
81.0
Mainstem and estuary
4,588
53.0
4,069
47.0
Total
21,902
51.0
20,811
49.0
24
Table 18. Distribution of LCLU activities on crown lands (“C”) compared to private lands (“P”)
within each fixed-width riparian buffers.
30
100
150
300
Ownership
C
P
C
P
C
P
C
P
Natural
Natural Stand
70.43
65.23
69.31
63.18
68.00
62.17
65.61
60.47
Wetland
6.86
4.45
4.22
2.61
3.50
2.06
3.39
2.02
Barren
0.14
0.99
0.10
0.82
0.00
0.55
0.07
0.55
Shrubland
2.77
5.61
2.07
4.71
2.00
4.35
2.07
4.10
Grassland
8.71
6.79
8.43
6.77
8.4
6.40
8.25
6.13
Total (%)
88.9
83.1
84.1
78.1
81.9
75.5
79.4
73.3
Disturbed
Secondary
Growth
6.1
5.95
9.72
8.28
11.2
9.66
12.78
11.30
Forest Loss
1.68
2.22
3.24
3.74
4
4.82
4.84
5.53
Urban
2.66
4.89
2.35
4.51
2.3
4.3
2.4
4.02
Agriculture
0.64
3.85
0.55
5.38
0.57
5.7
0.57
5.87
Total (%)
11.1
16.9
15.9
21.9
18.1
24.5
20.6
26.7
Total (km2)
45.05
40.28
150.5
132.4
411.1
332.57
214.26
186.79
OGF stands identified on Crown land under the Nova Scotia Old Growth Forest Policy
(Natural Resources and Renewables 2022) are granted proactive protection from timber
harvesting and other disturbances (Figure 9). In the St. Mary’s River watershed, 1,089.86
hectares of protected OGF overlap with a 100 m riparian buffer on Crown land, and
239.37 hectares of OGF are located in 100 m buffers within protected areas (Table 19).
These stands may include any forest types that meet the OGF minimum tree age
requirements present in s. 3 of the policy (e.g., 140 years for Tolerant Hardwood stands)
or forested areas that have yet to develop into old-growth forest but are expected to do
so with the passage of time” and are known as OGF restoration opportunity areas (Natural
Resources and Renewables 2022).
25
Figure 9. Land ownership as well as conservation and protected areas within the St. Mary’s River
watershed, Nova Scotia.
Table 19. Total area (in hectares) of OGF in the St. Mary’s River watershed that is protected
under the Nova Scotia Old Growth Forest Policy, split by land ownership.
Layer
Land
Type
Total
Area
(hectares)
100m
Riparian
Fixed-width
Buffer
(hectares)
OGF restoration opportunities outside of protected areas
Crown
3,469.13
1,089.86
OGF restoration opportunities inside of protected areas
Crown
942.34
239.37
Total OGF in the St. Mary’s River watershed
4,411.17
1,329.23
3.3 Weighting and Ranking
Three versions of ranking and weighting the LCLU classes, derived from the reclassified
raster (Figure 2), for “intactness” resulted in similar patterns (Figures 10,11 and 12). The
greatest amount of natural stand occurred in the north and western regions of the
watershed where Crown land was present, and most disturbance occurring on private
26
land, particularly in the southern and eastern regions of the watershed (Figures 10,11 and
12). Results of the weighting and ranking for all maps demonstrated trends where sub
watershed IDs 16, 54, 55, 57, 59, 60 and 62 would benefit from proactive protection
(Figures 10, 11, and 12). Meanwhile discrepancies in ranking and grouping of classes
into 8 or 9 classes highlighted additional sub watershed IDs 67,68,72,75, and 77 as
potential areas that may benefit from proactive protection. Sub watersheds identified as
potential areas that may benefit from proactive protection occurred predominately on
private land while areas that are highly natural across all weighted maps occurred on
areas of Crown land.
Grouping wetland with natural stand and performing no grouping of LCLU classes
resulted in an increase in less natural sub watersheds, whereas grouping LCLU classes
into only three classes (highly natural, regeneration likely, regeneration unlikely), resulted
in more natural sub watersheds identified. The discrepancy between grouping and
weighting maps occurred for several reasons. First, maps were calculated using a 100 m
riparian fixed-width buffer and as a result may only cover a significantly small portion of
the actual sub watershed. Therefore, there may be instances where a sub watershed is
indeed natural but appears less natural because it has a low proportion of land cover
compared to other sub watersheds. For example, sub watershed ID 68 covers
approximately 29.38 km2. Using the 100 m buffer the LCLU class covers a total of 6.08
km2 (Appendix D). In the resulting weighted maps sub watershed ID 68 appears less
natural in both the weighted and ranked 1-8 classes (Figure 10) and the non-grouped 1-
9 classes (Figure 11) compared to the 1-3 grouped classes (Figure 12) portrayed this sub
watershed as ‘Highly Natural’.
Second, the nature of the symbolized scale from 0-100 may demonstrate areas that are
natural as less natural due to a lower proportion of LCLU class obtained from the buffer
and size of the sub watershed. For example, in sub watershed ID 68 the weighted and
ranked 1-9 version resulted in the greatest proportion of LCLU being natural stand
(14.8%) followed by shrubland (3.6%), secondary growth (1.7%), wetland (1.6%),
grassland (1.5), urban (0.2%), barren (0.8%), forest loss (0.7%), and agriculture (0.1%).
However due to the low proportion of LCLU cover, the resulting map portrayed sub
watershed ID 68 as less natural as the greatest LCLU class is 14.8%. Similarly, in sub
watershed ID 68 of version 2 where classes are ranked from 1-8 natural stand & wetland
obtained a greater proportion of highly natural category (27.8%), followed by shrubland
(3.6%), secondary growth (1.7%), wetland (1.6%), grassland (1.5), urban (0.2%), barren
(0.8%), forest loss (0.7%), and agriculture (0.1%). However, weighting LCLU classes into
three groups highly natural (54.7%), regeneration likely (3.07%), and regeneration
unlikely (0.2%) demonstrated a higher proportion of highly natural for sub watershed ID
68 compared to grouping LCLU classes into 8 classes or no grouping at all. Therefore,
weighting results may be natural yet there may not be enough proportion of natural stand
to be reflected on the 0-100 scale depending on the grouping being analyzed.
Finally, differences may occur between maps as the nature of the classes grouped and
the weighting applied to each of the individual LCLU classes themselves slightly differ.
For example, grouping wetlands and natural stands together (version 1, 8 classes) as
one layer combined the sum of natural stand class area and the sum of the wetland class
areas together (Table 8, see methods 3.3). In version 2 (no grouping of classes, 1-9) the
27
natural stand and wetland classes were themselves separate layers that were each
weighted individually compared to version 1 where natural stand and wetland were
combined into one layer and weighted. In version 3, the classes grouped into three
categories appeared more natural compared to version 1 as a greater number of LCLU
classes (n=4) were combined in the highly natural class (Tables 7,8,9, methods section
2.6). For example, in version 3, highly natural combined the area sums for natural stand,
wetland, shrubland, and grassland classes for a weighted sum of 0.64%. Version 1
combined only natural stand and wetland and weighted grassland and shrubland
separately (3 layers and 3 weights to sum to 0.64%). Therefore, while the weights applied
to each version of weighting and ranking for natural stand each sum to 0.64% the input
layers being weighted themselves are slightly different (e.g. version 1 and 3: natural stand
and wetland together as a single layer and grassland and shrubland as two separately
weighted individual layers compared to version 2: where all layers were independent and
weighted as such).
Summarizing natural and disturbed areas using a 100 m riparian fixed-width buffer and a
weighted sum and ranking method for each sub watershed can appear misleading (less
natural) across various grouping of LCLU classes. If the weighted maps are to be used in
management decisions, the size of the sub watershed (area), the sum of the LCLU area,
the proportion of each LCLU class, and the size of the riparian buffer should all be
considered among and between sub watersheds (see Appendix D). Using the values
outlined in Appendix D for each version of ranking and assigned weighting, each map can
be validated.
28
Figure 10. Results of the weighted sum and ranking 1-9 using the classified land cover land use
classified raster clipped to and calculated based on a 100 m riparian fixed-width buffer. Results
determined the proportion of each land cover class within the fixed buffer and each sub
watershed. Classes were ranked from 1-9 where 1 was the most natural and 9 was least natural.
Results were based on a weighting of the ranking criteria from 0-100 (see methods section 2.6,
Table 8). Numbers within the map represent the sub watershed ID.
29
Figure 11. Results of the weighted sum and ranking 1-8 using the classified land cover land use
classified raster clipped to and calculated based on a 100 m riparian fixed-width buffer. Results
determined the proportion of each land cover class within the fixed buffer and each sub
watershed. Classes were ranked from 1-8: 1 was the most natural and 8 was least natural. Results
were based on a weighting of the ranking criteria from 0-100 (see methods section 2.6, Table 9).
Numbers within the map represent the sub watershed ID.
30
Figure 12. Results of the weighted sum and ranking 1-3 using the classified land cover land use
classified raster clipped to and calculated based on a 100 m riparian fixed-width buffer. Results
determined the proportion of each land cover class within the fixed buffer and each sub
watershed. Classes were ranked from 1-3: 1 was the most natural and 3 was least natural. Results
were based on a weighting of the ranking criteria from 0-100 (see methods section 2.6, Table 10).
Numbers within the map represent the sub watershed ID.
Using a 100 m riparian buffer, the class with the greatest proportion of area (m2) within a
sub watershed among all classes identified in the LCLU map and using each ranking
system was determined. Natural stand was the class with the greatest proportion of LCLU
among all ranking systems using a 100 m riparian buffer for each sub watershed
(Appendix D).
31
4. Discussion & Conclusion
Human land-use activities that occur near watercourses and waterbodies play a critical
role in the overall health, function, and biodiversity of the aquatic environment (Collison
2021). Riparian zones serve as an interface between aquatic and terrestrial habitats and
are sensitive to LCLU changes. LCLU activities that occur within the riparian zone may
have negative cascading effects on aquatic habitats and the species that live within them.
Understanding the types of LCLU that occur within an area provides a means to assess
current riparian buffer regulations, identify potential risks to species and associated
habitat, and identify areas that may benefit from proactive protection.
The results of this work demonstrate that the St. Mary’s River watershed is very natural.
Very little disturbance was found throughout the watershed from urban and agricultural
classes. Natural stand was found to be the dominant class occurring in the north and
western regions where crown land was present. Some areas of secondary growth on
private land occurred near agriculture and are likely regenerating crops. Within the St.
Mary's River watershed, the majority of disturbance occurred from the forest loss LCLU
class due to silvicultural practices, clear cuts, and partial clear cuts, as observed by the
NSFI layer. Furthermore, agriculture classes occurred more on private land that was
located along the main stem and tributaries within the watershed. Forest loss and
secondary growth occurred more in areas of crown land compared to private land. We
hypothesize that this may be a result of tree plantations or harvesting for forestry purposes
as observed in the NSFI and CLDFHD layers.
Validation points used for LCLU raster 1 did not include the detailed LCLU information
contained in the NS Forest Inventory, and instead were assembled from the results of
other classified land cover maps produced using different satellites/sensors. A limitation
of this approach was the different spatial resolution and image capture dates associated
with the validation land cover products in comparison with the Sentinel-2A satellite image
used in this study. Additional uncertainty should be considered when using and reviewing
the results generated in the study as validation points used for training and testing
required significant re-classification and verification against the RGB Sentinel-2A image
before classification. Reclassification of validation points was time intensive given the
considerable number of points used. Technicians may consider further reducing the
number of validation points used (e.g., minimum of 10 - 30 times the number of bands
depending on the classifier method) to decrease processing time and compare
accuracies (Li et al. 2014).
The 2019 classified raster achieved a lower overall accuracy (84.4%) compared to both
the AAFC in 2021 (87.94%) and NRCan classified Land Cover 2020 (86.9%) maps.
However, the reclassified raster achieved a higher accuracy (90.0%) compared to the
NRCan and AAFC classifications. It is possible that the initial classified raster (LCLU
raster 1) obtained a slightly lower accuracy compared to the AAFC and NRCan due to
the large number of validation points used for training and testing. While we try to examine
and validate every point it is possible that not every single point was correctly classified
as there were over 26,000 points. This was observed in the accuracy assessment where
some areas identified as urban (secondary roads) yet the image demonstrated forest had
32
started to regenerate and the secondary roads had become less easily identified causing
decreases in accuracies. These errors were often misclassified as forested or shrubland
LCLU classes. Consequently, some inaccuracies occurred due to too few numbers of
validation points for some classes such as shrubland, barren, grassland, and wetland.
Shrubland and grassland were often misclassified with agriculture while barren was often
misclassified with roads. Barren and urban classes are likely to be misclassified as they
are often flat and hard bare earth surfaces that are highly reflective. Similarly, pixel
reflectance values of low vegetation associated with shrubland and grassland are very
similar to those associated with agriculture. While the final reclassified raster included
agriculture LCLU on crown land (0.57% of crown land LCLU within 300 m fixed width
buffer), one can assume these are misclassified pixels as agricultural land use is likely
limited to private land. Furthermore, errors were identified in select areas of the watershed
classified as agriculture following discussions with the provincial Department of
Agriculture (NSDA). Future iterations of LCLU classification using this methodology could
be improved by accounting for land ownership type (i.e., crown versus private), and by
incorporating more field validation data on these specific LCLU types.
Validation points used in the reclassified raster (LCLU 2) incorporated more detailed
information on LCLU offered through the NS Forest Inventory dataset, which may have
increased accuracy. However, the majority of the forest inventory data available in the St.
Mary’s River watershed is based on aerial photography collected between 2007 and
2008, compared with the more recent 2019 satellite image used for this study. Obtaining
more recent field data to validate the analysis would provide more descriptive knowledge
on forest loss and disturbance within a region. Currently the NSFI data is the only
validation data that provides descriptive attributes on forest loss such as clear cut, partial
clear cut, burn, wind throw, etc. Other available data layers such as the Hansen Global
Forest Loss (Hansen et al. 2013) and the CLDFHD (used in this study) only provide
information on the year that forest loss occurred. By identifying the reason for forest loss,
one can obtain informative results on the types of LCLU activities and potential risks to
riparian and aquatic habitats. This knowledge may identify or rank areas based on
sensitivity to disturbance in addition to ranking LCLU class alone, and identify whether
change to landscapes are a result of natural or anthropogenic impacts. In Nova Scotia,
natural events such as Hurricane Dorian (2019) and post-tropical storm Fiona (2022)
have left drastic changes in forested habitat and likely riparian habitat. These changes
may only be observable by comparing satellite images and conducting change detection
analysis. Obtaining recent validation data and comparing several satellite images over
time together would contribute to accurate validation for classified LCLU maps,
particularly for forest loss and secondary growth. While limited on the ground field
validation data was available for this work, the LCLU maps generated by this study
provide relatively high resolution (10 m) baseline information required to examine LCLU
within riparian habitat to identify areas where disturbance and/or alterations to riparian
habitat may exist.
To date, the greatest resolution of classified LCLU maps across Canada is 30 m and is
generated on a yearly basis (i.e., AAFC dataset) or every 5 years (i.e., NRCan dataset).
However, LCLU products available may not adequately capture small scale changes due
to its coarse resolution. High resolution imagery offers a greater spatial detail with more
33
bands and therefore greater quantities of information that can be used to increase the
accuracy of classification compared to coarser resolution images such as those obtained
by Landsat (Fauvel et al. 2012; Lacelle & Shi 2021; Kamenova & Dimitrov 2021; Parker
& Lee 2016). This study contributed robust and simple methods to generate fine-scale
(10 m) classified LCLU maps that may be applied to examine riparian analysis on a
watershed-to-watershed scale and may be used to classify other watersheds in the
Maritimes region. Using these methods, technicians and researchers may generate fine
scale LCLU maps as they see fit (e.g., shorter temporal scales, between seasons,
between months, or on specific dates). Methods outlined in this study are replicable and
may be refined by researchers to examine LCLU changes to habitats. Geoprocessing
tools used to classify this watershed are robust enough to be automated using Python
scripts, with the exception of image download, cloud removal, and validating training and
testing points before classification. Moreover, scripts generated in RStudio can aid to
decrease, assemble, and combine multiple validation datasets together and automate
error matrices to determine the accuracy of classifications. Overall, this study provided a
comparative analysis on various fixed-width buffer sizes within the St. Mary’s River
watershed to complement literature reviews and examine LCLU classes against current
regulations in Nova Scotia.
Riparian LCLU analysis can help fill knowledge gaps required to identify sensitive areas
and determine the quality of aquatic habitat. Similar riparian spatial analysis and LCLU
classified maps have been used to characterize fish habitat (Bachiller-Jareno et al. 2019;
Jones et al. 2006; Kuiper et al. 2022; Tompalski et al. 2017), conduct threat assessments
of riparian areas (Coleshill & Watt 2017), determine suitable locations for riparian buffers
by sub watershed (Budlong 2004), conduct assessments for riparian buffer zone by
stream order among basins and ecoregions (Mary-Lauyé et al. 2022), assess the health
and vulnerability of watersheds (Roth et al., 2020) and examine exposure based
assessments and vulnerability from land-use threats (France & Pardy 2018).
This study faced some limitations including obtaining complete coverage of the study area
using the most recent satellite imagery data. To complete the analysis using the most up
to date image data, we attempted to use open-source high resolution satellite images
captured post-Hurricane Fiona (after September 23, 2022). However, the Sentinel 2
sensor did not consistently capture full coverage of the pre-planned tile footprint. As a
result, we could not find usable images captured post-Hurricane Fiona that could be
mosaicked together. The 2019 satellite image still provides a means to classify LCLU
within the SMRW where this data together with forest disturbance and regeneration
(secondary growth), has not previously been mapped. We recommend that our study is
replicated using high resolution satellite imagery that has been captured since Hurricane
Fiona and using images captured within the same season. Future work may consider
purchasing high resolution imagery with a more frequent re-visit time such as World View
2, World View 3, or Pléiades. Researchers may also consider the use of other image
collection applications such as UAVs, and aerial photography.
One limitation for this study exists within the validation data. Validation points used to
guide classification in this study were not conducted using true on the ground field
validation and therefore there is a need to build additional uncertainty into the
interpretation of these analyses. While classification products are readily available from
34
NRCan, AAFC, and NCC, validation data used by these organizations to generate these
products are not readily accessible. In many cases validation sources used by
organizations are created from scanning satellite imagery or in the case of the AAFC,
field validation data is collected. Obtaining access to on the ground field validation would
be useful to improve and validate classification accuracy as having access to only final
classified products means that our accuracy is limited to the accuracy achieved by input
data and therefore requires careful reclassification. This limitation restricted our ability in
automating classification across other watersheds. This limitation, however, can be
overcome by cleaning and reclassifying validation data. Results of this study are still valid
as they provide an overall high accuracy compared to coarser Landsat derived products
generated by the AAFC and NRCan. Although data may not be true on the ground field
validation, validation points used in training and testing were assessed individually
against the RGB satellite image prior to classification and therefore should accurately
validate and reflect LCLU classes within the St. Mary's River watershed.
Another limitation of this study exists within the classifier method. Due to the nature of
using a pixel-based supervised SVM classification approach, the resulting LCLU maps
produced a noisy result where individual or small groups of pixels are scattered or
appearing out of place. For example, several areas of the agriculture class may appear
outside of a clearly defined crop outline. LCLU classification may be improved using an
unsupervised or supervised Object Based Image Analysis (OBIA) method to classify data.
Previous studies have found that applying an OBIA approach to original bands, without
band indexes or ancillary data, yields greater (87-88%) overall accuracy using Sentinel-
2 compared to Landsat 8 (Sánchez-Espinosa & Schröder 2019). Alternatively,
researchers may need to develop and refine object and segmentation parameters to
reflect threshold sizes of the classes they wish to identify. Using the OBIA method
parameters may need to be refined at an individual watershed scale.
Future work may consider using remote sensing software such as PCI Geomatica, ENVI,
or ECognition where lookup tables can be generated to drive pixel-based image
classification. Generating lookup tables can help to automate classification and decrease
processing time required to classify large areas and examine landscape changes within
shorter time frames. Platforms such as Google Earth Engine handle multi-temporal
mosaics, cloud cover, and footprint cover well, however, the St. Mary's River watershed
was too large for this platform.
The classified LCLU map generated in this project may be used in species distribution
and/or habitat suitability modelling by providing information on riparian cover that can help
inform shading, bank stability, filtration and infiltration within the buffer zone. This work
could be further developed through the incorporation of LiDAR-derived canopy height
models to derive metrics of canopy shading in riparian areas (Kuiper et al. 2022;
Tompalski et al. 2017). Furthermore, imagery paired with the Nova Scotia Forest
Inventory layer can be used to further estimate and examine the percent of crown cover
for broad leaf tree species in examining shading which is critical for maintaining natural
variation in water temperature. Information on forest type and or height can also be used
to calculate a riparian score and help fill data gaps that exist within regulatory monitoring
such as the Canadian Aquatic Biomonitoring Network (CABIN) and may be embedded
within a suitability and/or species distribution model.
35
There is no one-size fits all approach in designing riparian fixed-width buffers. Rideout et
al. (2012) suggest considering what ecological functions or habitat components each
parameter contributes to or is influenced by, and where within the riparian zone is the
most sensitive to disturbance. Standardized methods to classify habitats at the watershed
or regional scale would be useful for consistency and comparing accuracies, yet this
outcome requires access to recent on the ground field validation, a standardized method
to reclassify validation data, and/or a standard method to generate validation data across
images. Furthermore, reviewing historical or longer-term changes across buffer sizes may
help to better understand and compare the performance of different buffer width
management measures and the impacts to aquatic ecosystems (Dey et al. 2021). This
report provided an example of robust methods to classify satellite imagery and analyze
results according to various fixed-width buffers, land ownership, and identify areas that
may benefit from proactive protection. LCLU information can help direct restoration
planning by identifying areas that are more degraded but where regeneration is possible.
Information of LCLU activities should be coupled with biological information on critical
areas for aquatic species to help assess risk and inform protection measures. The
methodology and results outlined in this report can help to inform riparian management
approaches to protect fish and fish habitat in the St. Mary’s River watershed, and beyond.
With continued data collection and analyses, this work can also set the foundation to
support long-term scientific and compliance monitoring.
36
Acknowledgements
The authors would like to thank Sarah Kingsbury, Gary Pardy, Tyler Veinot and Kasia
Rozalska for providing detailed reviews, advice, and support throughout this project.
37
References
Albertson, L. K., Ouellet, V., & Daniels, M. D. (2018). Impacts of stream riparian buffer
land use on water temperature and food availability for fish. Journal of
Freshwater Ecology, 33(1), 195-210.
https://doi.org/10.1080/02705060.2017.1422558
Agriculture & Agri-food Canada. (2021). ISO 19131 Annual Crop Inventory Data
Product Specifications (ISO 19131). Revision A. pp 1-30.
https://Agriculture.canada.ca/atlas/data_donnees/annualCropInventory/supportdo
cument_documentdesupport/en/ISO%2019131_AAFC_Annual_Crop_Inventory_
Data_Product_Specifications.pdf
Aresnault, M., O’Sullivan, A.M., Ogilvie, J., Gillis, C.A, Tommi, L., and Curry, R.A.
(2021). Remote sensing framework details riverscape connectivity fragmentation
and fish passability in a forested landscape. Journal of Ecohydraulics. 1-13.
https://doi.org/10.1080/24705357.2022.2040388
Bachiller-Jareno, N., Hutchins, M.G., Bowes, M.J., Charlton, M.B., & Orr, H.G. (2019). A
novel application of remote sensing for modelling impacts of tree shading on
water quality. Journal or Environmental Management. 230. 33-42.
https://doi.org/10.1016/j.jenvman.2018.09.037
Bowlby, H.D., Horsman, T., Mitchell, S.C., & Gibson, A.J.F. (2014). Recovery Potential
Assessment for Southern Upland Atlantic Salmon: Habitat Requirements and
Availability, Threats to Populations, and Feasibility of Habitat Restoration. DFO
Can. Sci. Advis. Sec. Res. Doc. 2013/006. vi + 155 p. https://waves-
vagues.dfompo.gc.ca/Library/359664.pdf.
Budlong R.C.(2004). The Use of Spatial Data in Creating a Riparian Buffer Suitability
Model: Whitewater River Watershed, Minnesota. United Sates Fish and Wildlife
Service. Department of Resource Analysis. St. Mary’s University of Minnesota,
Winona. 1-11. http://gis.smumn.edu/gradprojects/budlongb.pdf
Caskenette, A., Durhack, T., Hnytka, S., Kovachik, C., & Enders, E. (2021). Evidence of
effect of riparian attributes on listed freshwater fishes and mussels and their
aquatic critical habitat: a systematic map protocol. Environmental Evidence, 10
(1), 1-10. http://dx.doi.org/10.1186/s13750-021-00231-1
Cole, L. J., Stockan, J., & Helliwell, R. (2020). Managing riparian buffer strips to
optimise ecosystem services: A review. Agriculture, Ecosystems & Environment,
296, 106891. https://doi.org/10.1016/j.agee.2020.106891.
Coleshill, J. & Watt, G. (2017). Threat Assessment of Riparian Areas in the Kettle River
Watershed. Granby Wilderness Society. 1-94. http://kettleriver.ca/wp-
content/uploads/2017/05/KRWTA-Final-April-2017-1.pdf
38
Collison, B. R. (2021). Riparian management Considerations and Recommendations for
Ecologically Significant Areas: The St. Mary’s River Watershed. Lecture.
Fisheries and Oceans Canada- Maritimes Region, Regulatory Review Meeting.
July 14 2021.
Collison, B. R., & Gromack, A. G. (2022). Importance of riparian zone management for
freshwater fish and fish habitat protection: analysis and recommendations in
Nova Scotia, Canada. Can. Tech. Rep. Fish. Aquat. Sci. 3475: viii + 71 p.
https://publications.gc.ca/collections/collection_2022/mpo-dfo/Fs97-6-3475-
eng.pdf
COSEWIC. (2010). COSEWIC assessment and status report on the Atlantic Salmon
Salmo salar (Nunavik population, Labrador population, Northeast Newfoundland
population, South Newfoundland population, Southwest Newfoundland
population, Northwest Newfoundland population, Quebec Eastern North Shore
population, Quebec Western North Shore population, Anticosti Island population,
Inner St. Lawrence population, Lake Ontario population, Gaspé-Southern Gulf of
St. Lawrence population, Eastern Cape Breton population, Nova Scotia Southern
Upland population, Inner Bay of Fundy population, Outer Bay of Fundy
population) in Canada. Committee on the Status of Endangered Wildlife in
Canada. Ottawa. xlvii + 136 pp. www.sararegistry.gc.ca/status/status_e.cfm.
COSEWIC.(2012). Assessment and status report on the American eel Anguilla rostrata
in Canada. Committee on the Status of Endangered Wildlife in Canada. xlvii +
109 pp. www.sararegistry.gc.ca/status/status_e.cfm.
Cunningham, D. S., Braun, D. C., Moore, J. W., & Martens, A. M. (2023). Forestry
influences on salmonid habitat in the North Thompson River watershed, British
Columbia. Canadian Journal of Fisheries and Aquatic Sciences.
https://doi.org/10.1139/cjfas-2022-0255.
Daryaei, A., Sohrabi, H., Atzberger, C., & Immitzer, M. (2020). Fine-scale detection of
vegetation in semi-arid mountainous areas with focus on riparian landscapes
using Sentinel-2 and UAV data. Computers and Electronics in Agriculture, 177
(May), 105686. https://doi.org/10.1016/j.compag.2020.105686
De Sosa, L. L., Glanville, H. C., Marshall, M. R., Abood, S. A., Williams, A. P., & Jones,
D. L. (2018). Delineating and mapping riparian areas for ecosystem service
assessment. Ecohydrology, 11(2), e1928. https://doi.org/10.1002/eco.1928
Dey, C. J., Rego, A. I., Bradford, M. J., Clarke, K. D., McKercher, K., Mochnacz, N. J.,
... & Koops, M. A. (2021). Research priorities for the management of freshwater
fish habitat in Canada. Canadian Journal of Fisheries and Aquatic Sciences.
https://doi.org/10.1139/cjfas-2021-0002.
Eskandari, S., & Pourghasemi, H. R. (2022). Assessing and mapping distribution, area,
39
and density of riparian forests in southern Iran using Sentinel-2A, Google earth,
and field data. Environmental Science and Pollution Research, 0123456789.
https://doi.org/10.1007/s11356-022-21478-2
European Space Agency. (2012). Retrieved March 2, 2023, from RADARSAT
Constellation. https://www.eoportal.org/satellite-missions/rcm
European Space Agency. (2022a). Landsat Series. Retrieved February 28, 2023, from
https://earth.esa.int/eogateway/missions/landsat/description
European Space Agency. (2022b). Sentinel-2. Retrieved February 28, 2023, from
https://www.esa.int/Applications/Observing_the_Earth/Copernicus/Sentinel-2
Fauvel, M., Tarabalka, Y., Benediktsson, J.A., Chanussot, J., Tilton, J.C.,(2012).
Advances in Spectral-Spatial Classification of Hyperspectral Images.
IEEE.101(3). 652-675. https://doi.org/10.1109/JPROC.2012.2197589
Fisheries Act, R.S.C. 1985, c. F-14. Fisheries Act. https://www.laws-
lois.justice.gc.ca/eng/acts/F-14/
Fisheries and Oceans Canada. (2020). Guidance on the identification of Critical Habitat
in the riparian zone for freshwater species at risk. DFO Can. Sci. Advis. Sec. Sci.
Advis. Rep. 2020/040. 26pp. https://waves-vagues.dfo-
mpo.gc.ca/Library/40940469.pdf.
France, R.L., & Pardy, G. (2018). Spatially-Explicit, Exposure-Based Assessment of
Surface Water Vulnerability from Land Use Threats for Time-Efficient and Cost-
Effective Watershed Development Planning. Journal of Geoscience and
Environment Protection. (6). 35-55. https://doi.org/10.4236/gep.2018.66003
Fuller, M. R., Leinenbach, P., Detenbeck, N. E., Labiosa, R., & Isaak, D. J. (2022).
Riparian vegetation shade restoration and loss effects on recent and future
stream temperatures. Restoration Ecology, 30(7), e13626.
https://doi.org/10.1111/rec.13626.
Furuya, D. E. G., Aguiar, J. A. F., Estrabis, N. V., Pinheiro, M. M. F., Furuya, M. T. G.,
Pereira, D. R., Gonçalves, W. N., Liesenberg, V., Li, J., Junior, J. M., Osco, L. P.,
& Ramos, A. P. M. (2020). A machine learning approach for mapping forest
vegetation in riparian zones in an atlantic biome environment using sentinel-2
imagery. Remote Sensing, 12(24), 116. https://doi.org/10.3390/rs12244086
Government of Canada. (2018). Brook Floater (Alasmidonta varicose): management
plan. Management Plan for the Brook Floater (Alasmidonta varicosa) in Canada -
Canada.ca. https://www.canada.ca/en/environment-climate-
change/services/species-risk-public-registry/management-plans/brook-floater-
final.html
40
Government of Canada. (2021). St. Mary’s River. Geographical Names Board of
Canada. Place names - St. Marys River (nrcan.gc.ca).
https://geonames.nrcan.gc.ca/search-place-names/search
Government of Canada. (2022a). Critical Habitat of Species at Risk. Critical Habitat of
Species at Risk - Open Government Portal (canada.ca).
https://open.canada.ca/data/en/dataset/47caa405-be2b-4e9e-8f53-
c478ade2ca74
Government of Canada. (2022b). 2020 Land Cover of Canada. 2020 Land Cover of
Canada - Open Government Portal.
https://open.canada.ca/data/en/dataset/ee1580ab-a23d-4f86-a09b-
79763677eb47
Government of Canada. (2023). Species at Risk Act Implementation Guide for
Recovery Practitioners. Critical habitat identification toolbox: Species at Risk Act
guidance. https://www.canada.ca/en/environment-climate-
change/services/species-risk-public-registry/critical-habitat-
descriptions/identification-toolbox-guidance.html
Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S. A., Tyukavina,
A., et al. (2013). High-Resolution Global Maps of 21st-Century Forest Cover
Change. Science, 342, 850853. doi:10.1126/science.1244693
Jones, J. L. (2006). Side channel mapping and fish habitat suitability analysis using lidar
topography and orthophotography. Photogrammetric Engineering and Remote
Sensing, 72(11), 12021206. https://www.jstor.org/stable/26266006
Kamenova, I., & Dimitrov, P. (2021). Evaluation of sentinel-2 vegetation indices for
prediction of LAI, fAPAR and fCover of winter wheat in Bulgaria. Remote
Sensing. 54(1) pp. 89-90. https://doi.org/10.1080/22797254.2020.1839359
Kupier, S.D., Coops, N.C, Tomalski, P., Hinch, S.G., Nonia, A., White, J.C., Hamilton,
J., Davis, D.J.( 2022). Characterizing stream morphological features important for
fish habitat using airborne laser scanning data. Remote Sensing Environment
(272)112948. https://doi.org/10.1016/j.rse.2022.112948
Lacelle, C. & Shi, Z. (2021). Spectral detail versus spatial detail: A land cover
classification for Northeastern Georgian Bay using Sentinel-2 multispectral
imagery. University of Guelph. Department of Geography, Environment, and
Geomatics. https://geg.uoguelph.ca/sites/default/files/G6_AODA_W21.pdf
Lambin, E. F., Turner, B. L., Geist, H. J., Agbola, S. B., Angelsen, A., Bruce, J. W., ... &
Xu, J. (2001). The causes of land-use and land-cover change: moving beyond
the myths. Global environmental change, 11(4), 261-269.
https://doi.org/10.1016/S0959-3780(01)00007-3.
41
Lahey, W. (2018). The shores of watercourses (riparian areas) - An independent review
of forest practices in Nova Scotia, addendum.
https://novascotia.ca/natr/forestry/forest_review/FP_Addendum.pdf.
Li, C., Wang, J., Hu, L., & Gong, P. (2014). Comparison of Classification Algorithms and
Training Sample Sizes in Urban Land Classification with Landsat Thematic
Mapper Imagery. Remote Sensing, 6, pp. 964-983. doi:10.3390/rs6020964
Lind, L., Hasselquist, E. M., & Laudon, H. (2019). Towards ecologically functional
riparian zones: A meta-analysis to develop guidelines for protecting ecosystem
functions and biodiversity in agricultural landscapes. Journal of Environmental
Management, 249, 109391. https://doi.org/10.1016/j.jenvman.2019.109391.
Mary-Lauyé, A.L.,González, N.G., Somma, A., Silva, I., Lucas, C.M. (2022). Baseline
assessment of the hydrological network and land use in riparian buffers of
Pampean streams of Urugauy. Environ. Monit. Assess. 195(80).1-23.
https://doi.org/10.1007/s10661-022-10684-7
Natural Resources and Renewables. (2022). An Old-Growth Forest Policy for Nova
Scotia. 1-17.
https://novascotia.ca/natr/forestry/programs/ecosystems/oldgrowth.asp#:~:text=O
nce%20a%20hallmark%20of%20the%20Acadian%20Forest%2C%20old,potentia
l%20old-growth%20forest%20are%20in%20legally%20protected%20areas
Noseworthy & Beckley. (2020). Borealization of the New England-Acadian Forest: a
review of the evidence. Environmental Reviews. 28(3). https://doi.org/10.1139/er-
2019-0068
Nussey, P. & Noseworthy, J. (2020). The Active River Area for the Northern
Appalachian Acadian Region of Canada. Nature Conservancy of Canada.
https://2c1forest.databasin.org/datasets/0ce6df639f504fa9931a2cfe5d100d1b/
Parker, J., & Lee, J., (2016). Detection of landuse/landcover changes using remotely-
sensed data. Journal of Forestry Research. 27(6), pp. 1343-1350.
https://doi.org/10.1007/s11676-016-0270-x
Piedelobo, L., Taramelli, A., Schiavon, E., Valentini, E., Molina, J. L., Xuan, A. N., &
González-Aguilera, D. (2019). Assessment of green infrastructure in Riparian
zones using copernicus programme. Remote Sensing, 11(24).
https://doi.org/10.3390/rs11242967
Phiri, D., Simwanda, M., Salekin, S., Nyirenda, V. incen. R., Murayama, Y., &
Ranagalage, M. (2020). remote sensing Sentinel-2 Data for Land Cover / Use
Mapping : A Review. Remote Sensing, 12(2291), 135.
https://doi.org/https://doi.org/10.3390/rs12142291
42
Richardson, J. S., Naiman, R. J., & Bisson, P. A. (2012). How did fixed-width buffers
become standard practice for protecting freshwaters and their riparian areas from
forest harvest practices? Freshwater Science, 31(1), 232-238.
http://dx.doi.org/10.1899/11-031.1
Rideout, E. (2012). Setbacks and vegetated buffers in Nova Scotia. Nova Scotia
Environment, Canada: Hydrologic Systems Research Group.
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.712.407&rep=rep1&typ
e=pdf
Riis, T., Kelly-Quinn, M., Aguiar, F. C., Manolaki, P., Bruno, D., Bejarano, M. D., ... &
Dufour, S. (2020). Global overview of ecosystem services provided by riparian
vegetation. BioScience, 70(6), 501-514. https://doi.org/10.1093/biosci/biaa041
Roth, N., Wharton, C., Pickard, B., Sarkar, S., Lincoln, A.R. (2020). Chesapeake
Healthy Watersheds Assessment: Assessing the Health and Vulnerability of
Healthy Watershed within the Chesapeake Bay Watershed. Tetra Tech.
https://d18lev1ok5leia.cloudfront.net/chesapeakebay/chesapeake_healthy_water
sheds_assessment_report.pdf
Rusnák, M., goga, T., Michaleje, L., Michalková, M.S., Máčka, Z., Bertalan, L, &
Kidová, A. (2022). Remote Sensing of Riparian Ecosystems. Remote Sens.
14(2645). https://doi.org/10.3390/rs14112645
Sánchez-Espinosa, A., & Schröder, C. (2019). Land use and land cover mapping in
wetlands one step closer to the ground: Sentinel-2 versus landsat 8. Journal of
Environmental Management, 247(May), 484498.
https://doi.org/10.1016/j.jenvman.2019.06.084
Sim, J., & Wright, C. (2005). The kappa statistic in reliability studies: Use, interpretation,
and sample size requirements. Physical Therapy 85: 257-268.
https://doi.org/10.1093/ptj/85.3.257
Smokorowski, K. E., & Pratt, T. C. (2007). Effect of a change in physical structure and
cover on fish and fish habitat in freshwater ecosystemsa review and meta-
analysis. Environmental Reviews, 15, 15-41. https://doi.org/10.1139/a06-007.
Species at Risk Act, S.C. 2002, c. 29. https://www.laws-lois.justice.gc.ca/eng/acts/S-
15.3/
St. Mary’s River Association. (2019). St. Mary’s River. Retrieved March 14, 2024, from
https://www.stmarysriverassociation.com/st-marys-river.html
43
Stoffyn-Egli, P., & Duinker, P. N. (2013). An ecological approach to riparian-buffer
definition, and implications for timber harvests in Nova Scotia, Canada. Journal
of sustainable development, 6(12), 111. http://dx.doi.org/10.5539/jsd.v6n12p111.
Tiwari, T., Lundström, J., Kuglerová, L., Laudon, H., Öhman, K., & Ågren, A. M. (2016).
Cost of riparian buffer zones: A comparison of hydrologically adapted site
specific riparian buffers with traditional fixed widths. Water Resources Research,
52(2), 1056-1069. https://doi.org/10.1002/2015WR018014.
Tolkkinen, M. J., Heino, J., Ahonen, S. H., Lehosmaa, K., & Mykrä, H. (2020). Streams
and riparian forests depend on each other: A review with a special focus on
microbes. Forest Ecology and Management, 462, 117962.
https://doi.org/10.1016/j.foreco.2020.117962
Tompalski, P., Coops, N. C., White, J. C., Wulder, M. A., & Yuill, A. (2017).
Characterizing streams and riparian areas with airborne laser scanning data.
Remote Sensing of Environment, 192, 7386.
https://doi.org/10.1016/j.rse.2017.01.038
USGS and USDA, Natural Resources Conservation Service. (2013). Federal Standards
and Procedures for the National Watershed Boundary Dataset (WBD) (4 ed.):
U.S. Geological Survey Techniques and Methods 11A3, 63 p.
https://pubs.usgs.gov/tm/11/a3/pdf/tm11-a3_4ed.pdf
44
Appendix A Validation and Class Codes
Table A1. Reclassification of the validation points assembled and used to classify the St. Mary’s River watershed Nova Scotia, Canada.
Validation points were then reduced to 1000 points per class and split into training (80%) and testing (20%) datasets.
Original
dataset
Land Use
Reclass
Class/
FOR-
NON
Code
Class Name
Description by
Validation
Points Per
Class
AAFC
Agriculture
194
Nursery
Nursery
27
Agriculture
192
Sod
Sod
1,686
Agriculture
182
Blueberry
Blueberry
33,807
Agriculture
158
Soybeans
Soybeans
1,280
Agriculture
147
Corn
Corn
10,521
Agriculture
122
Pasture/forages
Periodically cultivated. Includes tame grasses and other perennial crops
such as alfalfa and clover grown alone or as mixtures for hay, pasture or
seed.
83,288
Grassland
110
Grassland
Predominantly native grasses and other herbaceous vegetation, may
include some shrubland cover.
60,082
Urban
34
Urban/developed
Land that predominantly built-up or developed and vegetation associated
with these land covers. This includes road surfaces, railway surfaces,
buildings and paved surfaces, urban areas, industrial sites, mine
structures, golf courses, etc.
175,792
Barren
30
Exposed
land/Barren
Land that is predominately non-vegetated and non-developed. Includes:
glacier, rock, sediments, burned areas, rubble, mines, other naturally
occurring non-vegetated surfaces. Excludes fallow agriculture.
129,987
NRCan
Grassland
10
Temperate or sub-
polar Grassland
Temperate or sub-polar grassland
1,015,111
Coniferous
1
Temperate or sub-
polar needleleaf
forest
Temperate or sub-polar needleleaf forest
4,320,910
Mixedwood
6
Mixed forest
Mixed forest
4,197,120
Urban
17
Urban and built-up
Urban and built-up
284,486
Wetland
14
Wetland
Wetland
24,075
45
Table A1 (continued). Reclassification of the validation points assembled and used to classify the St. Mary’s River watershed Nova
Scotia, Canada. Validation points were then reduced to 1000 points per class and split into training (80%) and testing (20%) datasets.
Original
dataset
Land Use
Reclass
Class/
FOR-
NON
Code
Class Name
Description by
Validation
Points Per
Class
Deciduous
5
Temperate or sub-
polar broadleaf
deciduous forest
Temperate or sub-polar broadleaf deciduous forest
4,835,441
Shrubland
8
Temperate or sub-
polar Shrubland
Temperate or sub-polar shrubland
1,178
Barren
16
Barren Lands
Barren Lands
667
Agriculture
15
Cropland
Cropland
136,606
NSTD
Urban
NA
Road
A vector file containing all the roads in Nova Scotia
374,066
NSFI
Natural
Stand
0
Natural Stand
Any forested stand which has not been treated silviculturally and does not
qualify under clear cut, partial cut, burn, old field, wind throw, alders,
brush or dead categories.
6,287,490
Secondary
Growth
2
Burn
Any stand that has been destroyed by fire leaving less than 25% crown
closure. In cases of partial burn, the remaining live stand is to be
categorized and not classed as burn.
174
Secondary
Growth
5
Old field
Any field that has an indication of merchantable tree species growing in
with less than 25% crown closure. All normal attributes are assigned to
existing commercial tree material as the main story.
20,685
Secondary
Growth
6
Wind throw
Any stand where more than 25% of the trees have been pushed over to
more than 45 degrees from the vertical by wind action. All normal
attributes are assigned to live tree material as the main story.
14,345
Secondary
Growth
1
Treated
Treatment not classified, an area where silviculture activity has occurred,
but the actual treatment is not identified in field data from other
Department programs. This treatment excludes stands that are defined by
other forest codes, such as plantations, Christmas trees, sugar bush, etc.
286,219
46
Table A1 (continued). Reclassification of the validation points assembled and used to classify the St. Mary’s River watershed Nova
Scotia, Canada. Validation points were then reduced to 1000 points per class and split into training (80%) and testing (20%) datasets.
Original
dataset
Land Use
Reclass
Class/
FOR-
NON
Code
Class Name
Description by
Validation
Points Per
Class
Secondary
Growth
12
Treated stand
Treatment classified-an area where silviculture activity has occurred, and
the actual treatment has been identified primarily by field data from other
Department programs. This treatment excludes stands that are defined by
other forest codes, such as plantations, Christmas trees, sugar bush etc.
204,866
Secondary
Growth
60
Clear cut
Any stand that has been completely cut and any residuals make up less
than 25% crown closure and with little or no indication of regeneration.
Site values are retained. Residual live commercial material is described
as the second story.
1,151,463
Secondary
Growth
61
Partial depletion
verified
Any stand that has been cut and residuals make up 25% or more of the
crown closure on the site. Site values are retained.
124,604
CLDFHD
Forest Loss
2011
2011
Forest harvest disturbance
78,995
Forest Loss
2012
2012
Forest harvest disturbance
88,075
Forest Loss
2013
2013
Forest harvest disturbance
106,317
Forest Loss
2014
2014
Forest harvest disturbance
78,098
Forest Loss
2015
2015
Forest harvest disturbance
97,213
Forest Loss
2016
2016
Forest harvest disturbance
123,155
Forest Loss
2017
2017
Forest harvest disturbance
134,182
Forest Loss
2018
2018
Forest harvest disturbance
135,738
Forest Loss
2019
2019
Forest harvest disturbance
95,379
Total
24,713,128
47
Appendix B Data cleaning and pre-processing
Figure B1. Data cleaning and classification process for the supervised support vector LCLU classification for the St. Mary’s River
watershed. Boxes in green were performed in Arc GIS Pro (v. 2.8), boxes in blue and light blue were performed in RStudio (v. 4.2.2).
Boxes in darker blue represent data and processes for the classified raster while boxes in light blue represent data and processes
used in the reclassified raster. Note all validation data must be checked against the desired image.
48
Appendix C Fixed-width Riparian Buffers (estuary coastline)
Figure C1. Land cover land use classification of a zoomed in version of the 100 m Fixed width riparian buffer along the estuary coastline
within the St. Mary’s River watershed. Scale 1:28,000.
49
Appendix D Proportion of LCLU Class per sub watershed ID (100 m riparian buffer)
Table D1. LCLU class distribution by sub watershed using the 100m riparian buffer zone & 9 classes.
Sub watershed
Sub watershed Percent (%) Distribution
Rank
1
2
3
4
5
6
7
8
9
ID
Area
(km2)
LCLU Class
Area (km2)
Natural
Stand
Wetland
Second.
Growth
Shrubland
Forest
Loss
Grassland
Agriculture
Barren
Urban
1
0.58
3.68
98.81
0.03
0.94
0.01
0.00
0.09
0.00
0.00
0.12
2
0.80
0.73
94.10
0.09
0.00
0.65
0.00
0.31
0.01
0.00
4.84
3
13.12
3.69
82.33
2.78
7.99
1.23
0.00
2.65
0.53
0.08
2.41
4
6.88
1.84
77.04
4.78
8.73
1.66
0.95
4.46
0.18
0.04
2.17
5
20.52
6.89
88.12
2.30
3.70
1.42
0.24
2.23
0.75
0.01
1.24
6
8.53
1.77
62.45
2.95
13.52
2.55
6.85
4.61
0.91
0.52
5.65
7
17.99
5.01
76.01
1.84
7.43
1.20
9.37
2.40
0.19
0.00
1.54
8
0.86
0.95
90.91
0.29
2.50
1.77
0.00
0.53
1.15
0.03
2.82
9
4.51
0.99
97.13
0.59
0.00
0.13
0.00
0.15
0.09
0.00
1.92
10
5.61
3.78
93.12
0.42
1.68
0.98
0.00
0.97
0.75
0.01
2.07
11
1.33
2.22
95.50
0.13
0.00
0.66
0.13
0.57
0.95
0.00
2.07
12
4.71
1.94
74.80
1.65
3.33
1.37
11.56
3.30
0.56
0.07
3.36
13
18.67
3.85
65.77
3.07
6.77
3.42
10.86
5.34
1.04
0.20
3.53
14
10.41
1.58
61.94
1.58
10.53
3.72
8.57
9.19
0.96
0.04
3.48
15
13.20
3.15
63.90
0.91
15.99
3.57
2.84
6.60
0.85
0.08
5.26
16
2.74
0.21
41.70
1.27
26.35
3.47
14.04
1.26
2.07
0.05
9.80
17
1.19
1.39
89.76
0.41
1.77
0.97
0.00
0.55
0.37
0.25
5.92
18
43.38
7.58
69.78
2.82
5.28
3.35
6.47
4.90
1.00
0.63
5.77
19
18.11
3.11
61.09
3.19
11.86
2.19
8.98
6.66
0.74
0.19
5.10
20
2.35
4.50
95.69
0.04
1.91
0.34
0.92
0.22
0.20
0.01
0.68
21
2.54
1.18
83.84
1.25
9.53
1.12
0.10
2.46
0.50
0.01
1.19
22
2.66
4.09
98.45
0.11
0.63
0.30
0.00
0.14
0.16
0.00
0.21
23
16.66
4.11
73.99
1.01
8.10
3.21
3.40
2.66
1.95
0.47
5.21
24
18.27
2.58
72.84
1.39
7.40
3.94
4.19
2.58
0.50
0.35
6.80
25
5.71
5.73
91.61
0.41
0.52
1.63
0.98
1.57
0.75
0.24
2.31
26
3.28
1.03
84.10
0.60
9.21
1.71
0.00
1.99
0.52
0.00
1.88
27
4.13
2.59
76.73
0.59
6.95
1.67
2.09
7.41
1.07
0.04
3.46
28
15.40
3.78
80.96
1.30
0.05
2.21
3.56
9.02
0.55
0.03
2.32
29
9.96
5.44
84.56
0.62
10.38
0.26
1.97
1.18
0.47
0.00
0.55
30
28.93
5.07
78.58
3.32
4.73
2.37
4.08
3.93
0.92
0.02
2.04
31
10.29
3.97
80.36
1.14
5.42
1.02
7.92
2.57
0.29
0.02
1.27
32
21.03
4.06
58.15
3.43
1.46
6.10
4.46
21.66
0.83
0.06
3.85
33
14.45
4.78
73.80
3.96
6.79
1.83
1.05
11.00
0.59
0.00
0.97
34
22.21
5.96
74.47
2.23
6.59
3.72
0.97
10.08
0.39
0.02
1.53
35
19.29
6.44
72.15
4.68
10.75
1.85
3.16
5.56
0.47
0.04
1.33
36
23.46
4.17
68.74
2.03
6.49
2.85
3.27
13.88
0.31
0.04
2.39
50
Table D1 (continued). LCLU class distribution by sub watershed using the 100m riparian buffer zone & 9 classes.
Sub watershed
Sub watershed Percent (%) Distribution
Rank
1
2
3
4
5
6
7
8
9
ID
Area
(km2)
LCLU Class
Area (km2)
Natural
Stand
Wetland
Second.
Growth
Shrubland
Forest
Loss
Grassland
Agriculture
Barren
Urban
37
35.64
6.63
65.79
2.18
3.05
4.11
5.79
14.59
0.78
0.05
3.65
38
17.43
6.07
74.57
2.50
11.38
1.70
0.34
6.53
0.23
0.00
2.74
39
45.97
8.30
66.25
2.52
9.61
3.37
7.74
6.72
0.97
0.14
2.69
40
28.82
8.87
76.01
3.33
5.38
1.66
1.40
9.57
0.54
0.01
2.09
41
38.89
8.22
68.73
3.16
5.57
3.78
2.36
12.41
1.03
0.08
2.90
42
37.25
7.52
71.96
3.95
10.04
1.22
2.12
8.40
0.15
0.07
2.09
43
13.04
3.58
85.21
4.60
4.01
0.96
0.84
3.68
0.13
0.03
0.55
44
0.93
0.76