Zhang Demirciyan 2025a - Revision history

Tomamil: Tomamil moved page Review 542427767720 to Zhang Demirciyan 2025a

2025-09-22T11:54:37Z

Tomamil moved page Review 542427767720 to Zhang Demirciyan 2025a

Cynthiazhang at 14:20, 7 August 2025

2025-08-07T14:20:38Z

Show changes

Cynthiazhang: Cynthiazhang moved page Draft Zhang 249907814 to Review 542427767720

2025-07-07T21:16:48Z

Cynthiazhang moved page Draft Zhang 249907814 to Review 542427767720

Cynthiazhang at 20:23, 7 July 2025

2025-07-07T20:23:51Z

Cynthiazhang at 20:22, 7 July 2025

2025-07-07T20:22:03Z

Cynthiazhang at 20:19, 7 July 2025

2025-07-07T20:19:42Z

Show changes

Cynthiazhang at 20:00, 7 July 2025

2025-07-07T20:00:29Z

Cynthiazhang at 19:57, 7 July 2025

2025-07-07T19:57:19Z

Cynthiazhang at 19:50, 7 July 2025

2025-07-07T19:50:53Z

Cynthiazhang at 19:48, 7 July 2025

2025-07-07T19:48:53Z

@@ Line 63: / Line 63: @@
 [[File:Draft_Zhang_249907814_3225_fig 4.png]]
-'''''Figure 4. Permutation importance of each environmental raster variable''' for each group’s CNN SDM. Error bars represent ± SD. - you also need to give more detail in your figure caption.''
+'''''Figure 4. Permutation importance of each environmental raster variable''' for each group’s CNN SDM. Error bars represent ± SD.''
 Features were randomly shuffled and the resulting decrease in accuracy score was measured to calculate permutation importances, which show the relative impact of each environmental raster variable on monarch occurrence predictions for each of the five groups.

@@ Line 20: / Line 20: @@
 ==== Species Occurrence Data and Group ====
 [[File:Draft_Zhang_249907814_7551_fig 1 (1).png]]
-'''''Figure 1. (a) Annual monarch butterfly distribution''''' ''from GBIF iNaturalist Research-grade observations (2010-2019) was grouped by '''(b) monthly distribution''' (Spring: Mar-May, Summer: May-Aug, Fall: Sep-Nov, Winter: Nov-Mar), then categorized into '''(c) five groups''' based on season, region, and grouping patterns from previous studies (11, 15). Figure generated by author.''
+'''''Figure 1. (a) Annual monarch butterfly distribution''''' ''from GBIF iNaturalist Research-grade observations (2010-2019) was grouped by '''(b) monthly distribution''' (Spring: Mar-May, Summer: May-Aug, Fall: Sep-Nov, Winter: Nov-Mar), then categorized into '''(c) five groups''' based on season, region, and grouping patterns from previous studies (''Oberhauser et al., 2016; Flockhart et al., 2013''). Figure generated by author.''
 <nowiki>Presence-only monarch butterfly (Danaus plexippus) occurrence data in the United States and Mexico from January 2010 to December 2019 was obtained from the Global Biodiversity Information Facility (GBIF) with their corresponding observation dates and latitude and longitude coordinates (GBIF.org (21 November 2024) GBIF Occurrence Download  https://doi.org/10.15468/dl.ybejfm). The dataset includes 46,588 occurrences from iNaturalist Research-grade Observations, which were selected and used for this study. Data points west of 106°W were dropped to eliminate potential presences of the western migratory monarch butterfly population that mostly resides west of the Rocky Mountains and overwinters along the coast of California.</nowiki>
@@ Line 105: / Line 105: @@
 ==References==

@@ Line 19: / Line 19: @@
 ==== Species Occurrence Data and Group ====
-[[File:Draft_Zhang_249907814_7551_fig 1 (1).png]]'''''Figure 1. (a) Annual monarch butterfly distribution''''' ''from GBIF iNaturalist Research-grade observations (2010-2019) was grouped by '''(b) monthly distribution''' (Spring: Mar-May, Summer: May-Aug, Fall: Sep-Nov, Winter: Nov-Mar), then categorized into '''(c) five groups''' based on season, region, and grouping patterns from previous studies (11, 15). Figure generated by author.''
+[[File:Draft_Zhang_249907814_7551_fig 1 (1).png]]
 <nowiki>Presence-only monarch butterfly (Danaus plexippus) occurrence data in the United States and Mexico from January 2010 to December 2019 was obtained from the Global Biodiversity Information Facility (GBIF) with their corresponding observation dates and latitude and longitude coordinates (GBIF.org (21 November 2024) GBIF Occurrence Download  https://doi.org/10.15468/dl.ybejfm). The dataset includes 46,588 occurrences from iNaturalist Research-grade Observations, which were selected and used for this study. Data points west of 106°W were dropped to eliminate potential presences of the western migratory monarch butterfly population that mostly resides west of the Rocky Mountains and overwinters along the coast of California.</nowiki>
@@ Line 31: / Line 32: @@
 ==== Convolutional Neural Network Species Distribution Models ====
-[[File:Draft_Zhang_249907814_7310_fig 2.png]]''Figure 2. Examples of '''(a) presences''' during the summer season (2010-2019) and corresponding '''(b) pseudo-absences''' generated. Figure generated by author.''
+[[File:Draft_Zhang_249907814_7310_fig 2.png]]
 For each of the five groups, pseudo-absence points were generated using a modified Surface Range Envelope (SRE) method at a 1:1 ratio with presence points, which was shown by Barbet-Massin et al. (2012) as the optimal method to maximize specificity or the true negative rate required for species conservation and reserve planning (24). Typically the SRE method randomly selects points outside of an envelope representing the species’ suitable range, where the envelope is defined by the maximum and minimum values of the environmental variables recorded at all species presence locations. However, this study uses environmental rasters around occurrence points, so pseudo-absence points were instead randomly selected from points with average environmental raster values falling outside the maximum and minimum values of those recorded at all species presences. The area of pseudo-absence selection was the contiguous United States west of the 106°W boundary for the Spring, Summer, Fall-North, and Fall-South groups; and Mexico for the Winter group.
-[[File:Draft_Zhang_249907814_6411_fig 3.png]]'''''Figure 3. Diagram of convolutional neural network species distribution model.''''' ''Figure generated by author.''
+[[File:Draft_Zhang_249907814_6411_fig 3.png]]
 Presence and pseudo-absence points were split into training (90%) and test (10%) sets. For each occurrence (presence or absence) point, its surrounding 64 x 64 pixel environmental neighborhood was extracted, with each pixel representing 1 km of data. For each occurrence, the input environmental tensor has a size of 64 x 64 x 8 pixels, containing a layer for each of the 8 environmental raster variables (19), where glyphosate use served as a predictor in only the United States groups (Spring, Summer, Fall-North, Fall-South), while forested land cover served as a predictor in only the Winter group. These 3-dimensional tensors underwent transformations, including convolution and pooling, to detect key features of the environmental rasters, creating 2-dimensional feature maps of lower dimensionality (19). The resulting 2-dimensional feature maps were further flattened into 1-dimensional feature vectors, which are fitted into dense neural networks with the binary cross-entropy loss function (Fig. 1). The ReLU and sigmoid activation functions were used for hidden layers and the output layer, respectively. The number of nodes and hidden layers for dense neural networks will be adjusted to decrease the loss of the training and test set and to maximize the CNN-SDMs’ true skill statistic (TSS) and area under curve of the receiver operating characteristic (AUC ROC) scores, which use sensitivity, or true positive rate, and specificity, or true negative rate, to indicate the ability of a binary classification model to correctly distinguish between positive and negative classes.
@@ Line 49: / Line 52: @@
 == Results ==
 ==== CNN-SDM Performance ====
 For each spatiotemporal group (Spring, Summer, Fall North, Fall South, Winter) representing spatiotemporal steps of the monarch annual cycle, a CNN-SDM was created to model and evaluate the ecological niche of the monarch butterfly throughout its migration. All CNN-SDMs except for the Fall South group had AUC-ROC and TSS statistics above 0.9, indicating high specificity and sensitivity in model predictions (Table 1).
 ==== Spatiotemporal Niche Shifts and Similarities ====
 '''''Figure 1. Permutation importance of each environmental raster variable''' for each group’s CNN SDM. Error bars represent ± SD. - you also need to give more detail in your figure caption.''
@@ Line 70: / Line 77: @@
 ==== Future Monarch Distributions for SSP-RCP Climate and Land Cover Scenarios ====
 '''''Figure 2. Modeled future distributions of (a) Spring, (b) Summer, (c) Fall North, (d) Fall South, and (e) Winter groups''' under future climate and land cover projections of three SSP-RCP scenarios — SSP1-2.6 (red), SSP2-4.5 (blue), SSP5-8.5 (turquoise) — and their resulting SHAP values for each environmental raster variable.''

@@ Line 19: / Line 19: @@
 ==== Species Occurrence Data and Group ====
-[[File:Draft_Zhang_249907814_7551_fig 1 (1).png]]
+[[File:Draft_Zhang_249907814_7551_fig 1 (1).png]]'''''Figure 1. (a) Annual monarch butterfly distribution''''' ''from GBIF iNaturalist Research-grade observations (2010-2019) was grouped by '''(b) monthly distribution''' (Spring: Mar-May, Summer: May-Aug, Fall: Sep-Nov, Winter: Nov-Mar), then categorized into '''(c) five groups''' based on season, region, and grouping patterns from previous studies (11, 15). Figure generated by author.''
-'''''Figure 3. (a) Annual monarch butterfly distribution''''' ''from GBIF iNaturalist Research-grade observations (2010-2019) was grouped by '''(b) monthly distribution''' (Spring: Mar-May, Summer: May-Aug, Fall: Sep-Nov, Winter: Nov-Mar), then categorized into '''(c) five groups''' based on season, region, and grouping patterns from previous studies (11, 15). Figure generated by student author.''
 <nowiki>Presence-only monarch butterfly (Danaus plexippus) occurrence data in the United States and Mexico from January 2010 to December 2019 was obtained from the Global Biodiversity Information Facility (GBIF) with their corresponding observation dates and latitude and longitude coordinates (GBIF.org (21 November 2024) GBIF Occurrence Download  https://doi.org/10.15468/dl.ybejfm). The dataset includes 46,588 occurrences from iNaturalist Research-grade Observations, which were selected and used for this study. Data points west of 106°W were dropped to eliminate potential presences of the western migratory monarch butterfly population that mostly resides west of the Rocky Mountains and overwinters along the coast of California.</nowiki>
@@ Line 32: / Line 31: @@
 ==== Convolutional Neural Network Species Distribution Models ====
-[[File:Draft_Zhang_249907814_7310_fig 2.png]]
+[[File:Draft_Zhang_249907814_7310_fig 2.png]]''Figure 2. Examples of '''(a) presences''' during the summer season (2010-2019) and corresponding '''(b) pseudo-absences''' generated. Figure generated by author.''
-Figure 4. Examples of (a) presences during the summer season (2010-2019) and corresponding (b) pseudo-absences generated. Figure generated by author.
 For each of the five groups, pseudo-absence points were generated using a modified Surface Range Envelope (SRE) method at a 1:1 ratio with presence points, which was shown by Barbet-Massin et al. (2012) as the optimal method to maximize specificity or the true negative rate required for species conservation and reserve planning (24). Typically the SRE method randomly selects points outside of an envelope representing the species’ suitable range, where the envelope is defined by the maximum and minimum values of the environmental variables recorded at all species presence locations. However, this study uses environmental rasters around occurrence points, so pseudo-absence points were instead randomly selected from points with average environmental raster values falling outside the maximum and minimum values of those recorded at all species presences. The area of pseudo-absence selection was the contiguous United States west of the 106°W boundary for the Spring, Summer, Fall-North, and Fall-South groups; and Mexico for the Winter group.
-[[File:Draft_Zhang_249907814_6411_fig 3.png]]
+[[File:Draft_Zhang_249907814_6411_fig 3.png]]'''''Figure 3. Diagram of convolutional neural network species distribution model.''''' ''Figure generated by author.''
-Figure 5. Diagram of convolutional neural network species distribution model. Figure generated by student author.
 Presence and pseudo-absence points were split into training (90%) and test (10%) sets. For each occurrence (presence or absence) point, its surrounding 64 x 64 pixel environmental neighborhood was extracted, with each pixel representing 1 km of data. For each occurrence, the input environmental tensor has a size of 64 x 64 x 8 pixels, containing a layer for each of the 8 environmental raster variables (19), where glyphosate use served as a predictor in only the United States groups (Spring, Summer, Fall-North, Fall-South), while forested land cover served as a predictor in only the Winter group. These 3-dimensional tensors underwent transformations, including convolution and pooling, to detect key features of the environmental rasters, creating 2-dimensional feature maps of lower dimensionality (19). The resulting 2-dimensional feature maps were further flattened into 1-dimensional feature vectors, which are fitted into dense neural networks with the binary cross-entropy loss function (Fig. 1). The ReLU and sigmoid activation functions were used for hidden layers and the output layer, respectively. The number of nodes and hidden layers for dense neural networks will be adjusted to decrease the loss of the training and test set and to maximize the CNN-SDMs’ true skill statistic (TSS) and area under curve of the receiver operating characteristic (AUC ROC) scores, which use sensitivity, or true positive rate, and specificity, or true negative rate, to indicate the ability of a binary classification model to correctly distinguish between positive and negative classes.
 To analyze the threats to monarch butterflies systematically across their annual cycle, a monarch butterfly CNN-SDM, as described above, was created for each of the five groups. To explain the ecological niche for each CNN-SDM, permutation importances were calculated for each environmental raster variable by randomly shuffling feature values and measuring the resulting decrease in accuracy score. Additionally, Shapley Additive Explanation (SHAP) values were calculated to assess each feature’s contribution on the model output for high and low feature values. Since the output is the probability of monarchs being present, the SHAP values will be a decimal from 0 to 1. Analysis of Variance (ANOVA) tests and the Holm-Bonferroni post-hoc test were performed to assess significance in permutation importance between groups for each variable.
 ==3 Bibliography<!--

@@ Line 18: / Line 18: @@
 == Methods ==
-==== Species Occurrence Data and Groups ====
+==== Species Occurrence Data and Group ====
 [[File:Draft_Zhang_249907814_7551_fig 1 (1).png]]
-==== Species Occurrence Data and Group[[File:Draft_Zhang_249907814_7551_fig 1 (1).png]] ====
 '''''Figure 3. (a) Annual monarch butterfly distribution''''' ''from GBIF iNaturalist Research-grade observations (2010-2019) was grouped by '''(b) monthly distribution''' (Spring: Mar-May, Summer: May-Aug, Fall: Sep-Nov, Winter: Nov-Mar), then categorized into '''(c) five groups''' based on season, region, and grouping patterns from previous studies (11, 15). Figure generated by student author.''
@@ Line 33: / Line 32: @@
 ==== Convolutional Neural Network Species Distribution Models ====
 For each of the five groups, pseudo-absence points were generated using a modified Surface Range Envelope (SRE) method at a 1:1 ratio with presence points, which was shown by Barbet-Massin et al. (2012) as the optimal method to maximize specificity or the true negative rate required for species conservation and reserve planning (24). Typically the SRE method randomly selects points outside of an envelope representing the species’ suitable range, where the envelope is defined by the maximum and minimum values of the environmental variables recorded at all species presence locations. However, this study uses environmental rasters around occurrence points, so pseudo-absence points were instead randomly selected from points with average environmental raster values falling outside the maximum and minimum values of those recorded at all species presences. The area of pseudo-absence selection was the contiguous United States west of the 106°W boundary for the Spring, Summer, Fall-North, and Fall-South groups; and Mexico for the Winter group.
 Presence and pseudo-absence points were split into training (90%) and test (10%) sets. For each occurrence (presence or absence) point, its surrounding 64 x 64 pixel environmental neighborhood was extracted, with each pixel representing 1 km of data. For each occurrence, the input environmental tensor has a size of 64 x 64 x 8 pixels, containing a layer for each of the 8 environmental raster variables (19), where glyphosate use served as a predictor in only the United States groups (Spring, Summer, Fall-North, Fall-South), while forested land cover served as a predictor in only the Winter group. These 3-dimensional tensors underwent transformations, including convolution and pooling, to detect key features of the environmental rasters, creating 2-dimensional feature maps of lower dimensionality (19). The resulting 2-dimensional feature maps were further flattened into 1-dimensional feature vectors, which are fitted into dense neural networks with the binary cross-entropy loss function (Fig. 1). The ReLU and sigmoid activation functions were used for hidden layers and the output layer, respectively. The number of nodes and hidden layers for dense neural networks will be adjusted to decrease the loss of the training and test set and to maximize the CNN-SDMs’ true skill statistic (TSS) and area under curve of the receiver operating characteristic (AUC ROC) scores, which use sensitivity, or true positive rate, and specificity, or true negative rate, to indicate the ability of a binary classification model to correctly distinguish between positive and negative classes.

← Older revision	Revision as of 11:54, 22 September 2025
(No difference)

← Older revision	Revision as of 21:16, 7 July 2025
(No difference)