{"slug":"change-detection","title":"change-detection","summary":"Change analysis, once the observations are comparable. Not for cases whose blocker is comparability itself: mixed sensors, product levels or processing baselines to remote-sensing-analysis, undocumented vertical datums to point-cloud-lidar, multi-decade archive trends over large ","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-09-18T14:08:01.884949Z","repo":{"url":"https://github.com/muend/geoai-skills","stars":43,"forks":5,"license":"MIT","updatedAt":"2026-10-17T04:13:45Z"},"bodyHtml":"<hr>\n<h2>name: change-detection\ndescription: &gt;-\nChange analysis, once the observations are comparable. Not for cases whose\nblocker is comparability itself: mixed sensors, product levels or processing\nbaselines to remote-sensing-analysis, undocumented vertical datums to\npoint-cloud-lidar, multi-decade archive trends over large areas to\ngoogle-earth-engine. Matching product level does not prove comparability.\nOtherwise invoke for what, where or how much changed: two-scene comparison,\ndeforestation, urban growth, disaster damage, parcel-change audits,\nbi-temporal differencing, post-classification comparison, adjusted area,\nbreak detection in a series in hand (BFAST/LandTrendr/CCDC). Seasonal\nmismatch is this skill's own confounder; a documented datum with a stated\naccuracy budget is settled comparability. Keep both.\nlicense: MIT\nmetadata:\nauthor: Muhammed Enes Duran</h2>\n<h1>Change Detection &amp; Spatio-temporal Analysis</h1>\n<p>Purpose: separate real surface change from the four great impostors —\nmisregistration, radiometric drift, phenology, and classification error.\nEvery method below exists to control one of them; skipping the controls\nproduces confident maps of nothing.</p>\n<h2>Preconditions (where change detection is won or lost)</h2>\n<p>Preconditions 1 and 2 are <em>checked</em> here but <em>established</em> elsewhere. When\neither fails, the owning skill leads and this skill resumes once comparable\nobservations exist. Precondition 3 is this skill's own problem and is never a\nreason to route away.</p>\n<ol>\n<li><strong>Co-registration</strong>: sub-pixel alignment between dates (AROSICS or\nmanual tie-points). Half a pixel of shift creates edge-shaped phantom\nchange everywhere. Verify: flicker-compare crisp features. For elevation\nsurfaces or point clouds, vertical datum agreement, co-registration and\nthe vertical-accuracy budget belong to <code>point-cloud-lidar</code> — a datum\noffset is not subsidence.</li>\n<li><strong>Radiometric consistency</strong>: same processing level (surface\nreflectance), same sensor, same processing baseline. If any of the three\ndiffer, this is a harmonization problem, not a thresholding one: hand it\nto <code>remote-sensing-analysis</code> (HLS for Landsat↔Sentinel-2, relative\nnormalization with PIFs, <code>BOA_ADD_OFFSET</code> across the Sentinel-2 2022\nbaseline change).</li>\n<li><strong>Same season / phenological stage</strong> for bi-temporal work — a May vs\nSeptember pair \"detects\" summer. If season can't be matched, use\ncomposites or time-series methods instead.</li>\n<li><strong>Cloud/shadow masks intersected across dates</strong>; analyze only mutually\nvalid pixels and report that coverage %.</li>\n</ol>\n<h2>Method selection</h2>\n<table>\n<thead>\n<tr>\n<th>Situation</th>\n<th>Method</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Two dates, continuous \"how much\"</td>\n<td>Index differencing (ΔNDVI, ΔNBR...) with statistical thresholding</td>\n</tr>\n<tr>\n<td>Two dates, categorical \"from-what-to-what\"</td>\n<td>Post-classification comparison (only with strong classifiers)</td>\n</tr>\n<tr>\n<td>Two dates, multivariate robust</td>\n<td>Change vector analysis (CVA); MAD/iMAD for sensor-robust detection</td>\n</tr>\n<tr>\n<td>Dense stack, gradual + abrupt</td>\n<td>Trend + break analysis (BFAST/LandTrendr/CCDC family; at archive scale → <code>google-earth-engine</code>)</td>\n</tr>\n<tr>\n<td>Structure change (buildings)</td>\n<td>DL bi-temporal segmentation (siamese U-Net) → <code>geo-deep-learning</code></td>\n</tr>\n<tr>\n<td>SAR pairs (clouds, disasters)</td>\n<td>Log-ratio of calibrated backscatter + speckle handling</td>\n</tr>\n<tr>\n<td>Vector vintages (parcels, buildings)</td>\n<td>Geometry+attribute diff with tolerance (below)</td>\n</tr>\n</tbody>\n</table>\n<h2>Thresholding — never eyeball it</h2>\n<p>Difference images need a defensible threshold: μ ± k·σ on the difference\nhistogram (report k), Otsu when bimodal, or supervised thresholds\ncalibrated on labeled change/no-change samples. Deliver the histogram with\nthe chosen cut marked. Sensitivity: report changed-area at k-0.5 and\nk+0.5; if the story flips, the detection is fragile — say so.</p>\n<h2>Post-classification comparison (PCC) — handle with care</h2>\n<p>PCC error compounds: two 90%-accurate maps yield ≤ ~81% change accuracy,\nand biased errors create systematic false transitions. Rules:</p>\n<ul>\n<li>Use ONE classifier trained on both dates' imagery (same legend, same\nfeatures) rather than two independent legacy maps.</li>\n<li>Build the full <strong>transition matrix</strong> (from-class → to-class areas), not\njust a change/no-change binary — impossible transitions (water→forest\nin 1 year) are your error detector.</li>\n<li>Apply a minimum mapping unit consistent across dates before differencing.</li>\n</ul>\n<h2>Time-series (dense stack) analysis</h2>\n<ul>\n<li>Build a gap-filled, cloud-masked index stack (xarray, time dimension).</li>\n<li>Decompose trend + seasonality + breaks; per-pixel linear trends need\nsignificance testing (Mann-Kendall + Sen's slope for monotonic trends —\nand FDR correction across millions of pixels, or your \"greening map\" is\nnoise).</li>\n<li>Label break DATES, not just presence — timing is usually the analytic\npayload (when did clearing start?).</li>\n<li>Validate detected breaks against known events (fires, construction\npermits, disaster dates) wherever records exist.</li>\n</ul>\n<h2>Vector change audit (two vintages of the same layer)</h2>\n<ul>\n<li>Match features by stable ID if it exists; else spatial matching with IoU\nthreshold (report it).</li>\n<li>Classify: added / removed / geometry-changed (area delta &gt; tolerance) /\nattribute-changed. Tolerances absorb digitization jitter — 1-2 m for\ncadastre-grade, more for digitized-from-imagery.</li>\n<li>Sum area deltas by class and reconcile totals; unexplained residual =\nmatching bugs.</li>\n</ul>\n<h2>Accuracy assessment (the deliverable's spine)</h2>\n<p>Change is rare, so random sampling wastes effort on stable pixels — use\n<strong>stratified sampling</strong> (strata: change/no-change or per-transition) with\ngood-practice area estimation (Olofsson et al. protocol): report\nuser's/producer's accuracy per stratum AND <strong>area estimates with\nconfidence intervals</strong> adjusted for map error. A raw pixel count of the\nchange map is a biased area estimate — always say the adjusted number.</p>\n<h2>Reporting template</h2>\n<pre><code>## Change: &lt;phenomenon&gt;, &lt;T1&gt; → &lt;T2 or period&gt;\n- Data: &lt;sensor/level&gt;, co-registration RMSE: &lt;px&gt;, valid overlap: &lt;%&gt;\n- Method: &lt;...&gt; threshold/params: &lt;...&gt; (sensitivity: &lt;stable/fragile&gt;)\n- Transitions: &lt;matrix or top-5 list with areas ± CI&gt;\n- Accuracy: stratified n=&lt;&gt;, UA/PA per class, adjusted areas ± CI\n- Impostor controls: season &lt;matched?&gt;, radiometry &lt;harmonized?&gt;\n</code></pre>\n<h2>Pitfalls checklist</h2>\n<ul>\n<li>Phantom edge-change from misregistration.</li>\n<li>Seasonal difference sold as land cover change.</li>\n<li>PCC with two independently produced legacy maps.</li>\n<li>Threshold chosen \"because it looked right\", no sensitivity.</li>\n<li>Raw changed-pixel counts reported as area (no error-adjusted estimate).</li>\n<li>Trend maps without multiple-testing control.</li>\n<li>SAR change on unfiltered linear-power images.</li>\n</ul>\n<h2>Execution contract</h2>\n<ul>\n<li><strong>Workflow:</strong> define the change question; harmonize extent, season, radiometry, resolution, and registration; select method; estimate change; validate; report uncertainty.</li>\n<li><strong>Decision rules:</strong> use direct differencing only for comparable continuous signals, post-classification comparison for stable class legends, and time-series methods when a dense temporal stack exists.</li>\n<li><strong>Verification protocol:</strong> quantify co-registration, valid overlap, threshold sensitivity, transition accounting, and accuracy-adjusted area with confidence intervals.</li>\n<li><strong>Failure modes:</strong> reject causal change claims when season, sensor, clouds, registration, or independent map errors can explain the signal.</li>\n<li><strong>Deliverables:</strong> change map, transition or trend table, parameter record, validation sample and metrics, adjusted-area estimate, and limitations.</li>\n<li><strong>Source freshness:</strong> consult <a href=\"references/authoritative-sources.md\">the authoritative source registry</a> before using version-sensitive products or APIs and record the checked date.</li>\n</ul>\n","files":[{"path":"agents/openai.yaml","sizeBytes":204,"isText":true},{"path":"references/authoritative-sources.md","sizeBytes":854,"isText":true},{"path":"SKILL.md","sizeBytes":7719,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-09-18T14:08:25.380329Z","sha256":"AC33A964CACF2A5A2437B961B7EBBBA569B8CEAB6656DEE6A0853C539605DCE8","sizeBytes":4670},"review":null,"source":{"repositoryUrl":"https://github.com/muend/geoai-skills","path":"skills/change-detection","license":"MIT","commit":"19937ce14c7d60d764732cc2688c6a3dd92f6d0e","subtreeSha":"E4C18A9DAE20791CC8823F1EEA092B2F2A07CF2D9BE9766C4847BC96855B1F56","lastSyncedAt":"2026-11-08T21:45:06.252907Z"},"reviewedAt":"2026-09-18T14:09:29.962248Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/muend/geoai-skills/tree/main/skills/change-detection"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install muend-geoai-skills@llmmart"},{"target":"git","command":"git clone https://github.com/muend/geoai-skills.git"}]}