-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy path05-aim3.html
More file actions
1469 lines (1436 loc) · 135 KB
/
Copy path05-aim3.html
File metadata and controls
1469 lines (1436 loc) · 135 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
<meta charset="utf-8">
<meta name="generator" content="quarto-1.5.57">
<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
<title>6 Proposed Deep Learning Framework for UOG Site Monitoring (Aim 3) – Portfolio</title>
<style>
code{white-space: pre-wrap;}
span.smallcaps{font-variant: small-caps;}
div.columns{display: flex; gap: min(4vw, 1.5em);}
div.column{flex: auto; overflow-x: auto;}
div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
ul.task-list{list-style: none;}
ul.task-list li input[type="checkbox"] {
width: 0.8em;
margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */
vertical-align: middle;
}
/* CSS for syntax highlighting */
pre > code.sourceCode { white-space: pre; position: relative; }
pre > code.sourceCode > span { line-height: 1.25; }
pre > code.sourceCode > span:empty { height: 1.2em; }
.sourceCode { overflow: visible; }
code.sourceCode > span { color: inherit; text-decoration: inherit; }
div.sourceCode { margin: 1em 0; }
pre.sourceCode { margin: 0; }
@media screen {
div.sourceCode { overflow: auto; }
}
@media print {
pre > code.sourceCode { white-space: pre-wrap; }
pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
}
pre.numberSource code
{ counter-reset: source-line 0; }
pre.numberSource code > span
{ position: relative; left: -4em; counter-increment: source-line; }
pre.numberSource code > span > a:first-child::before
{ content: counter(source-line);
position: relative; left: -1em; text-align: right; vertical-align: baseline;
border: none; display: inline-block;
-webkit-touch-callout: none; -webkit-user-select: none;
-khtml-user-select: none; -moz-user-select: none;
-ms-user-select: none; user-select: none;
padding: 0 4px; width: 4em;
}
pre.numberSource { margin-left: 3em; padding-left: 4px; }
div.sourceCode
{ }
@media screen {
pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
}
</style>
<script src="site_libs/quarto-nav/quarto-nav.js"></script>
<script src="site_libs/quarto-nav/headroom.min.js"></script>
<script src="site_libs/clipboard/clipboard.min.js"></script>
<script src="site_libs/quarto-search/autocomplete.umd.js"></script>
<script src="site_libs/quarto-search/fuse.min.js"></script>
<script src="site_libs/quarto-search/quarto-search.js"></script>
<meta name="quarto:offset" content="./">
<link href="./06-technical-report.html" rel="next">
<link href="./04-prelim-analysis.html" rel="prev">
<link href="./pic/cover.png" rel="icon" type="image/png">
<script src="site_libs/quarto-html/quarto.js"></script>
<script src="site_libs/quarto-html/popper.min.js"></script>
<script src="site_libs/quarto-html/tippy.umd.min.js"></script>
<script src="site_libs/quarto-html/anchor.min.js"></script>
<link href="site_libs/quarto-html/tippy.css" rel="stylesheet">
<link href="site_libs/quarto-html/quarto-syntax-highlighting.css" rel="stylesheet" id="quarto-text-highlighting-styles">
<script src="site_libs/bootstrap/bootstrap.min.js"></script>
<link href="site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
<link href="site_libs/bootstrap/bootstrap.min.css" rel="stylesheet" id="quarto-bootstrap" data-mode="light">
<script id="quarto-search-options" type="application/json">{
"location": "sidebar",
"copy-button": false,
"collapse-after": 3,
"panel-placement": "start",
"type": "textbox",
"limit": 50,
"keyboard-shortcut": [
"f",
"/",
"s"
],
"show-item-context": false,
"language": {
"search-no-results-text": "No results",
"search-matching-documents-text": "matching documents",
"search-copy-link-title": "Copy link to search",
"search-hide-matches-text": "Hide additional matches",
"search-more-match-text": "more match in this document",
"search-more-matches-text": "more matches in this document",
"search-clear-button-title": "Clear",
"search-text-placeholder": "",
"search-detached-cancel-button-title": "Cancel",
"search-submit-button-title": "Submit",
"search-label": "Search"
}
}</script>
<style>html{ scroll-behavior: smooth; }</style>
<script src="https://cdnjs.cloudflare.com/polyfill/v3/polyfill.min.js?features=es6"></script>
<script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
<script type="text/javascript">
const typesetMath = (el) => {
if (window.MathJax) {
// MathJax Typeset
window.MathJax.typeset([el]);
} else if (window.katex) {
// KaTeX Render
var mathElements = el.getElementsByClassName("math");
var macros = [];
for (var i = 0; i < mathElements.length; i++) {
var texText = mathElements[i].firstChild;
if (mathElements[i].tagName == "SPAN") {
window.katex.render(texText.data, mathElements[i], {
displayMode: mathElements[i].classList.contains('display'),
throwOnError: false,
macros: macros,
fleqn: false
});
}
}
}
}
window.Quarto = {
typesetMath
};
</script>
</head>
<body class="nav-sidebar docked">
<div id="quarto-search-results"></div>
<header id="quarto-header" class="headroom fixed-top">
<nav class="quarto-secondary-nav">
<div class="container-fluid d-flex">
<button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
<i class="bi bi-layout-text-sidebar-reverse"></i>
</button>
<nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item"><a href="./05-aim3.html"><span class="chapter-number">6</span> <span class="chapter-title">Proposed Deep Learning Framework for UOG Site Monitoring (Aim 3)</span></a></li></ol></nav>
<a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
</a>
<button type="button" class="btn quarto-search-button" aria-label="Search" onclick="window.quartoOpenSearch();">
<i class="bi bi-search"></i>
</button>
</div>
</nav>
</header>
<!-- content -->
<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article">
<!-- sidebar -->
<nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation docked overflow-auto">
<div class="pt-lg-2 mt-2 text-left sidebar-header">
<div class="sidebar-title mb-0 py-0">
<a href="./">Portfolio</a>
<div class="sidebar-tools-main">
<a href="" class="quarto-reader-toggle quarto-navigation-tool px-1" onclick="window.quartoToggleReader(); return false;" title="Toggle reader mode">
<div class="quarto-reader-toggle-btn">
<i class="bi"></i>
</div>
</a>
</div>
</div>
</div>
<div class="mt-2 flex-shrink-0 align-items-center">
<div class="sidebar-search">
<div id="quarto-search" class="" title="Search"></div>
</div>
</div>
<div class="sidebar-menu-container">
<ul class="list-unstyled mt-1">
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./index.html" class="sidebar-item-text sidebar-link">
<span class="menu-text">Preface</span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./intro.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">1</span> <span class="chapter-title">Introduction</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./01-proposal.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">2</span> <span class="chapter-title">Literature Analytics and Methodological Gap Analysis in Geospatial Epidemiology</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./02-data-collection.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">3</span> <span class="chapter-title">Data Collection</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./03-literature-review.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">4</span> <span class="chapter-title">Literature Review</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./04-prelim-analysis.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">5</span> <span class="chapter-title">Preliminary Analysis</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./05-aim3.html" class="sidebar-item-text sidebar-link active">
<span class="menu-text"><span class="chapter-number">6</span> <span class="chapter-title">Proposed Deep Learning Framework for UOG Site Monitoring (Aim 3)</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./06-technical-report.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">7</span> <span class="chapter-title">Technical Report</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./summary.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">8</span> <span class="chapter-title">Summary</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./references.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">9</span> <span class="chapter-title">References</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./update.html" class="sidebar-item-text sidebar-link">
<span class="menu-text">Project Update</span></a>
</div>
</li>
<li class="sidebar-item sidebar-item-section">
<div class="sidebar-item-container">
<a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-1" role="navigation" aria-expanded="true">
<span class="menu-text">Appendices</span></a>
<a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-1" role="navigation" aria-expanded="true" aria-label="Toggle section">
<i class="bi bi-chevron-right ms-2"></i>
</a>
</div>
<ul id="quarto-sidebar-section-1" class="collapse list-unstyled sidebar-section depth1 show">
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./appendix-a-etl-rag.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">A</span> <span class="chapter-title">appendix-a-etl-rag.html</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./appendix-b-lca-rf.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">B</span> <span class="chapter-title">Preliminary LCA Cover crop Analysis</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./appendix-c-hardware.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">C</span> <span class="chapter-title">Appendix C: Computational Hardware</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./appendix-d-neo4j-queries.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">D</span> <span class="chapter-title">Knowledge Graph Queries and Supporting Data</span></span></a>
</div>
</li>
<li class="sidebar-item">
<div class="sidebar-item-container">
<a href="./appendix-e-db-queries.html" class="sidebar-item-text sidebar-link">
<span class="menu-text"><span class="chapter-number">E</span> <span class="chapter-title">Database Queries</span></span></a>
</div>
</li>
</ul>
</li>
</ul>
</div>
</nav>
<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
<!-- margin-sidebar -->
<div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
<nav id="TOC" role="doc-toc" class="toc-active">
<h2 id="toc-title">Table of contents</h2>
<ul>
<li><a href="#introduction-and-foundational-acknowledgement" id="toc-introduction-and-foundational-acknowledgement" class="nav-link active" data-scroll-target="#introduction-and-foundational-acknowledgement"><span class="header-section-number">7</span> Introduction and Foundational Acknowledgement</a></li>
<li><a href="#rationale-overcoming-the-limitations-of-traditional-analyses" id="toc-rationale-overcoming-the-limitations-of-traditional-analyses" class="nav-link" data-scroll-target="#rationale-overcoming-the-limitations-of-traditional-analyses"><span class="header-section-number">8</span> Rationale: Overcoming the Limitations of Traditional Analyses</a></li>
<li><a href="#study-scope-and-future-directions" id="toc-study-scope-and-future-directions" class="nav-link" data-scroll-target="#study-scope-and-future-directions"><span class="header-section-number">9</span> Study Scope and Future Directions</a></li>
<li><a href="#potential-risks-and-limitations" id="toc-potential-risks-and-limitations" class="nav-link" data-scroll-target="#potential-risks-and-limitations"><span class="header-section-number">10</span> Potential Risks and Limitations</a></li>
<li><a href="#ethical-considerations-and-confidentiality" id="toc-ethical-considerations-and-confidentiality" class="nav-link" data-scroll-target="#ethical-considerations-and-confidentiality"><span class="header-section-number">11</span> Ethical Considerations and Confidentiality</a></li>
<li><a href="#model-interpretability-and-transparency" id="toc-model-interpretability-and-transparency" class="nav-link" data-scroll-target="#model-interpretability-and-transparency"><span class="header-section-number">12</span> Model Interpretability and Transparency</a></li>
<li><a href="#reproducibility-and-open-science" id="toc-reproducibility-and-open-science" class="nav-link" data-scroll-target="#reproducibility-and-open-science"><span class="header-section-number">13</span> Reproducibility and Open Science</a></li>
<li><a href="#phase-1-uog-site-identification-via-cnn-object-detection" id="toc-phase-1-uog-site-identification-via-cnn-object-detection" class="nav-link" data-scroll-target="#phase-1-uog-site-identification-via-cnn-object-detection"><span class="header-section-number">14</span> Phase 1: UOG Site Identification via CNN Object Detection</a>
<ul>
<li><a href="#software-tools-and-libraries" id="toc-software-tools-and-libraries" class="nav-link" data-scroll-target="#software-tools-and-libraries"><span class="header-section-number">14.1</span> Software Tools and Libraries</a></li>
<li><a href="#data-acquisition-and-preprocessing" id="toc-data-acquisition-and-preprocessing" class="nav-link" data-scroll-target="#data-acquisition-and-preprocessing"><span class="header-section-number">14.2</span> Data Acquisition and Preprocessing</a></li>
<li><a href="#training-data-creation" id="toc-training-data-creation" class="nav-link" data-scroll-target="#training-data-creation"><span class="header-section-number">14.3</span> Training Data Creation</a></li>
<li><a href="#model-architecture-and-training" id="toc-model-architecture-and-training" class="nav-link" data-scroll-target="#model-architecture-and-training"><span class="header-section-number">14.4</span> Model Architecture and Training</a></li>
</ul></li>
<li><a href="#phase-2-methane-monitoring-with-spatiotemporal-models" id="toc-phase-2-methane-monitoring-with-spatiotemporal-models" class="nav-link" data-scroll-target="#phase-2-methane-monitoring-with-spatiotemporal-models"><span class="header-section-number">15</span> Phase 2: Methane Monitoring with Spatiotemporal Models</a>
<ul>
<li><a href="#methane-data-acquisition" id="toc-methane-data-acquisition" class="nav-link" data-scroll-target="#methane-data-acquisition"><span class="header-section-number">15.1</span> Methane Data Acquisition</a></li>
<li><a href="#spatiotemporal-analysis-with-convlstm" id="toc-spatiotemporal-analysis-with-convlstm" class="nav-link" data-scroll-target="#spatiotemporal-analysis-with-convlstm"><span class="header-section-number">15.2</span> Spatiotemporal Analysis with ConvLSTM</a></li>
<li><a href="#leak-detection-and-site-attribution" id="toc-leak-detection-and-site-attribution" class="nav-link" data-scroll-target="#leak-detection-and-site-attribution"><span class="header-section-number">15.3</span> Leak Detection and Site Attribution</a></li>
</ul></li>
<li><a href="#phase-3-epidemiological-linkage-and-analysis" id="toc-phase-3-epidemiological-linkage-and-analysis" class="nav-link" data-scroll-target="#phase-3-epidemiological-linkage-and-analysis"><span class="header-section-number">16</span> Phase 3: Epidemiological Linkage and Analysis</a>
<ul>
<li><a href="#health-outcome-data-acquisition-of-restricted-access-microdata" id="toc-health-outcome-data-acquisition-of-restricted-access-microdata" class="nav-link" data-scroll-target="#health-outcome-data-acquisition-of-restricted-access-microdata"><span class="header-section-number">16.1</span> Health Outcome Data: Acquisition of Restricted-Access Microdata</a></li>
<li><a href="#analytical-approach-prediction-and-inference" id="toc-analytical-approach-prediction-and-inference" class="nav-link" data-scroll-target="#analytical-approach-prediction-and-inference"><span class="header-section-number">16.2</span> Analytical Approach: Prediction and Inference</a></li>
<li><a href="#predictive-modeling-and-pattern-discovery" id="toc-predictive-modeling-and-pattern-discovery" class="nav-link" data-scroll-target="#predictive-modeling-and-pattern-discovery"><span class="header-section-number">16.3</span> Predictive Modeling and Pattern Discovery</a></li>
<li><a href="#statistical-modeling-for-inference-and-effect-size-estimation" id="toc-statistical-modeling-for-inference-and-effect-size-estimation" class="nav-link" data-scroll-target="#statistical-modeling-for-inference-and-effect-size-estimation"><span class="header-section-number">16.4</span> Statistical Modeling for Inference and Effect Size Estimation</a>
<ul class="collapse">
<li><a href="#methodology-negative-binomial-regression" id="toc-methodology-negative-binomial-regression" class="nav-link" data-scroll-target="#methodology-negative-binomial-regression"><span class="header-section-number">16.4.1</span> Methodology: Negative Binomial Regression</a></li>
<li><a href="#confounder-control" id="toc-confounder-control" class="nav-link" data-scroll-target="#confounder-control"><span class="header-section-number">16.4.2</span> Confounder Control</a></li>
<li><a href="#full-model-specification" id="toc-full-model-specification" class="nav-link" data-scroll-target="#full-model-specification"><span class="header-section-number">16.4.3</span> Full Model Specification</a></li>
<li><a href="#key-outputs" id="toc-key-outputs" class="nav-link" data-scroll-target="#key-outputs"><span class="header-section-number">16.4.4</span> Key Outputs</a></li>
</ul></li>
</ul></li>
<li><a href="#model-evaluation-and-validation-metrics" id="toc-model-evaluation-and-validation-metrics" class="nav-link" data-scroll-target="#model-evaluation-and-validation-metrics"><span class="header-section-number">17</span> Model Evaluation and Validation Metrics</a>
<ul>
<li><a href="#phase-1-cnn-object-detector-quality" id="toc-phase-1-cnn-object-detector-quality" class="nav-link" data-scroll-target="#phase-1-cnn-object-detector-quality"><span class="header-section-number">17.1</span> Phase 1: CNN Object Detector Quality</a></li>
<li><a href="#phase-2-methane-model-quality" id="toc-phase-2-methane-model-quality" class="nav-link" data-scroll-target="#phase-2-methane-model-quality"><span class="header-section-number">17.2</span> Phase 2: Methane Model Quality</a></li>
<li><a href="#phase-3-epidemiological-model-quality" id="toc-phase-3-epidemiological-model-quality" class="nav-link" data-scroll-target="#phase-3-epidemiological-model-quality"><span class="header-section-number">17.3</span> Phase 3: Epidemiological Model Quality</a></li>
</ul></li>
<li><a href="#projected-timeline-and-milestones-for-aim-3" id="toc-projected-timeline-and-milestones-for-aim-3" class="nav-link" data-scroll-target="#projected-timeline-and-milestones-for-aim-3"><span class="header-section-number">18</span> Projected Timeline and Milestones for Aim 3</a>
<ul>
<li><a href="#year-1-focus-on-phase-1---cnn-model-development-and-site-inventory-creation" id="toc-year-1-focus-on-phase-1---cnn-model-development-and-site-inventory-creation" class="nav-link" data-scroll-target="#year-1-focus-on-phase-1---cnn-model-development-and-site-inventory-creation"><span class="header-section-number">18.1</span> Year 1: Focus on Phase 1 - CNN Model Development and Site Inventory Creation</a></li>
<li><a href="#year-2-spatiotemporal-analysis-health-data-acquisition-and-epidemiological-linkage" id="toc-year-2-spatiotemporal-analysis-health-data-acquisition-and-epidemiological-linkage" class="nav-link" data-scroll-target="#year-2-spatiotemporal-analysis-health-data-acquisition-and-epidemiological-linkage"><span class="header-section-number">18.2</span> Year 2: Spatiotemporal Analysis, Health Data Acquisition, and Epidemiological Linkage</a></li>
</ul></li>
</ul>
</nav>
</div>
<!-- main -->
<main class="content" id="quarto-document-content">
<header id="title-block-header" class="quarto-title-block default">
<div class="quarto-title">
<div class="quarto-title-block"><div><h1 class="title"><span id="sec-aim-three" class="quarto-section-identifier"><span class="chapter-number">6</span> <span class="chapter-title">Proposed Deep Learning Framework for UOG Site Monitoring (Aim 3)</span></span></h1><button type="button" class="btn code-tools-button" id="quarto-code-tools-source"><i class="bi"></i> Code</button></div></div>
</div>
<div class="quarto-title-meta">
</div>
</header>
<section id="introduction-and-foundational-acknowledgement" class="level1" data-number="7">
<h1 data-number="7"><span class="header-section-number">7</span> Introduction and Foundational Acknowledgement</h1>
<p>This chapter outlines the proposed methodology for Specific Aim 3: the development and application of a novel deep learning framework to identify, classify, and monitor unconventional oil and gas (UOG), or “fracking,” sites. The successful completion of the preceding project aims; engineering a sophisticated ETL+RAG pipeline and conducting a baseline machine learning analysis, directly highlighted the critical need for more advanced spatiotemporal models in environmental health. This aim is designed to address that identified gap head-on.</p>
<p>The conceptualization and design of the methodologies presented here were made possible by the foundational knowledge acquired in the “AI with Geospatial Big Data” (GOG 594) course at the University at Albany. The course, taught not only the theoretical underpinnings of advanced neural networks but also the practical, hands-on skills necessary to apply them to complex, real-world geospatial problems. Its approach to data processing, model implementation, and the use of tools like Quarto for reproducible research provided the essential toolkit for envisioning and structuring this final stage of the dissertation.</p>
<p>The objective of this aim is twofold. First, to create a high-resolution, actionable dataset of UOG infrastructure and activity by applying state-of-the-art computer vision techniques to satellite imagery. Second, to leverage this dataset to monitor associated environmental indicators, specifically methane emissions, providing a new layer of data for future epidemiological studies. To ensure methodological depth and feasibility within the doctoral timeline, this chapter will develop and validate this entire framework through a case study of UOG development in the Commonwealth of Pennsylvania, a region with a long history of activity and excellent data availability.</p>
<p>The scientific novelty of this research is threefold. First, it pioneers a multi-modal deep learning framework that moves beyond static well counts to create a dynamic, activity-based UOG exposure inventory from satellite imagery. Second, by linking this high-resolution exposure data to census-tract level health microdata, it directly confronts and mitigates the core methodological challenges of ecological fallacy and exposure misclassification that have limited previous studies. Finally, it integrates these components within a hybrid analytical framework, using AI for predictive discovery and established statistical models for formal inference, providing a comprehensive and robust model for environmental health risk assessment.</p>
</section>
<section id="rationale-overcoming-the-limitations-of-traditional-analyses" class="level1" data-number="8">
<h1 data-number="8"><span class="header-section-number">8</span> Rationale: Overcoming the Limitations of Traditional Analyses</h1>
<p>The impetus for this aim stems directly from the findings and limitations identified in a <a href="./04-prelim-analysis.html#sec-prelim-analysis">preliminary analysis of environmental exposures and health outcomes</a>. That baseline study, which used a Random Forest model to examine county-level ozone exposure and cardiovascular disease (CVD) mortality, was successful in establishing a statistical association. More importantly, however, it served as a crucial diagnostic, revealing the methodological barriers inherent in using aggregated, publicly available data for environmental health research.</p>
<p>Specifically, the preliminary work highlighted three challenges that this aim is designed to overcome:</p>
<ul>
<li><strong>The Ecological Fallacy:</strong> By operating at the county level, the analysis could only establish that counties with higher average exposures also had higher mortality rates. It was impossible to determine if the individuals who died were the ones most exposed, a fundamental limitation known as the ecological fallacy.</li>
<li><strong>Exposure Misclassification:</strong> Using county-wide averages for exposure metrics masks the vast differences in exposure that occur within a large geographic area. Residents living next to a source are grouped with those living miles away, leading to significant exposure misclassification that biases results, typically towards finding no effect.</li>
<li><strong>Artifacts from Smoothed Health Data:</strong> The preliminary analysis relied on publicly available, “spatiotemporally smoothed” mortality rates. This statistical smoothing, while useful for stabilizing rates in small populations, artificially introduces spatial and temporal patterns into the data. Advanced spatiotemporal models, such as the ones proposed here, risk simply “re-discovering” the signature of this smoothing algorithm rather than a true environmental health signal.</li>
</ul>
<p>Therefore, Aim 3 is predicated on a foundational principle: to robustly investigate the health impacts of localized environmental exposures like UOG development, one must create a high-resolution exposure dataset and link it to granular, unsmoothed health microdata. This chapter details the creation of the former, while the subsequent phase outlines the plan to acquire and analyze the latter. In doing so, this research presents a significant methodological advancement designed to produce a more accurate and granular understanding of the potential health risks of UOG development.</p>
</section>
<section id="study-scope-and-future-directions" class="level1" data-number="9">
<h1 data-number="9"><span class="header-section-number">9</span> Study Scope and Future Directions</h1>
<p>The scope of this chapter is to develop, execute, and validate the complete, end-to-end methodology within the defined study area of Pennsylvania. This includes the creation of a UOG inventory for the state and the subsequent epidemiological linkage with Pennsylvania health data.</p>
<p>This approach allows for a deep analysis, ensuring the development of a well-validated system. The successful completion of this state-level case study will serve as a pilot project and proof-of-concept. The validated models, established data pipelines, and initial findings will form the preliminary data for a future, national-scale research initiative, which would be the subject of a post-doctoral project or a grant proposal.</p>
</section>
<section id="potential-risks-and-limitations" class="level1" data-number="10">
<h1 data-number="10"><span class="header-section-number">10</span> Potential Risks and Limitations</h1>
<p>Although carefully designed, potential risks are anticipated. Data acquisition delays from satellite imagery providers or health data access could impact project timelines. Annotation workflows may require more time or resources than initially planned, particularly if annotation complexity is higher than expected. Computational constraints, such as GPU bottlenecks during model training, may necessitate additional optimization or hardware adjustments. Contingency plans, such as leveraging alternative imagery sources, adjusting annotation sample sizes, or optimizing GPU usage through batch size tuning, are prepared to mitigate these challenges.</p>
</section>
<section id="ethical-considerations-and-confidentiality" class="level1" data-number="11">
<h1 data-number="11"><span class="header-section-number">11</span> Ethical Considerations and Confidentiality</h1>
<p>Given the use of restricted-access health microdata, ethical considerations regarding data privacy and confidentiality are paramount. This project will rigorously adhere to NCHS confidentiality requirements, IRB guidelines, and best practices for data security. All sensitive analyses will be conducted in secure environments, ensuring individual privacy and confidentiality through aggregate reporting only.</p>
</section>
<section id="model-interpretability-and-transparency" class="level1" data-number="12">
<h1 data-number="12"><span class="header-section-number">12</span> Model Interpretability and Transparency</h1>
<p>Model interpretability is essential to validate predictive accuracy and ensure methodological transparency. Techniques such as SHAP (SHapley Additive exPlanations) will be applied to quantify and visualize the contribution of different predictors, as previously demonstrated in the <a href="./04-prelim-analysis.html#sec-prelim-results">preliminary analysis</a>. For CNN and GNN architectures, visual methods such as Grad-CAM or attention maps will be employed to demonstrate model decision-making processes transparently.</p>
</section>
<section id="reproducibility-and-open-science" class="level1" data-number="13">
<h1 data-number="13"><span class="header-section-number">13</span> Reproducibility and Open Science</h1>
<p>This research commits to open science and reproducibility principles. All analytical workflows, scripts, and models will be documented thoroughly using reproducible Quarto notebooks and made publicly available through GitHub repositories, facilitating validation, reuse, and future research.</p>
<hr>
</section>
<section id="phase-1-uog-site-identification-via-cnn-object-detection" class="level1" data-number="14">
<h1 data-number="14"><span class="header-section-number">14</span> Phase 1: UOG Site Identification via CNN Object Detection</h1>
<div id="fig-well-concept" class="quarto-float quarto-figure quarto-figure-left anchored">
<figure class="quarto-float quarto-float-fig figure">
<div aria-describedby="fig-well-concept-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
<img src="figures\contrst.PNG" class="img-fluid figure-img">
</div>
<figcaption class="quarto-float-caption-bottom quarto-float-caption quarto-float-fig" id="fig-well-concept-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
Figure 14.1: Well pad Concept
</figcaption>
</figure>
</div>
<section id="software-tools-and-libraries" class="level2" data-number="14.1">
<h2 data-number="14.1" class="anchored" data-anchor-id="software-tools-and-libraries"><span class="header-section-number">14.1</span> Software Tools and Libraries</h2>
<p>This project leverages open-source software tools to ensure reproducibility, efficiency, and ease of collaboration:</p>
<ul>
<li><strong>PyTorch / PyTorch Lightning:</strong> efficient neural network training and inference</li>
<li><strong>Ultralytics YOLOv8:</strong> framework for CNN-based object detection</li>
<li><strong>Hugging Face Model Hub:</strong> Source of pre-trained models, including YOLOv8 checkpoints for transfer learning</li>
<li><strong>CVAT (Computer Vision Annotation Tool):</strong> software for bounding box annotation and iterative labeling workflows</li>
<li><strong>GDAL, rasterio, geopandas:</strong> Tools for geospatial data acquisition, preprocessing, and spatial masking workflows</li>
</ul>
<p>These tools ensure reproducibility, efficiency, and ease of collaboration, enabling integration of complex components within the proposed methodological framework.</p>
</section>
<section id="data-acquisition-and-preprocessing" class="level2" data-number="14.2">
<h2 data-number="14.2" class="anchored" data-anchor-id="data-acquisition-and-preprocessing"><span class="header-section-number">14.2</span> Data Acquisition and Preprocessing</h2>
<p><strong>Satellite Imagery:</strong><br>
We will acquire high-resolution multispectral satellite imagery (from Sentinel-2, Planet Labs, or NAIP) for regions with known UOG development.</p>
<p><strong>Training Data Scaffolding (Permit-Guided):</strong><br>
To efficiently generate a dataset for model training, this project will leverage publicly available oil and gas well permit databases from state regulatory agencies. The latitude and longitude coordinates from these official records will be used to programmatically clip thousands of small image “snippets” from the satellite imagery. This approach ensures the manual annotation effort is focused on relevant imagery containing known positive examples, forming the foundation for the training process described in the next section.</p>
<p><strong>Inference Search Area Definition (Tax Parcel-Guided):</strong><br>
For the final model deployment phase, where the goal is to discover all sites (both known and unknown), a broad search area will be defined using county tax parcel data. This data will create a “mask” to programmatically exclude areas where UOG development is highly improbable (e.g., dense residential zones, protected lands). Running the model over this large, plausible search area is what enables the discovery of unrecorded or legacy sites.</p>
</section>
<section id="training-data-creation" class="level2" data-number="14.3">
<h2 data-number="14.3" class="anchored" data-anchor-id="training-data-creation"><span class="header-section-number">14.3</span> Training Data Creation</h2>
<p>Creating a large, accurate, and diverse training dataset is critical, yet manual annotation at a national scale is not possible for a single researcher. Therefore, a modern, “human-in-the-loop” workflow will be employed to make this large-scale labeling task manageable and to produce a highly robust final training set. The process is structured as follows:</p>
<ul>
<li><p><strong>Seed Dataset Creation (Manual Annotation):</strong><br>
A manageable initial subset of image snippets (2,000–5,000) will be labeled by hand in the CVAT annotation tool to create the initial high-quality ground truth.</p></li>
<li><p><strong>Initial Model Training (V1):</strong><br>
This seed dataset will be used to train a preliminary version (V1) of the YOLO/EfficientDet object detection model.</p></li>
<li><p><strong>Semi-Automated Labeling:</strong><br>
The V1 model will then be deployed on the remaining tens of thousands of unlabeled snippets to generate “pre-labels.” The human annotator’s role then shifts to that of a reviewer, rapidly confirming correct predictions and correcting inaccurate ones.</p></li>
<li><p><strong>Final “Expert” Model Training:</strong><br>
The full, high-quality dataset comprising both the manually labeled seed set and the much larger set of human-corrected semi-automated labels will be used to train the final, “expert” version of the model, likely by fine-tuning the V1 model.</p></li>
</ul>
<p>This entire iterative process will be used to draw bounding boxes around key features, including:</p>
<ul>
<li><strong>Well Pads:</strong> The overall footprint of the development site.</li>
<li><strong>Storage Tanks:</strong> Identifying individual or clusters of tanks.</li>
<li><strong>Active Equipment:</strong> Such as drilling rigs or vehicle fleets, which serve as a proxy for site activity.</li>
<li><strong>Abandoned Sites:</strong> Identifying sites with visible infrastructure but no signs of recent activity.</li>
</ul>
<p>To ensure the objectivity and reproducibility of these labels, an <strong>Inter-Annotator Agreement (IAA)</strong> analysis will be performed on a subset of the data. By having a second independent annotator label the same set of images and calculating metrics such as Intersection over Union (IoU) for bounding boxes and Cohen’s Kappa for categorical labels, the reliability of the annotation process will be quantitatively validated.</p>
<hr>
</section>
<section id="model-architecture-and-training" class="level2" data-number="14.4">
<h2 data-number="14.4" class="anchored" data-anchor-id="model-architecture-and-training"><span class="header-section-number">14.4</span> Model Architecture and Training</h2>
<p>Detailed information on the specialized computational hardware used for training and inference (multi-GPU workstation with 48GB GPUs) is provided in <a href="./appendix-c-hardware.html#sec-appendix-c-hardware">Appendix C: Computational Hardware</a>. Leveraging these dedicated computational resources, we will employ a transfer learning approach, which is standard practice for complex object detection tasks. Specifically, a state-of-the-art YOLO (You Only Look Once) variant, such as YOLOv8 pre-trained on the COCO dataset, will be fine-tuned on our annotated UOG site dataset. This approach enables the model to leverage its existing knowledge of general shapes and textures, adapting rapidly to the specific task of identifying UOG infrastructure. Utilizing our robust computational setup significantly accelerates training times, facilitates larger batch sizes, and ensures high-quality, generalizable model performance.</p>
<ul>
<li><p><strong>Data Curation:</strong><br>
The training dataset will be constructed using stratified sampling to ensure representation from different regions within Pennsylvania’s Appalachian Basin (the wet gas southwestern region vs. the dry gas northeastern region) and across different seasons (leaf-on vs. leaf-off). This approach ensures the model is robust for the case study area, while the methodology itself is designed to be directly scalable to other basins in future work.</p></li>
<li><p><strong>Data Augmentation:</strong><br>
The training data will be programmatically augmented by applying random geometric (flips, rotations) and photometric (brightness, contrast, hue) transformations. This forces the model to learn the fundamental features of UOG infrastructure rather than superficial characteristics tied to a specific region or season.</p></li>
</ul>
<hr>
</section>
</section>
<section id="phase-2-methane-monitoring-with-spatiotemporal-models" class="level1" data-number="15">
<h1 data-number="15"><span class="header-section-number">15</span> Phase 2: Methane Monitoring with Spatiotemporal Models</h1>
<section id="methane-data-acquisition" class="level2" data-number="15.1">
<h2 data-number="15.1" class="anchored" data-anchor-id="methane-data-acquisition"><span class="header-section-number">15.1</span> Methane Data Acquisition</h2>
<p>We will acquire publicly available methane concentration data from sources such as the Sentinel-5P TROPOMI instrument or the more recent MethaneSAT. This provides regular, global coverage of atmospheric methane levels.</p>
</section>
<section id="spatiotemporal-analysis-with-convlstm" class="level2" data-number="15.2">
<h2 data-number="15.2" class="anchored" data-anchor-id="spatiotemporal-analysis-with-convlstm"><span class="header-section-number">15.2</span> Spatiotemporal Analysis with ConvLSTM</h2>
<p>To model the movement and concentration of methane plumes over time, we will use a spatiotemporal deep learning model such as a ConvLSTM (Convolutional Long Short-Term Memory) network. This architecture is ideal for this task because:</p>
<ul>
<li>The <strong>Convolutional (CNN)</strong> component analyzes the spatial structure of the methane plume in each individual satellite image.</li>
<li>The <strong>Recurrent (LSTM)</strong> component analyzes the sequence of these images over time, learning how plumes emerge, move with wind patterns, and dissipate.</li>
</ul>
<p>This model will be trained to predict methane concentrations across our study region.</p>
</section>
<section id="leak-detection-and-site-attribution" class="level2" data-number="15.3">
<h2 data-number="15.3" class="anchored" data-anchor-id="leak-detection-and-site-attribution"><span class="header-section-number">15.3</span> Leak Detection and Site Attribution</h2>
<p>By overlaying the outputs from the ConvLSTM model with the geolocated UOG site inventory from Phase 1, we can systematically:</p>
<ul>
<li><strong>Identify Anomalous Methane Plumes:</strong> Detect methane concentrations that are significantly higher than the regional baseline.</li>
<li><strong>Attribute Emissions:</strong> Correlate these plumes with specific active UOG sites to monitor their emissions.</li>
<li><strong>Detect Leaks from Abandoned Sites:</strong> Critically, we can also identify significant methane sources originating from sites previously classified as abandoned, providing a powerful tool for environmental monitoring and public health risk assessment.</li>
</ul>
<p>The final output of this aim will be a dynamic, spatially explicit dataset linking UOG infrastructure to its real-world methane emissions, creating a novel and valuable resource for the environmental health sciences community.</p>
<hr>
</section>
</section>
<section id="phase-3-epidemiological-linkage-and-analysis" class="level1" data-number="16">
<h1 data-number="16"><span class="header-section-number">16</span> Phase 3: Epidemiological Linkage and Analysis</h1>
<p>The UOG and methane emissions inventory created in Phases 1 and 2 will serve as a high-resolution spatiotemporal exposure dataset. The final phase of this aim is to link this environmental data with health outcome data to investigate potential associations. This requires acquiring restricted-access health microdata and applying appropriate statistical methods to control for confounding variables.</p>
<hr>
<section id="health-outcome-data-acquisition-of-restricted-access-microdata" class="level2" data-number="16.1">
<h2 data-number="16.1" class="anchored" data-anchor-id="health-outcome-data-acquisition-of-restricted-access-microdata"><span class="header-section-number">16.1</span> Health Outcome Data: Acquisition of Restricted-Access Microdata</h2>
<p>To overcome the limitations of publicly available aggregated data, this research will use individual-level, anonymized health data containing fine-scale geographic identifiers. The target dataset is the National Center for Health Statistics (NCHS) National Vital Statistics System (NVSS) Mortality Data. This dataset contains individual death records, including cause of death and, critically, the census tract of residence.</p>
<p>Acquiring this data is a formal process that must be conducted through the New York Federal Statistical Research Data Center (NYRDC), of which the University at Albany is a member. The process involves the following key steps:</p>
<ol type="1">
<li><p><strong>Proposal Development:</strong><br>
A detailed research proposal will be submitted to the NYRDC. This proposal will outline the research questions, the specific data variables required (year of death, census tract, cause of death, age, race, sex), the statistical methodology, and a plan for ensuring data confidentiality.</p></li>
<li><p><strong>Institutional Review Board (IRB) Review:</strong><br>
The proposal must be approved by the University at Albany’s IRB. This ensures the research is conducted ethically and that all protocols for protecting sensitive human subjects’ data are in place.</p></li>
<li><p><strong>Execution of a Data Use Agreement (DUA):</strong><br>
Upon provisional approval, a DUA from NCHS will be processed through the University’s Office of Research and Regulatory Compliance. This legal agreement governs the terms of data use, storage, and security.</p></li>
<li><p><strong>Secure Data Analysis:</strong><br>
All analyses using the restricted microdata will be performed within the secure environment of an NYRDC facility to ensure compliance with all federal data protection requirements.</p></li>
</ol>
<hr>
</section>
<section id="analytical-approach-prediction-and-inference" class="level2" data-number="16.2">
<h2 data-number="16.2" class="anchored" data-anchor-id="analytical-approach-prediction-and-inference"><span class="header-section-number">16.2</span> Analytical Approach: Prediction and Inference</h2>
<p>To fully address the research questions, a hybrid analytical framework will be employed, leveraging the distinct strengths of deep learning for prediction and discovery, and established statistical models for inference. This two-pronged strategy allows the project to predict where health risks may be highest and quantify the magnitude of those risks in a more traditional statistically rigorous manner.</p>
<hr>
</section>
<section id="predictive-modeling-and-pattern-discovery" class="level2" data-number="16.3">
<h2 data-number="16.3" class="anchored" data-anchor-id="predictive-modeling-and-pattern-discovery"><span class="header-section-number">16.3</span> Predictive Modeling and Pattern Discovery</h2>
<p>The primary objective of this phase is to achieve the highest possible predictive accuracy for health outcomes and to uncover complex, non-linear spatiotemporal patterns that traditional models often miss.</p>
<ul>
<li><strong>Methodology:</strong> The high-resolution UOG and methane exposure datasets (from Phases 1 & 2) will be used as inputs for the trained Convolutional Neural Network (CNN) and Graph Neural Network (GNN) architectures. These models are specifically designed to learn from the spatial grid and network structures of the data, capturing localized patterns and inter-county dependencies (i.e., spatial spillover effects).</li>
<li><strong>Key Outputs:</strong>
<ul>
<li><strong>Predictive Risk Maps:</strong> Generation of spatially explicit maps identifying census tracts predicted to have the highest health risk in future time periods.<br>
</li>
<li><strong>AI-Driven Feature Importance:</strong> Application of model interpretation techniques (SHAP) to visualize the non-linear effects and complex interactions discovered by the deep learning models.</li>
</ul></li>
</ul>
<hr>
</section>
<section id="statistical-modeling-for-inference-and-effect-size-estimation" class="level2" data-number="16.4">
<h2 data-number="16.4" class="anchored" data-anchor-id="statistical-modeling-for-inference-and-effect-size-estimation"><span class="header-section-number">16.4</span> Statistical Modeling for Inference and Effect Size Estimation</h2>
<p>The objective of this phase is to complement the predictive findings by quantifying the specific, interpretable effect size of UOG exposure on mortality while controlling for a wide range of confounders.</p>
<section id="methodology-negative-binomial-regression" class="level3" data-number="16.4.1">
<h3 data-number="16.4.1" class="anchored" data-anchor-id="methodology-negative-binomial-regression"><span class="header-section-number">16.4.1</span> Methodology: Negative Binomial Regression</h3>
<p>The relationship between UOG exposure and health outcomes (mortality counts) will be analyzed using Negative Binomial regression. This type of Generalized Linear Model (GLM) is standard for analyzing count data that exhibits overdispersion (i.e., where the variance in the data is greater than the mean), which is common in epidemiological data.</p>
<p>The selection of a Negative Binomial model over more conventional methods like simple Linear Regression or ANOVA is a deliberate choice dictated by the fundamental nature of the outcome data. Linear models, including ANOVA, are inappropriate for this analysis as they assume a continuous, normally distributed outcome with constant variance.</p>
<p>Applying them to mortality counts which are discrete, non-negative, and exhibit increasing variance with the mean, would violate these core assumptions and could lead to biased estimates and impossible predictions, such as negative death counts. The Negative Binomial framework is purpose-built to overcome these challenges, properly handling count data and its characteristic overdispersion while allowing for the estimation of interpretable risk ratios.</p>
<p>A Negative Binomial model is defined by two key components:</p>
<ol type="1">
<li><p><strong>The Probability Distribution</strong><br>
The number of deaths <span class="math inline">\((y)\)</span> in a given census tract is assumed to follow a Negative Binomial distribution. This distribution is defined by its mean <span class="math inline">\((\mu)\)</span> and a dispersion parameter <span class="math inline">\((\theta)\)</span> that models the extra variation not captured by a simpler Poisson model. The probability of observing exactly <span class="math inline">\((y)\)</span> deaths is given by the Probability Mass Function (PMF):</p>
<p><span class="math display">\[
P(Y=y \mid \mu,\theta) = \frac{\Gamma(y+\theta)}{\Gamma(y+1)\,\Gamma(\theta)} \left(\frac{\theta}{\theta+\mu}\right)^{\theta} \left(\frac{\mu}{\theta+\mu}\right)^{y}
\]</span></p>
<p>where <span class="math inline">\(y\)</span> is the observed mortality count, <span class="math inline">\(\mu\)</span> is the expected mean count for an observation, <span class="math inline">\(\theta\)</span> is the dispersion parameter, and <span class="math inline">\(\Gamma()\)</span> is the gamma function.</p></li>
<li><p><strong>The Link Function and Regression Structure</strong><br>
The model connects the expected mean count <span class="math inline">\((\mu_i)\)</span> for each census tract <span class="math inline">\((i)\)</span> to a set of predictor variables <span class="math inline">\((X)\)</span> using a natural logarithm link function. This ensures the predicted counts are always positive and that the predictors have a multiplicative effect on the outcome. The general form of the regression equation is:</p>
<p><span class="math display">\[
\log(\mu_i) = \beta_0 + \beta_1 X_{i1} + \beta_2 X_{i2} + \dots + \beta_p X_{ip}
\]</span></p>
<p>where <span class="math inline">\(\beta_0\)</span> is the intercept and <span class="math inline">\(\beta_1, \dots, \beta_p\)</span> are the regression coefficients representing the change in the log of the mean count for a one-unit change in each predictor.</p></li>
</ol>
<p>Insights from the AI model’s SHAP analysis will directly inform the structure of this regression model. For example, identified non-linear exposure-response relationships will be modeled using flexible terms such as restricted cubic splines to ensure the inferential model is as robust and realistic as possible.</p>
</section>
<section id="confounder-control" class="level3" data-number="16.4.2">
<h3 data-number="16.4.2" class="anchored" data-anchor-id="confounder-control"><span class="header-section-number">16.4.2</span> Confounder Control</h3>
<p>To account for other factors that influence health, census-tract level covariates will be incorporated from the CDC’s Social Vulnerability Index (SVI) (data on poverty, education, income). Individual-level demographic variables available in the restricted health data (age, race, sex) will be included directly in the model.</p>
</section>
<section id="full-model-specification" class="level3" data-number="16.4.3">
<h3 data-number="16.4.3" class="anchored" data-anchor-id="full-model-specification"><span class="header-section-number">16.4.3</span> Full Model Specification</h3>
<p>To properly analyze mortality <strong>rates</strong> rather than raw counts, the model must account for the population size in each census tract. This is accomplished by including the logarithm of the population as an offset term, which effectively acts as a predictor with a coefficient fixed at 1. The full conceptual model for this analysis is:</p>
<p><span class="math display">\[
\log(\mu_i) = \beta_0 + \beta_1\,s(\text{UOG Exposure}_i) + \sum_{j=2}^{p} \beta_j\,X_{ij} \;+\; \text{offset}\!\bigl(\log(\text{Population}_i)\bigr)
\]</span></p>
<p>Here, <span class="math inline">\(s(\text{UOG Exposure}_i)\)</span> represents the potentially non-linear term for UOG exposure, and <span class="math inline">\(\sum_{j=2}^{p} \beta_j\,X_{ij}\)</span> represents all other confounders (demographic and socioeconomic). This formulation is equivalent to modeling the mortality rate <span class="math inline">\((\mu_i/\text{Population}_i)\)</span>, and the resulting coefficients <span class="math inline">\((\beta_j)\)</span> are exponentiated <span class="math inline">\((e^{\beta_j})\)</span> to yield interpretable Risk Ratios (RR).</p>
</section>
<section id="key-outputs" class="level3" data-number="16.4.4">
<h3 data-number="16.4.4" class="anchored" data-anchor-id="key-outputs"><span class="header-section-number">16.4.4</span> Key Outputs</h3>
<ul>
<li><strong>Risk Ratios (RR):</strong> A quantifiable measure of association (the percent increase in mortality risk for each unit increase in UOG exposure), complete with 95% confidence intervals and p-values for formal hypothesis testing.</li>
</ul>
<p>By integrating these two approaches, this research will not only predict where risk is highest but also explain how much risk is attributable to UOG exposure, providing a comprehensive and actionable set of findings.</p>
<hr>
</section>
</section>
</section>
<section id="model-evaluation-and-validation-metrics" class="level1" data-number="17">
<h1 data-number="17"><span class="header-section-number">17</span> Model Evaluation and Validation Metrics</h1>
<p>An evaluation plan will be employed to quantify the performance of each model component to ensure the reliability of the final results.</p>
<section id="phase-1-cnn-object-detector-quality" class="level2" data-number="17.1">
<h2 data-number="17.1" class="anchored" data-anchor-id="phase-1-cnn-object-detector-quality"><span class="header-section-number">17.1</span> Phase 1: CNN Object Detector Quality</h2>
<p>The quality of the UOG site detector will be assessed using standard computer vision metrics on a held-out test set:</p>
<ul>
<li><strong>Intersection over Union (IoU):</strong> Measures the accuracy of predicted bounding boxes against ground-truth boxes.</li>
<li><strong>Precision & Recall:</strong> Measure the rates of false positives and false negatives, respectively.</li>
<li><strong>Mean Average Precision (mAP):</strong> The primary, single-number metric that summarizes overall performance across all object classes and detection thresholds.</li>
</ul>
</section>
<section id="phase-2-methane-model-quality" class="level2" data-number="17.2">
<h2 data-number="17.2" class="anchored" data-anchor-id="phase-2-methane-model-quality"><span class="header-section-number">17.2</span> Phase 2: Methane Model Quality</h2>
<p>The spatiotemporal ConvLSTM model for methane prediction is a regression task and will be evaluated using:</p>
<ul>
<li><strong>Root Mean Squared Error (RMSE) & Mean Absolute Error (MAE):</strong> To quantify the average prediction error in parts-per-billion (ppb).</li>
<li><strong>R-squared (R²):</strong> To measure the proportion of variance in the real methane data explained by the model.</li>
</ul>
</section>
<section id="phase-3-epidemiological-model-quality" class="level2" data-number="17.3">
<h2 data-number="17.3" class="anchored" data-anchor-id="phase-3-epidemiological-model-quality"><span class="header-section-number">17.3</span> Phase 3: Epidemiological Model Quality</h2>
<p>The two prongs of the health analysis require different evaluation methods:</p>
<ul>
<li><strong>Predictive AI Models (CNN/GNN):</strong> Performance will be measured using RMSE, MAE, R², and Pearson Correlation between the predicted and actual health outcomes on a held-out test set of census tracts/years.</li>
<li><strong>Inferential Statistical Model (Negative Binomial):</strong> The focus is on model fit and the validity of the conclusions. Quality will be assessed via:
<ul>
<li><strong>Goodness-of-Fit Statistics (AIC/BIC):</strong> To compare model variations and select the best-fitting specification.</li>
<li><strong>Significance and Confidence Intervals:</strong> The p-value and 95% CI for the UOG exposure term will determine the statistical significance and precision of the primary finding.</li>
<li><strong>Residual Analysis:</strong> To validate the model’s underlying statistical assumptions.</li>
</ul></li>
</ul>
<hr>
</section>
</section>
<section id="projected-timeline-and-milestones-for-aim-3" class="level1" data-number="18">
<h1 data-number="18"><span class="header-section-number">18</span> Projected Timeline and Milestones for Aim 3</h1>
<p>While timing may vary, a realistic timeline for this aim, spans two academic years:</p>
<section id="year-1-focus-on-phase-1---cnn-model-development-and-site-inventory-creation" class="level2" data-number="18.1">
<h2 data-number="18.1" class="anchored" data-anchor-id="year-1-focus-on-phase-1---cnn-model-development-and-site-inventory-creation"><span class="header-section-number">18.1</span> Year 1: Focus on Phase 1 - CNN Model Development and Site Inventory Creation</h2>
<ul>
<li><strong>Months 1–4: Data Acquisition and Preparation.</strong> This period will be dedicated to acquiring high-resolution satellite imagery and the state-level UOG permit databases. The primary milestone will be the programmatic generation of tens of thousands of image snippets centered on permitted well locations.<br>
</li>
<li><strong>Months 4–8: Iterative Dataset Annotation</strong> This phase involves the multi-step process of creating the seed dataset, training the V1 model, and performing semi-automated labeling and review to build the full training dataset for the Pennsylvania case study. The IAA analysis will also be conducted.</li>
<li><strong>Months 8–12: “Expert” Model Training and Large-Scale Deployment.</strong> This final phase involves training the expert model on the full dataset, validating its performance, and deploying it over the tax-parcel-masked search area to generate the final, comprehensive UOG inventory.</li>
</ul>
</section>
<section id="year-2-spatiotemporal-analysis-health-data-acquisition-and-epidemiological-linkage" class="level2" data-number="18.2">
<h2 data-number="18.2" class="anchored" data-anchor-id="year-2-spatiotemporal-analysis-health-data-acquisition-and-epidemiological-linkage"><span class="header-section-number">18.2</span> Year 2: Spatiotemporal Analysis, Health Data Acquisition, and Epidemiological Linkage</h2>
<ul>
<li><p><strong>Months 1–9 (Parallel Track 1): Methane & Exposure Modeling</strong><br>
Acquiring and processing methane data and integrating it with the UOG inventory from Phase 1 to create the final, dynamic exposure dataset.</p></li>
<li><p><strong>Months 1–9 (Parallel Track 2): Health Data Acquisition</strong><br>
the formal application process for the restricted-access NCHS mortality data will be undertaken through the NYRDC. This includes proposal writing, IRB submission, and DUA processing, which can take months.</p></li>
<li><p><strong>Months 9–12: Final Integration and Analysis.</strong></p>
<ul>
<li>Linking the final exposure dataset with the approved NCHS health microdata.<br>
</li>
<li>Executing the hybrid analytical plan: running the predictive GNN/CNN models and the inferential Negative Binomial models.<br>
</li>
<li>Generating final results, including predictive maps and risk ratios.<br>
</li>
<li>Completing the initial draft of the chapter’s findings.</li>
</ul></li>
</ul>
<p>The final output of this aim will culminate in a comprehensive environmental health assessment. This includes both the creation of a novel, high-resolution dataset linking UOG infrastructure to methane emissions and its application in a hybrid analytical framework. By integrating deep learning for predictive risk mapping with rigorous statistical modeling for inferential risk quantification, this research will provide a multi-faceted understanding of the potential public health consequences of UOG development, offering a valuable new resource for researchers, policymakers, and communities.</p>
<hr>
<p>As the methodology is developed, this chapter will be updated.</p>
<!-- -->
</section>
</section>
</main> <!-- /main -->
<script id="quarto-html-after-body" type="application/javascript">
window.document.addEventListener("DOMContentLoaded", function (event) {
const toggleBodyColorMode = (bsSheetEl) => {
const mode = bsSheetEl.getAttribute("data-mode");
const bodyEl = window.document.querySelector("body");
if (mode === "dark") {
bodyEl.classList.add("quarto-dark");
bodyEl.classList.remove("quarto-light");
} else {
bodyEl.classList.add("quarto-light");
bodyEl.classList.remove("quarto-dark");
}
}
const toggleBodyColorPrimary = () => {
const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
if (bsSheetEl) {
toggleBodyColorMode(bsSheetEl);
}
}
toggleBodyColorPrimary();
const icon = "";
const anchorJS = new window.AnchorJS();
anchorJS.options = {
placement: 'right',
icon: icon
};
anchorJS.add('.anchored');
const isCodeAnnotation = (el) => {
for (const clz of el.classList) {
if (clz.startsWith('code-annotation-')) {
return true;
}
}
return false;
}
const onCopySuccess = function(e) {
// button target
const button = e.trigger;
// don't keep focus
button.blur();
// flash "checked"
button.classList.add('code-copy-button-checked');
var currentTitle = button.getAttribute("title");
button.setAttribute("title", "Copied!");
let tooltip;
if (window.bootstrap) {
button.setAttribute("data-bs-toggle", "tooltip");
button.setAttribute("data-bs-placement", "left");
button.setAttribute("data-bs-title", "Copied!");
tooltip = new bootstrap.Tooltip(button,
{ trigger: "manual",
customClass: "code-copy-button-tooltip",
offset: [0, -8]});
tooltip.show();
}
setTimeout(function() {
if (tooltip) {
tooltip.hide();
button.removeAttribute("data-bs-title");
button.removeAttribute("data-bs-toggle");
button.removeAttribute("data-bs-placement");
}
button.setAttribute("title", currentTitle);
button.classList.remove('code-copy-button-checked');
}, 1000);
// clear code selection
e.clearSelection();
}
const getTextToCopy = function(trigger) {
const codeEl = trigger.previousElementSibling.cloneNode(true);
for (const childEl of codeEl.children) {
if (isCodeAnnotation(childEl)) {
childEl.remove();
}
}
return codeEl.innerText;
}
const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
text: getTextToCopy
});
clipboard.on('success', onCopySuccess);
if (window.document.getElementById('quarto-embedded-source-code-modal')) {
// For code content inside modals, clipBoardJS needs to be initialized with a container option
// TODO: Check when it could be a function (https://github.com/zenorocha/clipboard.js/issues/860)
const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
text: getTextToCopy,
container: window.document.getElementById('quarto-embedded-source-code-modal')
});
clipboardModal.on('success', onCopySuccess);
}
const viewSource = window.document.getElementById('quarto-view-source') ||
window.document.getElementById('quarto-code-tools-source');
if (viewSource) {
const sourceUrl = viewSource.getAttribute("data-quarto-source-url");
viewSource.addEventListener("click", function(e) {
if (sourceUrl) {
// rstudio viewer pane
if (/\bcapabilities=\b/.test(window.location)) {
window.open(sourceUrl);
} else {
window.location.href = sourceUrl;
}
} else {
const modal = new bootstrap.Modal(document.getElementById('quarto-embedded-source-code-modal'));
modal.show();
}
return false;
});
}
function toggleCodeHandler(show) {
return function(e) {
const detailsSrc = window.document.querySelectorAll(".cell > details > .sourceCode");
for (let i=0; i<detailsSrc.length; i++) {
const details = detailsSrc[i].parentElement;
if (show) {
details.open = true;
} else {
details.removeAttribute("open");
}
}
const cellCodeDivs = window.document.querySelectorAll(".cell > .sourceCode");
const fromCls = show ? "hidden" : "unhidden";
const toCls = show ? "unhidden" : "hidden";
for (let i=0; i<cellCodeDivs.length; i++) {
const codeDiv = cellCodeDivs[i];
if (codeDiv.classList.contains(fromCls)) {
codeDiv.classList.remove(fromCls);
codeDiv.classList.add(toCls);
}
}
return false;
}
}
const hideAllCode = window.document.getElementById("quarto-hide-all-code");
if (hideAllCode) {
hideAllCode.addEventListener("click", toggleCodeHandler(false));
}
const showAllCode = window.document.getElementById("quarto-show-all-code");
if (showAllCode) {
showAllCode.addEventListener("click", toggleCodeHandler(true));
}
var localhostRegex = new RegExp(/^(?:http|https):\/\/localhost\:?[0-9]*\//);
var mailtoRegex = new RegExp(/^mailto:/);
var filterRegex = new RegExp('/' + window.location.host + '/');
var isInternal = (href) => {
return filterRegex.test(href) || localhostRegex.test(href) || mailtoRegex.test(href);
}
// Inspect non-navigation links and adorn them if external
var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
for (var i=0; i<links.length; i++) {
const link = links[i];
if (!isInternal(link.href)) {
// undo the damage that might have been done by quarto-nav.js in the case of
// links that we want to consider external
if (link.dataset.originalHref !== undefined) {
link.href = link.dataset.originalHref;
}
}
}
function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
const config = {
allowHTML: true,
maxWidth: 500,
delay: 100,
arrow: false,
appendTo: function(el) {
return el.parentElement;
},
interactive: true,
interactiveBorder: 10,
theme: 'quarto',
placement: 'bottom-start',
};
if (contentFn) {
config.content = contentFn;
}
if (onTriggerFn) {
config.onTrigger = onTriggerFn;
}
if (onUntriggerFn) {
config.onUntrigger = onUntriggerFn;
}
window.tippy(el, config);
}
const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
for (var i=0; i<noterefs.length; i++) {
const ref = noterefs[i];
tippyHover(ref, function() {
// use id or data attribute instead here
let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
try { href = new URL(href).hash; } catch {}
const id = href.replace(/^#\/?/, "");
const note = window.document.getElementById(id);
if (note) {
return note.innerHTML;
} else {
return "";
}
});
}
const xrefs = window.document.querySelectorAll('a.quarto-xref');
const processXRef = (id, note) => {
// Strip column container classes
const stripColumnClz = (el) => {
el.classList.remove("page-full", "page-columns");
if (el.children) {
for (const child of el.children) {
stripColumnClz(child);
}
}
}
stripColumnClz(note)
if (id === null || id.startsWith('sec-')) {
// Special case sections, only their first couple elements
const container = document.createElement("div");
if (note.children && note.children.length > 2) {
container.appendChild(note.children[0].cloneNode(true));
for (let i = 1; i < note.children.length; i++) {
const child = note.children[i];
if (child.tagName === "P" && child.innerText === "") {
continue;
} else {
container.appendChild(child.cloneNode(true));
break;
}
}
if (window.Quarto?.typesetMath) {
window.Quarto.typesetMath(container);
}
return container.innerHTML
} else {
if (window.Quarto?.typesetMath) {
window.Quarto.typesetMath(note);
}
return note.innerHTML;
}
} else {
// Remove any anchor links if they are present
const anchorLink = note.querySelector('a.anchorjs-link');
if (anchorLink) {
anchorLink.remove();
}
if (window.Quarto?.typesetMath) {
window.Quarto.typesetMath(note);
}
// TODO in 1.5, we should make sure this works without a callout special case
if (note.classList.contains("callout")) {
return note.outerHTML;
} else {
return note.innerHTML;
}
}
}
for (var i=0; i<xrefs.length; i++) {
const xref = xrefs[i];
tippyHover(xref, undefined, function(instance) {
instance.disable();
let url = xref.getAttribute('href');
let hash = undefined;
if (url.startsWith('#')) {
hash = url;
} else {
try { hash = new URL(url).hash; } catch {}
}
if (hash) {
const id = hash.replace(/^#\/?/, "");
const note = window.document.getElementById(id);
if (note !== null) {
try {
const html = processXRef(id, note.cloneNode(true));
instance.setContent(html);
} finally {
instance.enable();
instance.show();
}
} else {
// See if we can fetch this
fetch(url.split('#')[0])
.then(res => res.text())
.then(html => {
const parser = new DOMParser();
const htmlDoc = parser.parseFromString(html, "text/html");
const note = htmlDoc.getElementById(id);
if (note !== null) {
const html = processXRef(id, note);
instance.setContent(html);
}
}).finally(() => {
instance.enable();
instance.show();
});
}
} else {
// See if we can fetch a full url (with no hash to target)
// This is a special case and we should probably do some content thinning / targeting
fetch(url)
.then(res => res.text())
.then(html => {
const parser = new DOMParser();
const htmlDoc = parser.parseFromString(html, "text/html");
const note = htmlDoc.querySelector('main.content');
if (note !== null) {
// This should only happen for chapter cross references
// (since there is no id in the URL)
// remove the first header
if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
note.children[0].remove();
}
const html = processXRef(null, note);
instance.setContent(html);
}
}).finally(() => {
instance.enable();
instance.show();
});
}
}, function(instance) {
});
}
let selectedAnnoteEl;
const selectorForAnnotation = ( cell, annotation) => {
let cellAttr = 'data-code-cell="' + cell + '"';
let lineAttr = 'data-code-annotation="' + annotation + '"';
const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
return selector;
}
const selectCodeLines = (annoteEl) => {
const doc = window.document;
const targetCell = annoteEl.getAttribute("data-target-cell");
const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
const lines = annoteSpan.getAttribute("data-code-lines").split(",");
const lineIds = lines.map((line) => {
return targetCell + "-" + line;
})
let top = null;
let height = null;