Research Question

How is the spatial presence of postfire woodpecker nests related to the spatial presence of salvage-logged forest stands?

• How are woodpecker nests clustered within survey units? (Exercise 1 and 3)
• How does this clustering relate to salvage treatment units within the survey units? (Exercise 2 and 3)

After evaluating my data with several spatial analyses methods, this research question still seems applicable to my project goals. A possible revision to this research question is to include phrasing about the internal spatial characteristics of the nest and salvage harvest location datasets aside from their relationship to each other.

Dataset

This project uses postfire datasets targeting the 2015 Canyon Creek fire complex on the Malheur National Forest in eastern Oregon. A salvage logging operation occurred in the burn area in July 2016. My geospatial analyses examine the relationships between woodpecker nest site locations and salvage harvest unit locations before and after the logging occured. This will lead to correlating woodpecker population dynamics with salvage treatment effects. Woodpeckers are an ideal indicator species for postfire ecosystem monitoring due to their heavy reliance on burned snags for bark-beetle foraging and cavity-nesting. Their population dynamics indicate habitat quality for all snag-dependent wildlife. Previous research indicates woodpeckers exhibiting species-specific habitat preferences toward burned forest variables like stand density, diameter class, height, and snag decay class. 2016 pre-salvage and 2017 post-salvage lidar datasets are in processing for eventual correlation between these 3D forest structure variables and woodpecker nest site selection before and after harvest. However, the spatial analyses featured in this project focused on the intial task of relating 2D nest and salvage unit locations.

My research is in conjunction with a Rocky Mountain Research Station study examining salvage logging effects on three woodpecker species in the Canyon Creek complex. In 2016 and 2017 I led crews on this project collecting extensive postfire woodpecker occupancy and nest productivity datasets for black-backed, white-headed, and Lewis’s woodpecker populations. This resulted in a 148-nest dataset for 2016 and 2017, representing woodpecker populations before and after salvage logging. A polygon shapefile outlining ten RMRS woodpecker point count survey units serves as the area of interest (6 treatment, 4 control). Within the 6 treatment units, another polygon shapefile outlining 35 salvage harvest units indicates treatment areas. Three silvicultural prescriptions replicating optimal habitat types for each woodpecker species designate target salvage variables like average post-harvest stand density and diameter class. Each salvage unit adheres to one of these three harvest prescriptions. Supplementary geospatial data includes a 2015 WorldView-3 1 m raster and ArcGIS basemaps.

Above: The 2015 Canyon Creek Fire burning near John Day, OR.

Black-backed woodpecker (Picodies arcticus)     White-headed woodpecker (Picoides albolarvatus)                       Lewis’s woodpecker (Melanerpes lewis)

Above: The three target woodpecker species. All species are Oregon Sensitive Species. The white-headed woodpecker is a U.S. Fish and Wildlife Species of Concern.

Above: Survey units and corresponding salvage treatments for each unit. Each treatment type is based on a silvicultural prescription designed to benefit target woodpecker species. Based on previous research, each target species demonstrates preferred habitat characteristics in variables like stand density, diameter class, height, and decay class.

Above: The three salvage treatment types and the control treatment depicted as overhead stem maps. Each polygon represents a snag of indicated diameter class. The salvage treatments retain more trees of smaller diameter moving from Treatment 1 to Treatment 3, with the control units retaining all trees.

Above: A salvage treatment in the Crazy Creek woodpecker survey unit (Treatment 1).

Above: An early project map showing rough locations of Rocky Mountain Research Station woodpecker survey units. Since this map was created, two more survey units were added for 10 total (6 treatment, 4 control). This map effectively shows the woodpecker survey units occupying areas of highest burn severity.

Above: The Canyon Creek fire complex as a false color WorldView-3 1 m raster. The area of interest includes 10 survey units in blue, labeled with yellow text (6 treatment, 4 control). This visual orients the survey units to an area in eastern Oregon southeast of John Day and Canyon City. The false color image displays healthy vegetation as red, with the darkest areas displaying high burn severity. The survey units are found within some of the highest burn severity areas in the fire complex.

Above: A close-up of the 35 salvage treatment polygons outlined in red and labeled with white text. Control units lack red salvage polygons. This image does not include Overholt Creek.

Above: A subset of the 2016 (orange) and 2017 (pink) nest points featuring survey (gold) and salvage unit (green) polygons.

Hypotheses

I expected to see dispersed nests in 2016 with possible trends indicating species habitat preferences. Previous research indicates species-specific preferences for certain forest habitat variables. Black-backed woodpeckers prefer dense, small-diameter stands for foraging and nest excavation. White-headed woodpeckers prefer a mosaic of live and dead variable density and diameter stands for pine cone foraging. Lewis’s woodpeckers prefer tall medium to large-diameter stands for aerial foraging maneuvers. I expected to see nest sites in both years clustered in areas with these forest structures. In 2017 I also expected to see nest sites clustered near salvage treatments implemented for each species. Overall I expected the control units to exhibit nest dispersal and high woodpecker activity.

Approaches

Exercise 1: Multi-Distance Spatial Cluster Analysis (Ripley’s K)

Multi-Distance Spatial Cluster Analysis (Ripley’s K) analyzes point data clustering over a range of distances. Ripley’s K indicates how spatial clustering or dispersion changes with neighborhood size. If the user specifies distances to evaluate, starting distance, or distance increment, Ripley’s K identifies the number of neighboring features associated with each point if the neighboring features are closer than the distance being evaluated. As the evaluation distance increases, each feature will typically have more neighbors. If the average number of neighbors for a particular evaluation distance is higher/larger than the average concentration of features throughout the study area, the distribution is considered clustered at that distance.

Referenced from ArcGIS Desktop 9.3 Help (http://resources.esri.com/help/9.3/arcgisdesktop/com/gp_toolref/spatial_statistics_tools/how_multi_distance_spatial_cluster_analysis_colon_ripley_s_k_function_spatial_statistics_works.htm)

Exercise 2: Multiple Buffer Distance Analysis

This tool creates shapefiles of concentric circles around features based on user-input values. Users can dissolve buffers to create a single feature class per buffer distance. Buffer widths overlapping with features of interest can indicate spatial relationships between target subjects. In this case, intersecting woodpecker nest buffers with salvage harvest polygons may reveal trends in woodpeckers selecting nest sites closer to or farther from salvage units. An equation producing percent area of each nest buffer intersected by a salvage unit serves as an index.

Exercise 3: Nearest Neighbor

The Average Nearest Neighbor tool measures the distance between each feature centroid and its nearest neighbor’s centroid location and averages the distances for a nearest neighbor ratio. If the ratio is less than the average for a hypothetical random distribution, the feature distribution is clustered. If the ratio is greater than a hypothetical random distribution average, the features are dispersed.

Results

Exercise 1: Multi-Distance Spatial Cluster Analysis (Ripley’s K)

Above: Notable clustering results from Ripley’s K analysis. Most sample sizes were too small to produce significant results, but Big Canyon and Crawford Gulch in 2016 and Lower Fawn Creek in 2017 showed evidence of clustering.

Exercise 2: Multiple Buffer Distance Analysis

Above: The difference in results between 50 and 100 meter buffer distances for the 2016 nest dataset demonstrate the utility of graphs created from buffer analysis. These results indicate that woodpeckers consistently choose to place their nests in areas completely surrounded by salvage treatment 3. The results also indicate that woodpeckers consistently place their nests within varying distances of salvage treatment 1 with trends indicating average nest placement near but not completely within those salvage treatments.

Exercise 3: Nearest Neighbor

Above: Nearest neighbor results for woodpecker nests in each survey unit in 2016 and 2017. The NN Ratio is a threshold for expected clustering or dispersion. NN Ratio values less than 1 indicate clustering and NN Ratio values greater than 1 indicate dispersion. In 2017, green cells indicate an increase in value and red cells indicate a decrease in value. NN Ratio cells with an increased value in 2017 indicate units where nests increased dispersion. NN Ratio cells with a decreased value in 2017 indicate units where nests increased clustering. Alder Gulch (Treatment 3), Lower Fawn Creek (Treatment 2), and Upper Fawn Creek (Treatment 2) demonstrated increased clustering in 2017. Big Canyon (Treatment 1), Crazy Creek (Treatment 1), and Sloan Gulch (Treatment 3) experienced increased nest dispersion in 2017. All control units experienced increased or present dispersion in 2017.

Significance

I am processing two lidar datasets of the study area from 2016 and 2017. These datasets were acquired before and after the salvage logging treatments occurred. I will produce forest metrics such as stand density, diameter class, and height in salvage and survey units. I will then correlate geospatial and statistical properties of the nest datasets to quantified forest variables affecting woodpecker nest site selection. I will examine trends between 2016 and 2017 nest selection to understand the effects of harvest treatments on woodpecker populations. At least two more years of woodpecker data will exist for 2018 and 2019, so future research will add these datasets to the analyses. I would like to see a machine learning algorithm developed from this study that could predict areas of optimal habitat suitability for snag-dependent wildlife species. Postfire wildlife habitat prediction will be crucial to resource managers as wildfires increase in the coming decades alongside accelerated landscape restoration.

This spatial problem is important to science and resource managers as climate change amplifies wildfire effects. Using 3D remote sensing datasets for resource management is the future trend across all disciplines. Increased wildfire activity around the world necessitates cutting-edge methods for active fire and postfire ecosystem quantification. In the Blue Mountains ecoregion in eastern Oregon, Rocky Mountain Research Station, Malheur National Forest, and Blue Mountains Forest Partners rely on this project’s lidar quantification for their research and management decisions. Determining acceptable levels of salvage harvest for wildlife habitat affects whether government agencies and rural economies in this region will allow small businesses to profit from sustainable harvest operations. Science overall will benefit from the continued investigation of wildlife response to anthropogenic disturbance, specifically postfire forest management decisions and the controversial practice of salvage logging.

Learning Software

For my analyses I used mostly ArcMap, ArcGIS Pro, and R. I am very familiar with ArcMap, but it was my first time using ArcGIS Pro extensively. I enjoyed applying my R programming experience to my own datasets and creating visual aids for my research.

Learning Statistics

The statistical analyses I implemented for my research did an adequate job characterizing my datasets. However, I would try a different set of analyses that fit my small sample sizes better. A spatial autocorrelation test may support these analyses. I used R heavily in my Ripley’s K and buffer analyses to create graphs

Research Question: How is watershed recession behavior related to the spatial pattern of watershed lithological composition in Oregon coastal freshwater systems?

Data

Recession Analysis

For the recession analysis performed in this exercise, I used daily records of precipitation from the PRISM database. Daily data is the highest temporal resolution that PRISM has to offer. The PRISM time series were downloaded from cells nearest the gage. Time series were from 2001 to April 2019. For further methodology on recession analysis, refer to Exercise 1. River discharge data was collected by the U.S. Geological Survey at the watershed pour point used here. Discharge data is recorded every 15 minutes, which was upscaled to daily average to match the resolution of the precipitation data. As a result, recession analysis results were on the daily timestep from April 2019 to 2001.

Geophysical data

Lithological data was sourced and derived from the Oregon Department of Geology and Mineral Industries state geological map. General lithology type is a listed variable in the attribute table (GEN_LITH_TY), and ArcGIS was used to dissolve that attribute across the study area. The spatial resolution of this dataset is unclear. Additionally, the data is representing “immutable” features, and therefore there is no associated temporal scale.

Hypotheses

Streams and rivers in watersheds with predominantly volcanic geology will recede faster than those with predominantly sedimentary geology.

Streams and rivers in watersheds with similar underlying lithology will have similar recession characteristics.

Approaches

To perform this analysis, I followed the following steps:

1. Recession analysis on the daily timescale across eight watersheds. Statistical methods and other analytical procedures are outlined in Exercise 1.
2. Profile lithological composition within each of 35 study watersheds across the Oregon Coast Range. Percentages of metamorphic, plutonic, sedimentary, surficial sediments, tectonic, and volcanic lithologies were quantified for each study watershed. Technical procedures are outlined in Exercise 3.
3. Perform hierarchical clustering analysis on lithological composition profile to see which watersheds are most lithologically similar and dissimilar.
4. Compare recession analysis coefficients to clustering results and lithologic profiles to see if recession behavior appears to be different in different predominant lithologies.

Results

Results from the clustering analysis demonstrated that a most coastal watersheds are predominantly underlain by sedimentary and volcanic lithologies, and that sedimentary lithologies dominate the region. However, volcanic lithologies are well-represented, primarily at the northernmost latitudes. Because of this stratification, recession analyses could be performed at watersheds in both lithologies and compared to test whether lithology is a dominant control. However, because I conducted the recession analysis before profiling lithological composition, only one of the eight hydrologically analyzed watersheds is in predominantly volcanic geology (Figure 1). The watersheds grouped under the sedimentary branch span values of 54-100% sedimentary lithology. Watersheds grouped under the volcanic branch span values of 47-74% volcanic geology.

Figure 1. Results from hierarchical clustering analysis depicted in a dendrogram. Watersheds in which recession analyses were performed are circled.

When recession coefficients from the recession analysis are compared with the clustering analysis results, we see that coefficient a is fairly consistent among all watersheds, while coefficient b is much larger in the predominantly volcanic watershed (Table 1). The Wilson River near Tillamook, Oregon, is characterized as 50% volcanic lithology, which is the highest among the studied watershed. Recession analysis coefficient b from the Wilson River is 1.726 is also the highest in the eight studied systems.

Table 1. Table showing lithological composition of watersheds for which recession analysis was performed. Coefficient results from the analysis are also listed.

Significance

While the results of this study are inconclusive given the small sample size, it does give indication that lithology is an important control on watershed recession behavior and should therefore be included in analysis of temporal patterns of hydrological behavior. It is relevant to resource managers in coastal PNW regions because it indicates a watershed characteristic that influences storage and release behavior. Such information may be valuable for identifying systems that are more or less susceptible to drought and changes in the climatic patterns.

Learning Outcomes

• Watersheds can be delineated in Matlab, and it is much faster than ArcGIS.
• dplyr is a very important and useful package for reformatting data in a logical way
• I learned more about spatial autocorrelation from conversations with other students about their analyses, especially given the breadth of research topics in the classroom.
• Probably the two most important outcomes of this project were the recession analysis and the clustering analysis. The recession analysis involved multi-variate regression, which when applied to my data and research question, makes far more sense. The clustering analysis helped get at the question I have had for some time: how similar/dissimilar are my study watersheds? For this project, I focused on lithology, but I think it will be a useful tool for analyzing the other variables as well. Additionally the clustering analysis will provide a useful method for stratifying my sampling design into different lithological groups.

Final Project Blog: Characterizing the exclusion of refugee settlements from global settlement datasets

Anna Ballasiotes

RESEARCH QUESTION

This research is a small part of a larger question: where are refugees? Which in itself is the basic question to a larger problem: how will the global community provide resources to those who need help — and in order to do so, understanding where these vulnerable populations exist is particularly important. In considering the hierarchy of understanding where IDPs and refugees exist, there are certain drivers of the creation of a ‘refugee population,’ notably conflict (most often varying levels of violent conflict) and more recently, climate events (including natural disaster events). Beyond the drivers of refugee populations, there are different typologies of refugee populations: refugees within cities and blending into existing infrastructure, refugees in settlements and camps that are close to borders, refugees in settlements close to urban centers, and refugees in settlements and camps in areas where populations have not been established and as such there is available land. My focus lies largely in the last item of that list: where populations have not been formally established but now camps and settlements are expanding. It is a known problem that global settlement datasets are not doing an adequate job of recognizing these camps and settlements as “population.” As such, my goal was to characterize how these global settlement datasets were not recognizing these camps and settlements — and thus excluding these populations. This asks the question: to what extent do global settlement datasets fail to detect known refugee camps and settlements that form in areas of low density population?

I originally sought to answer the question, “what about the classification metrics of global settlement dataset fails to detect settlements known to OSM.” I wasn’t able to truly answer this question because I needed to spend more time  on actually characterizing this exclusion before understanding why these global population datasets were unsuccessful. Characterizing where this exclusion occurred was difficult enough and took most of the class, especially as I explored what these refugee settlements looked like and considered how they may have formed.

DATA

I used three main data sources: Refugee settlement boundaries from OpenStreetMap, Facebook High Resolution Settlement Dataset, and the World Settlement Footprint. Only data within Uganda was used for these data, given the overlap of data availability and project partners with reliable data. The Facebook High Resolution Settlement Dataset is trained on 2018 imagery and is presented at a 30 meter resolution. World Settlement Footprint is trained on 2015 imagery and is also presented at 30 meter resolution, but with a minimum mapping unit mask that creates clusters of pixels that make the data appear to have a coarser resolution. The refugee settlement boundaries from OSM range in temporality from 2016 to 2019; however, this is only when the boundaries were added to the OpenStreetMap database and is not necessarily when camps or settlements were established. I had initially included the Global Human Settlement Layer, which is also trained on 2015 imagery, but GHSL did not appear to detect any settlement in the refugee camp boundaries within OSM, so I was not able to perform any meaningful analysis with it.

HYPOTHESES: predictions of patterns and processes.

I wanted to understand in what ways the global settlement datasets excluded the known refugee settlements — thus, I wanted to understand what percentage of refugee settlements were captured and whether larger settlements or smaller settlements showed a higher percentage of captured area, or if settlements that had a higher percentage of captured area showed a faster green-up around the settlement as one moved away from the boundary. I suspected that smaller settlements would have a higher percentage of area captured, mostly because I predicted that smaller settlements conformed more to the actual built-up of settlement area, whereas the larger settlements included more land cover types that would be less likely detected as “built-up.” I also suspected that settlements that showed a higher detection percentage would have a more distinct natural boundary — that the green-up around the settlements would occur faster if a higher percentage was detected.  

APPROACHES: analysis approaches used.

I had initially measured the spatial autocorrelation of the refugee settlements using ArcGIS Pro, but did not find this analysis to provide too much significant information for me. I already knew that my settlements were quite different from each other in size; any clustering would have really come from other factors that are not obviously visible, like the way land was distributed or delegated for refugee settlements.

In order to characterize the percentage of area of refugee settlements captured by each settlement dataset, I brought everything into Earth Engine and calculated pixel area of every pixel both within the refugee boundaries and across WSF and Facebook datasets — counting only the pixels that have area detected as settlement.

In order to refine the comparison of area detected within refugee settlements, I also needed to really only look at pixels that look like “built-up” within the refugee settlements, since the boundary themselves have quite a lot of green space and other land cover types. also performed a binary unsupervised classification based on the known locations of refugee settlements. Ultimately, I used a confusion matrix to compare this binary unsupervised classification to WSF and Facebook to refine the “percent area detected.”

I also looked for the relationship of detection and greenness moving away from the boundary. To do this, I created a multi-ring buffer out to 1 kilometer at 100 meter increments, calculated the average maximum NDVI in each “donut buffer” for 2018 using Landsat 8 imagery. I then plotted these NDVI values over increasing distance from the refugee boundary to see the green-up.

RESULTS

I produced a map of the spatial confusion of overlap between WSF, Facebook, and a binary unsupervised classifier within my settlement boundaries. I have refined this map a bit, but there are still further statistical comparisons that could be done to better characterize the overlap of WSF, Facebook, and the K-Means classifier.

I also have all of the settlement boundaries with the percent of area captured by each global dataset within each boundary: this could be a sorting mechanism in which I could perform furt

Ayilo 1 & 2 Refugee Settlements; Classifications shown

her analysis to better answer my hypothesis, which I didn’t necessarily answer well, specifically with regards to the NDVI prediction. Within the exercises, I answered this question based on geography rather than percent detection — did settlements in the north vs. south show different pattern of vegetation greenup around the settlement?

SIGNIFICANCE What did you learn from your results? How are these results important to science? to resource managers?

I learned that across the board, refugee settlements ARE systematically excluded from global datasets but can easily be detected at the local scale, and thus this systematic bias of these datasets must be addressed and solved. Those that are modeling systems of migration or providing access to humanitarian resources need to have more accurate representations of settlements. Beyond this, I was really hoping to be able to characterize the spatial patterns of the exclusion better than I was able to, and I think that’s where my analysis fell short and where my next hypothesis would manifest.

LEARNING: what did you learn about software (a) Arc-Info, (b) Modelbuilder and/or GIS programming in Python, (c) R, (d) other? What about statistics? including (a) hotspot, (b) spatial autocorrelation (including correlogram, wavelet, Fourier transform/spectral analysis), (c) regression (OLS, GWR, regression trees, boosted regression trees), (d) multivariate methods (e.g., PCA),  and (e) or other techniques?

I didn’t truly dive more deeply into these softwares than I have in the past; I think I just became more acutely aware of the statistical methods that exist in Arc toolboxes and became more comfortable manipulating data in R.  

I think that the statistical learning was less specific or intense than I was hoping, but also in part because I think I was really challenged to think statistically and spatially-statistically about my research question. I wish I had spent more time understanding statistical methods rather than trying to mold my question into something that seemed to fit my elementary understanding of spatial statistics.

My project focused on a set of grain size and cross-sectional change data from a long term research project in the Andrews Forest. (Maps and study description from Blog Post 1)

I wanted to explore two questions:

• How do stream channel erosion, deposition, and particle size vary over time and space?
• How do these changes relate to adjacent changes in the across-stream or along-stream direction?

Hypotheses:

I tried to address these nested hypotheses about the system:

1. Existing stream bed morphology drives spatial patterns in cross-sectional change.
   1. Unit-scale stream morphology should lead to patchy transport and different levels of autocorrelation of the change at different spatial scales
   2. Either hillslope or hydrodynamic processes or both should result in relatively high cross sectional change near banks.
2. Extreme flow events alter the size distribution of in-channel sediment
   1. High energy flows associated with high peak discharge events should increase median grain size
   2. Transport of hillslope material (also associated with high peak discharge events) should decrease sorting
3. Extreme flow events reduce the relative impact of bed morphology in cross-sectional change.

Here’s a conceptual model to investigate the patterns of change

Overall, I expected that different processes would drive change at different scales, so I expected the spatial and temporal patterns of change to vary based on scale as well.

Spatially:

• At reach scales: patterns of change are driven by reach level features including slope, watershed size, and land use history.
• At unit scales and smaller, patterns of change are driven by unit-scale bed morphology and positions of adjacent features.

Temporally:

• During time periods that include extreme flows, patterns of change are driven by the magnitude of the flow.
• During time periods of less extreme flows, patterns of change are more dependent on local features as described above

Approaches:

Due to the study design and to limits in data availability, it made sense for me to use one-dimensional analysis methods and correlation or autocorrelation tools on much of the data. I mostly worked in R. I also started but didn’t finish work on a network model using a tool specifically designed for stream network analysis.

Technical limitations:

1-dimensional models don’t perfectly capture the complex realities of stream networks and irregularly spaced spatial data, so I ran into a few hurdles trying to parse analysis results. I also couldn’t do some of my desired analyses due to the lack of shared coordinate systems for different data sets or for adjacent cross sections. However, thinking about these analyses helped me better understand exactly which data I would need to collect in order to do them in the future.

I am still having some trouble with a stream network analysis add-on in ArcGIS, and I am in the process of reaching out to experienced users who I hope know the way around my particular error.

Results:

• Results from Exercise #1 generally supported hypothesis 1.a.
• Results from Exercise # 2 might somewhat support hypothesis 1.b., but I think I need a more robust way to test this.
• Results from Exercise # 3 don’t appear to support hypothesis 2.a, but they might show some support for hypotheses 2.b. if the effect is lagged by one year.
• Patterns in Exercise 1 might show some support for hypothesis 3, but I really need a more robust analysis to pull apart the effects.

I started the stream network analysis partly because I wanted more context for the results of exercise 3. Some 2017 REU students collected grain size data at sites that overlapped with the long term cross sectional study sites. I haven’t finished the network analysis, but I’ve mapped the REU student data below.

Here is a map of the median grain size (D50) at the REU student sites:

Here is a map of half the difference between the 84^th and 16^th percentile grain size at the REU student sites. This value is one metric of sorting, and it would equal the standard deviation if the sizes were normally distributed.

These results indicate that grain sizes and sorting in the grain sizes display more complex patterns than a simple fining of grain size downstream or consistent gradations in sorting.

For all of the analyses, I need to spend a little more time figuring out which of these results are useable and might reflect reality in the field and which ones are more likely showing artifacts of data collection or processing

Significance:

I hope to use these results to add more depth to my research and place my research in a broader context. I personally learned more about strengths and limitations of data set

The broader purpose of entire project is to see how streams like these respond over time. How much of their changes are influenced by hydrologic rather than hillslope processes? How big of a hydrologic event needs to happen to do substantial work on the channel? The project could be useful to land managers because channel mobility could damage streamside infrastructure or alter benthic ecology

My learning (technical):

From a technical standpoint, I think improved ggplot skills. I started using cowplot and learned more about custom themes. I learned a lot about preparing data using the STARS toolbox, and I am excited to learn more about the SSN R package once I get my data working

My learning (statistics):

In this class, I learned more about autocorrelation methods and correlograms. Hope to learn more about network analysis soon too.

Question

My research question for this project was to investigate how spatial patterns of land use/land cover (LULC) change were related and/or associated with spatial patterns of artisanal, small-scale gold mining (ASGM) establishment. A possible mechanism I considered as influencing these patterns would be deforestation associated with the establishment of ASGM operations, as miners require significant amounts of timber to bolster mine shafts, in addition to constructing shelters around the mine shafts and for the miners themselves. As such, my research question can be phrased thusly: “How is the spatial pattern of LULC change related to the spatial pattern of ASGM establishment via the mechanism of deforestation?” Additionally, I wanted to examine whether alternative factors may affect LULC change around sites; in this case, I examined the proximity of a site to a primary road and whether or not that related to rates of LULC change.

Dataset

My dataset for this project had several components. In previous work, I used very high resolution satellite imagery to map out the locations and spatial extent of nine ASGM sites, as well as nine non-ASGM sites to compare to (these sites are shown in fig. 1). After having mapped these sites, I acquired Landsat 5 and Landsat 8 imagery at 30m spatial resolution from the USGS for April 2007 and April 2018, respectively. Using this Landsat imagery, I calculated normalized difference vegetation indices (NDVIs) for each year, and then subtracted the former from the latter to obtain a dataset showing NDVI loss between each timeframe. This dataset enabled me to examine NDVI change around each of my 18 sites to evaluate LULC change in the form of NDVI change. In addition to these datasets, I also acquired a road network layer from OpenStreetMaps for Senegal, which I selected to only include “primary” roads in Kedougou. Together, these datasets enabled to me to investigate LULC change around ASGM sites via NDVI change and also how this change relates to proximity to primary roads.

Fig 1: Study Site

Hypotheses

My two hypotheses for this project are that 1) NDVI would be more negatively impacted near ASGM sites compared to non-ASGM sites as a result of deforestation, and 2) NDVI losses at ASGM sites could be attributed in part to each site’s proximity to primary roads.

Patterns of higher NDVI loss as a percentage of the total would occur at ASGM sites compared to non-ASGM sites. This may be because, as the larger populations near ASGM sites and also the activity itself would necessitate more timber harvesting, there would be more negative NDVI change in the immediate surroundings compared to sites without ASGM activity. The spatial process I would expect to see would be areas of higher NDVI loss nearer to the center of ASGM villages. These areas would be fragmented, indicating differences in vegetative cover around the village, but, overall, more negative than around non-ASGM villages. This pattern results from miners and villagers harvesting the nearby timber to bolster mine shafts, as well as for construction of structures around each mine and their own huts. This hypothesis is an explanation for the pattern of NDVI loss around a village.
ASGM sites may be, overall, closer to primary roads than non-ASGM sites, and so their growth and attendant NDVI loss is in part attributable to proximity to major travel corridors and the relative ease of reaching the sites. If a mine is closer to a primary road than another mine, it’s possible that this proximity may result in larger spatial patterns of NDVI loss. This hypothesis is an explanation for the process of deforestation around different sites.

The first hypothesis here is an attempt at understanding what spatial patterns may be occurring around ASGM sites; that is, it is an attempt to understand what NDVI losses may look like. The second hypothesis is an attempt to understand why those patterns may be occurring, and in finding some sort of correlation between the observed pattern and an alternative possible explanation for that pattern.

Approaches
Exercise 1: Moran’s I

Before examining how patterns of NDVI loss are shaped around ASGM and non-ASGM sites, I first wanted to examine how the patterns of NDVI loss and gain were self-related; that is, I wanted to know whether or not patterns of NDVI change occur near areas of similar change (spatially autocorrelated or clustered) or whether or not these patterns occur away from self-similar patterns (dispersed). To do this, I took my layer of NDVI change and ran a script in Rstudio to generate a Moran’s I value for the global layer, which is a measure of spatial autocorrelation: it results in values between -1 and 1, where values closer to -1 indicate negative spatial autocorrelation (dispersion) and values closer to 1 indicate positive spatial autocorrelation (clustering). I also performed this analysis at a smaller scale around one mining village to see how Moran’s I may change on a different scale.

Exercise 2: Neighborhood Analysis

After determining the spatial autocorrelation of patterns of NDVI change, I could examine how those patterns were related to and varied between ASGM sites and non-ASGM sites. To do this, I performed a neighborhood analysis. At each village, I found its center using the Mean Center tool in ArcGIS Pro. Around each center point, I generated “donut” buffers using the Buffer tool at 250 meter increments, up to 2 kilometers, for eight different donuts (e.g. 0-250m, 251-500m, 501-750m, etc.). I then used each of these buffers to Clip the global NDVI layer (which I had also edited to omit values over rivers and Mali), generating eight raster layers representing NDVI change values between each spatial step. From there, I simply generated histograms for each step and counted the values below 0, indicating NDVI loss from 2007 to 2018, and found their percentage of the total. With these values, I could then see how NDVI loss as a percentage of total pixels in each donut varied with distance from the center of each village, and could more readily understand how NDVI values change with distance from both ASGM and non-ASGM villages.

Exercise 3: Pearson’s R Correlation

After examining how NDVI change across space near ASGM and non-ASGM villages, I wanted to further understand how these patterns may be related to another factor: proximity to a primary road. To do this, I used my road network layer from OpenStreetMaps showing locations of all primary roads in Kedougou and used the Near tool in ArcGIS Pro to find the shortest distance from each of my 18 sites to the nearest primary road. Then, I took only the first values of NDVI change (from 0m to 250m), and, using these values, along with each site’s distance to the nearest primary road, I performed a Pearson’s R calculation (fig. 2). This approach thus attempts to explain how values of NDVI loss around both ASGM and non-ASGM sites may be related to a site’s proximity to a primary road, providing some more information on how different rates of NDVI loss may be affected by the ability of people to access a particular village.

Fig 2: Pearson’s R Correlation. x, in this case, is distance (m) from nearest primary road and y is NDVI loss at the first ring.

Results

I had several results from these three approaches.

Firstly, I produced a global Moran’s I of 0.6716816. The second Moran’s I produced, at a smaller scale, was .8079745. These results helped me to understand that, generally, patterns of NDVI loss are spatially autocorrelated, meaning that areas of loss and growth are clustered near self-similar areas. However, this analysis occurred over such a large area that it is possible the global Moran’s I may be accounting more for patterns of near riparian corridors, which are of course autocorrelated, especially areas where water bodies are visible. As such, this result needs to be interpreted with caution.

Secondly, the results from my neighborhood analysis are presented in fig. 3 [insert graph of NDVI changes for fig. 3; include key in caption]. These results indicate that, on average, villages which have ASGM activity present see more NDVI loss in their immediate surroundings than non-ASGM villages, up to about 2km radially from their center. This means that, generally, villages with ASGM activities experience reduced NDVI, which may indicate vegetation loss or loss healthy vegetation, than villages without ASGM present.

Fig 3: NDVI results. Light blue lines indicate ASGM mines; the dark blue line is their average. Light red lines indicate non-ASGM villages; the dark red line is their average.

Finally, the results of my Pearson’s R correlation indicate relatively weak correlations between a village’s NDVI loss (at least, within the first ring) and its proximity to the nearest primary road. The results are shown in table 1 [insert table screen snip]. For ASGM villages, their correlation was -0.1418, whereas non-ASGM villages have a Pearson’s R of -0.0578. Both of these are weak correlations, there is some indication that ASGM villages and their attendant NDVI losses are more correlated with proximity to a primary road (i.e., the closer they are to a primary road, the more NDVI loss they experience) than non-ASGM villages.

Table 1: Pearson’s R Results

Significance

These results are significant for several reasons. These showed to me that there is an observable pattern of NDVI loss that is greater around villages which develop or engage in ASGM than in villages which do not engage in ASGM. However, there is still some work to do in order to understand the process by which NDVI varies, and it may not be explainable by a village’s proximity to a primary road. There are several possibilities that may explain the process: deforestation is more prevalent around the gold mines for reasons to do with securing timber for construction, or because an inherently larger population at gold mines necessitates more forest product harvesting. More work should be undertaken to understand these processes. On the whole, however, these results are useful for governments or NGOs which seek to help better the conditions which ASGM practitioners participate and engage in, now that they know that ASGM villages experience greater rates of NDVI loss, which may indicate environmental degradation.

Learning

This process taught me several techniques. Firstly, I had no experience at all with R and R Studio, though I do understand some coding. As such, this helped me learn more about a really useful language that I did not have any experience with. Secondly, any practice with ArcGIS Pro is good practice for me, and so I got a lot of time and work done using that piece of software. In particular, I go some good time in with the model builder to help expedite work, as generating eight different ring buffers and then clipping them to an NDVI layer for 18 different villages would have been very time-consuming otherwise. Finally, I gained some good experience performing a statistical analysis, in the form of Pearson’s R. I have very little experience performing statistics, so this was really useful for me to help understand and interpret statistical analyses and results

Statistics

With regard to statistics, I learned several things. The three statistical analyses I used were Moran’s I for spatial autocorrelation, neighborhood analyses for comparing spatial patterns, and Pearson’s R for calculating a simple statistical correlation.

With regards to Moran’s I, I learned the importance of examining patterns at different scales, as a pattern which is apparent at one scale may not be so at another scale, and so interpretation and set-up must be conducted carefully.

Concerning neighborhood analyses, I learned the importance of dilligent workflow and juggling several different variables without muddying the results. This also demonstrated how to examine and structure a problem which seeks to understand change both across space and scross time.

Finally, conducting a Pearson’s R helped me to understand better how statistical analyses are constructed and what their significance is. Additionally, this taught me the value of interpreting a statistical result, as their significance may not always be indicative of anything. Null results or negative results are just as valuable as positive ones, and this analysis showed me that. Now I know more about what process to examine and how to examine them!

Introduction:

I have looked at the spatial distribution of depth to first lava and the spatial relationship between the depth to first lava and the depth to first water. I found that the distance from volcanic centers might affect the depth we find the contact with first lava. I did not find any strong correlation between first water and first lava, but that does not preclude the idea that first water correlates to deeper lava contacts.

Now I want to look at what else might be effecting the spatial distributions of lava in the region. Note, that the techniques I used for looking at lava can also be applied to looking at the spatial distribution for water in the field area.

Figure 1: The style of active faulting in the Pit River Study area (delineated roughly in green). N-S trending faults accommodate dip slip (up-down) motion, while NW-Se trending faults accommodate oblique slip motion (up-down and side to side). Structurally controlled volcanoes often lie at the elbows of intersection between the N-S and NW-SE trending faults. Modified from Blakely et al 1997.

The Northern California Field area lies at the intersection of the Walker Lane Fault Zone, the Basin and Range and the Cascadia subduction zone. For the purpose of this examination I will focus on the relationship between the spatial distribution between Walker Lane style faulting (fig 1) and the spatial distribution of the depth to first water and the depth to first lava in the region.

The topography in the region reflects the interplay between active Walker Lane Style Faulting in the region and the higher frequency volcanism that mutes it. For, example, in the northern portion of the field area, Medicine Lake Volcano, with flanks covered by volcanism less than 10 kyr (fig 2), active faults disappear beneath the volcanic edifice. They do not appear in the topography, but they exist in the subsurface. However, in regions where faults are well expressed, enough displacement has occurred to 1) be resistant to burial by periodic lavas, and 2) to either serve as conduits to lava flow, or to displace lava itself.

The Data:

Figure 2: Map of the Field area with active faults (pink lines), the Pit River (blue line), rough locations of ground water discharge (yellow circles), young lava flows (green outlines) and all of the available well data (blue dots).

The subset of the data we used in figure 2 is depicted in figure 3. Temporally they range from wells dug in the1950s to the present. It is important to note that depths to first water might have changed in the 60 years since the older well logs were recorded. For this reason I did not use many of the older well logs in my analyses.

Figure 3: The center of the township and range sections for the wells used and corresponding depths to first lava. Purple logs are the shallowest and Blue logs are the deepest.

Hypotheses:

Figure 3: Conceptual design for the project. A) Configuration before slip upon the fault. The lava flow is continuous. B) If a normal fault displaces a lava flow, then one side will go down with respect to the other. The lower block experiences deposition, and the development of a soil profile. If enough time has passed, then there will be a difference between the depth to first lava between the upper and lower block.

I predict that faults will run parallel to and coincide with large step changes in the depth to first lava.

If there are no large step changes in depth to first lava, I predict that these changes will be driven by changes in elevation and correspond to channelized lava flows rather than faults.

Methods:

I used the Threshold Kriging function in Arc to make a surface that corresponded to the likelihood of the depth to first lava being less than 30 feet deep. The Kriged surface then interpolates the probability that location at other points in the map correspond to less than 30 feet deep. When the surface has a value of 1, values are close to less than 30ft deep. When it is close to zero, values are greater than 30 feet deep.

I then preformed a confusion matrix to see how my results compared to what I expected. This part I did by eye.

Results:

Figure 4: Kriged Surface with faults superimposed.

Table 1: Corresponding confusion matrix. There were 257 High-Low values that corresponded to parallel faults, but there were 315 Low-Low or High Values that also corresponded to parallel faults. See discussion below.

Figure 5: Kriged surface with contours.

Discussion:

257 out of 790 of the faults were gradient parallel. If, corresponded to step function in kriged depth to first lava. However, many more did not. What is happening here? If you look to figure 5, you can see that while some of the high gradients correspond to rapid changes in elevation. This block is the footwall of a range bounding normal fault. That fault is not mapped because it is no longer active. But it has large amounts of uplift upon it, and likely corresponds to a different lithology entirely and so does not fall into our hypothesis.

If I do this analysis for every lava layer I find, and can correctly link the well logs together, then this analysis could help me constrain slip upon the faults in the region. This information is good for hazards assessment in the region as PG&E has a hydroelectric dam in the area.

Learning outcomes:

Throughout the course of this class, I learned a few of the nuances of R, though my work here has just begun. I also feel like I have a better understanding of how to think about spatial processes, which I why I enrolled in the first place. Future goals include better locating my data in space. This way, many of the techniques I used in this class will pick up on actual processes, and not on the gridded spacing of my data.

Sources:

Blakely, Richard et al. “Gravity Anomalies, Quaternary Vents, and Quaternary Faults in the Southern Cascade Range, Oregon and California: Implications for Arc and Backarc Evolution.” Journal of Geophysical Research: Solid Earth, vol. 102, no. B10, 1997, pp. 22513–22527.

Background

The objective of this research was to find a way to attribute proper classification of diseased versus non-diseased regions on a turnip leaf. To do this I assessed the classification accuracy of the unsupervised machine learning classification algorithm, segmentation, with manual classification. Additionally, I hoped to reveal some spatial characteristics about the lesions on the leaf for future work.

Dataset

The cumulative data set is roughly 500 turnip leaves, harvested here in the Willamette valley during the months of February to April of 2019. For the sake of this project I have selected 5 images in which I have worked with to complete the basis of this analysis. The spatial resolution is just under 1 mm and images are in raster format of single turnip leaves with five bands (blue, green, red, far-red, NIR). For this analysis only, the red band was used. The camera being used is a Micasense Red-edge which has a 16-bit radiometric resolution.

Questions

These were not all questions that I had anticipated answering at the beginning of the term, but have developed over the course of the term as I ran into issues or found intriguing results.

• Can diseased patches be successfully identified and isolated by computer image classification and through manual classification?
If yes;
o can the diseased patches based on image classification and manually classified diseased patches be characterized?
o How accurate is the segmentation classification?
o Are more raster cells being classified as false negatives or false positives when misclassified?
o What is going undetected in segmentation?

Hypothesis

My original hypothesize was that spatial patterns of pixels based on image classification are related to manually classified spatial patterns of observed disease on turnip leaves because disease has a characteristic spectral signature of infection on the leaves.

Approaches and results

I began in ArcGIS Pro by using the: segmentation -> mask -> clip -> raster to polygon -> merge -> polygon to raster. This yielded a layer which included only the computer generated segmented diseased patches which were exported as Tiff files. I followed a similar process with my manually classified diseased patches but used the ‘training sample manager’ instead of segmentation. Both these Tiff files were uploaded to Fragstats where I did patch analysis between the two to characterize the diseased patches. I found this not entirely helpful, but if I dug deeper into patch analysis I could be provided with some worthwhile information about the diseased lesion characteristics.

After characterizing the patches, I performed an error analysis using a confusion matrix in R. The two Tiff files were uploaded to R and converted to rasters to check quality and then to tables. The tables were imported into excel and values were adjusted to 1 for diseased patches and 0’s for non-diseased. Anything outside the area of the leaf was deleted. The tables were imported back to R and a confusion matrix comparing the computer classified segmentation versus the manually classified patches was conducted. All the classification matrices were combined for the overall accuracy (Table 1). This aspect was a bit time consuming as was the original image processing but was a very useful analysis.

One other approach I used to help with visualization was showing the leaf with false positives and false negatives included (Image 1). This was necessary to help explain results more clearly to others and show results outside of the statistics.

Significance

I have created a method for getting from image acquisition all the way to statistical results. This work is first off significant to my thesis research. This will be something I refine to help streamline but is a very good starting point for me. Second, I got statistically significant results based on p-value and a high accuracy. This will be helpful to those in agronomy or crop consulting which are using remote sensing for disease detection. At this point I don’t think it is practical for application of agronomist but with further work I hope to make this methodology an available resource for other researchers to work off of and for farmers to potentially try and utilize.

Hypothesis revised

After receiving these results and considering my next steps, my hypothesis has been refined. I now hypothesize that segmentation classification has an equal ability to classify diseased patches as manual classification because of the spectral values associated with diseased pixels.

New research questions

This research is far from complete and has led to a new set of questions to be answered.

• Is percent classification of diseased pixels the best metric for how well the classification method works?
• What is the accuracy of a support vector machine versus manually classified pixels?
• Is this classification translatable to other brassica species such as canola, radish, kale, etc.?

Future research

To answer some of my new questions I believe it is critical I learn how to create band composites between the five available bands. Currently, my issue is the bands do not overlap. My approach for solving this includes the use of georeferenced points/pixels on each image to ensure they overlap. I think if I use the 4 corners of the image I should get the overlap I want. By creating band composites, I can begin using vegetative indices like NDVI to help more accurately distinguish pixels than can be done with one band alone.

Another concern I have is despite classifying pixels accurately I am weary about this being an all determining method of classification. I have instances in my current classification where only 2 or 3 of the 6 or so lesions are actually having pixels inside its perimeter classified as diseased. I think a more accurate or a helpful assessment would quantify the percent of lesions that are captured by the segmentation process out of the total number of lesions on the leaf.

Using an unsupervised classification scheme for this work was good to lay the groundwork. In future work I would like to move to supervised machine learning classification schemes. My reason being; when attempting to translate this algorithm used on turnip classification to others such as canola, radish, etc. I believe it will have a higher level of accuracy and consider more spatial characteristics. Currently, my classification scheme only uses the spectral values of the red band where I would like to implement variables such as lesion size, proximity from margin, proximity to other lesions, shape, NDVI values, etc. This works into my last question concerning the ability of the classification to be applied to broader ranges of plants.

Technical and software limitations

Some of my limitations are apparent in above sections because they led to my future research questions. One of my greatest limitations was my ability to effectively and efficiently use ArcGIS Pro as well as code in R. This led to longer measures of time conducting image processing which may have been streamlined if I was better in ArcGIS Pro. The major roadblocks I hit in ArcGIS Pro were related to the overlap of different bands that I mentioned earlier in addition to the use of the support vector machine. I have plans to effectively solve the band composite issue, but am still unsure about the support vector machine function in ArcGIS Pro. I believe the way it was setup is not designed for collecting training data for many images but collecting this data from one image. My alternative for the SVM is to try conducting the analysis in R.

In R I struggled in my ability to code. This was a major strain on my time and required the assistance of others to help troubleshoot on many occasions. This can be overcome with more practice and is not so much a limitation of the software, but the user.

My learning

I learned the most in ArcGIS Pro I would say. The image processing took lots of time and many steps which exposed me to a whole set of tools and analysis I was previously unaware of. I don’t think my skills in R increased much but I now have the code and steps for the analysis I would like to complete on my future thesis work, the confusion matrix. Another program I gained some experience in was Fragstats. It was useful for characterizing the diseased patches and may be an option for some of my future work when looking at the size, shape distance, etc. of these diseased patches.

Appendix

Table 1. Confusion matrix for pixel classification of five leaves.

Image 1. Example of one leaf analysis where the green cells are manually classified non-diseased pixels, white pixels inside the leaf are segmentation classified diseased pixels and block pixels are the manually classified diseased pixels.

Question Asked

The question that I sought to answer in this class was “How is the spatial pattern of forage fish assemblage in the California Current System related to the spatial pattern of seascapes based on the sea-surface conditions used to classify the seascapes?”  This is outlined in my first blog post, which can be found here. Additionally, I was seeking to characterize the spatial distribution and extent of seascape classes and forage fishes. Seascape classes are used as a way to simplify the myriad of physical, biological, and chemical processes that can affect organisms in the ocean. Eventually, the focus of the analyses shifted to young-of-the-year rockfish, as opposed to forage fish. All analyses mentioned in this blog post were conducted using the YOY-Rockfish data as opposed to the forage fish data.

Data

Midwater trawls have been conducted annually by the National Oceanic and Atmospheric Administration’s (NOAA) Southwest Fisheries Science Center (SWFSC) in an attempt to monitor the recruitment of pelagic rockfish (Sebastes spp.) and other epipelagic micronekton at SWFSC stations off California. The trawls have informed a dataset that represents overall abundance of all midwater pelagic species that commonly reside along the majority of the nearshore coast of California from 1998 to 2015. Each trawl collected information about both fish abundance, recorded in absolute abundance, and location data, recorded in the form of latitude and longitude. The dataset also includes a breakdown of species by taxa.

Seascapes have been classified using a combination of in-situ data (from the trawls) and remotely sensed data from NASA’s MODIS program. Seascapes were classified using the methods described in Kavanaugh et al., 2014 and represent the seascape class in the immediate area that each trawl occurred. Seascapes are classified at 1 km and 4 km spatial resolution and at 8-day and monthly temporal resolution. Each seascape has been assigned an ID number which is used to identify similar conditions throughout the dataset.

The current seascape classification methods are able to classify surface conditions into one of 33 distinct seascape classes. The variables used to classify the seascapes are listed below:

• Sea-surface temperature (Celsius)
• Sea-surface salinity (Practical salinity units)
• Absolute Dynamic Topography (meters)
• Ice cover (% of area)
• Colored dissolved organic matter (m^-1)
• Spatial averaged sea-surface chlorophyll a (mg m^-3)
• Phytoplankton physiological state (NFLH) (W m^-2 um^-1 sr^-1)
• NFLH:Chlorophyll ratio

It is inferred that certain, carefully-selected sea-surface conditions can imply favorable conditions to support healthy ocean ecosystems in the water column and near the ocean floor. The conditions listed above have been studied and selected for this reason. However, this may not be a perfect science – any attempt to delineate and simplify the multitude of factors and conditions that facilitate life in the ocean is likely to leave out important factors. This underscores the importance of research that tests these classification methods and measures their ability to replicate and predict the true natural environment. The analyses conducted for this class attempt to do that in the context of one functional group of fishes within one major ecosystem in the Pacific Ocean.

Hypotheses

Simply put, I hypothesized that any measurable spatial changes in the spatial extent of certain seascape classes will also be identifiable in the spatial variability of forage fish assemblage over time. The California Current ecosystem is meticulously studied and examined by a myriad of researchers from a number of different affiliate institutions. Studies reviewing the physical and biogeochemical conditions of the area indicate that from an environmental perspective, many areas off of the coast of California should support areas of high fish abundance.

Specifically, I was expecting areas of high forage fish and young of the year rockfish abundance to exist in seascape classes that represent nutrient-rich, upwelled water. These conditions have been shown to support thriving ecosystems throughout the water column due to an abundance of energy and food for fishes that live higher in the water column. Upwelling brings cold, nutrient-rich water to the surface and occurs seasonally in most places along the California Coast. Using the variables listed above, that would mean below average sea-surface temperature, high dissolved organic matter, and high chlorophyll a. Some classes that represent conditions similar to this are:

• Temperate Transition (Class 7)
• Tropical/Subtropical Upwelling (Class 11)
• Subpolar (Class 12)
• Temperate Blooms Upwelling (Class 14)
• Warm, Blooms, High Nutrients (Class 21)

Preliminary multivariate community structure analysis has shown some statistically significant relationships between certain species and certain seascape classes using this data. If spatial patterns do exist, I expect there to be some relationship between the surface conditions and the fish found at depth of the midwater trawls.

Since my shift focused from forage fish to YOY rockfish, my hypothesis shifted from a prediction of correlation between certain seascape classes and forage fish abundance to relationships between rockfish and other seascape classes (as determined by the aforementioned multivariate analyses).

My hypothesis, taking into account all of the aforementioned considerations, can be restated as follows:

It is expected that the spatial pattern of areas of high young-of-the-year rockfish abundance will be related to the spatial patterns of seascape classes that represent areas of high nutrient availability and upwelled water along the California Coast.

Additionally, I expect even higher areas of abundance to occur in areas where two or more seascapes representing these favorable conditions converge or border one another. These border areas are likely to indicate much larger swaths of ocean that hold habitable conditions that are likely to be able to support entire communities rather than smaller numbers of fishes. While this hypothesis was not tested in this project, future work can be conducted using FRAGSTATS, patch analysis software, and other landscape ecology methods to seek an answer related to this prediction.

Approaches Used

Exercise 1: For the first exercise, different interpolation methods were used to explore their effects on the point trawls data. The interpolations (Kriging and IDW were tested) were used to model the abundance of rockfish in four selected years.

Exercise 2: Next, a neighborhood analysis was used to determine what the dominant seascapes were in the areas around trawls that produced high rockfish abundance (both according to the data and the interpolation) and low rockfish abundance. Buffers were set at 5, 10, and 25km distances around two trawls of each kind and seascape classes were measured as a function of % area of each buffer.

Exercise 3: Finally, a confusion matrix was calculated that measured the agreement between occurrence of the significant seascapes (as determined by exercise 2) and areas of high abundance (as determined by the interpolations in exercise 1).

Results

Exercise 1: The interpolations produced a number of maps of varying spatial extents and with varying resolution. Methods had to be refined multiple times to procure workable results, but maps interpolating rockfish abundance for 4 distinct years were eventually created and used.

Exercise 2: The neighborhood analysis produced statistics related to the percent of each buffer area occupied by each seascape class. This information was then used to create two plots – one for areas of high abundance and one for areas of low abundance. The most important result was an understanding of which seascape classes represented areas of high rockfish abundance. The dominant seascape classes in areas of high abundance turned out to match two of the seascape classes predicted in my hypothesis: Temperate Transition (Class 7) and Subpolar (Class 12).

Exercise 3: The final exercise produced a confusion matrix that measured true positives, true negatives, false positives, and false negatives as they related to the agreement between the seascape classes and the interpolated maps. The final matrix also produced an overall agreement score (Kappa).

Significance

The confusion matrix ended up being a bit of a failure, as both the seascape information and interpolated information had to be turned into presence-absence data for the matrix to be calculated. This eliminated some of the fine-tuned resolution associated with the interpolation and ultimately led to a Kappa score that suggested less agreement than would be expected by chance. While this specific result is not important to science or managers, these methods can be refined to improve the agreement between the two sources, ultimately providing a way to statistically link physical ocean conditions and rockfish abundances in California.

For these reasons, I consider this project to be an excellent pilot study for these data – one that introduces me to the types of analyses available to researchers who work with these types of data and their strengths and shortcomings.

Learning Outcomes

Software and Statistics

This class gave me an opportunity to brush up on my Arcmap and Modelbuilder skills, which I used to be very proficient in but had lost. I’ve also had the opportunity to explore some analyses through the lenses of R and Python, though I did not end up executing any parts of my project in these programs. The most important statistical learning outcomes for me had to do with interpolation and confusion matrices. The first exercise taught me all about the different ways point data can be interpolated and the consequences that that can have for results. I was familiar with the different types of hotspot analysis available to users in ArcGIS, but was not aware of their differences. The confusion matrix introduced me to an entirely new way to connect independent modeling methods. Methodologically, I learned about the importance of recording your methods as you execute processes and about the benefits of naming files sequentially. Had I not been organized, I would have had a hard time keeping track of the different processes executed and files used during each step.