Add Gamma SBAS production workflow

2026-05-21 04:53:39 +08:00
parent 9000feeee8
commit cc22b9ac2d
46 changed files with 11386 additions and 332 deletions
@@ -0,0 +1,551 @@
+# Gamma IPTA LT1 SBAS Trial Runbook
+
+Date: 2026-05-18
+
+## Goal
+
+Validate whether the current local LT1 data pool and WSL Gamma installation can produce one usable SBAS/IPTA trial result with the official Gamma toolchain.
+
+This is not yet a production integration design. The immediate goal is to run one conservative stack, inspect failures and product quality, then decide what should be productized in the system.
+
+Actual trial root used in this repository:
+
+```text
+D:\Code\Insar_management_system_v2\backend\runtime\gamma_ipta_trials\lt1b_r114_e1312_n438_20240516_20251002
+```
+
+## Current Environment
+
+- WSL distro configured by the project: `Ubuntu-24.04`.
+- Gamma install found in WSL: `/usr/local/GAMMA_SOFTWARE-20240627`.
+- Installed package name present on disk: `GAMMA_SOFTWARE-20240627_MSP_ISP_DIFF_IPTA.linux64_ubuntu2404.tar.gz`.
+- Installed Gamma modules include `MSP`, `ISP`, `DIFF`, `DISP`, and `IPTA`.
+- `GEO` is not a separate directory in this install, but DIFF contains the relevant geocoding tools, including `gc_map`, `geocode`, and `geocode_back`.
+- Project Gamma environment script: `deploy/wsl/profiles/gamma_env.sh`.
+- Project WSL Python: `/home/administrator/miniconda3/envs/insar_wsl_v1/bin/python`.
+
+Important command checks already performed:
+
+- `IPTA/bin/ts_rate` runs and prints usage. No license-denied error observed.
+- `IPTA/bin/multi_def_pt` runs and prints usage. No license-denied error observed.
+- `ISP/bin/par_LT1_SLC` exists and prints LT1 SLC conversion usage.
+- `ISP/scripts/LT1_precision_orbit.py` runs with the project conda Python and prints usage.
+
+## Data Pool Findings
+
+Configured LT1 source pool:
+
+- `D:\LuTan1_Image_Pool`
+
+Configured LT1 precise orbit pool:
+
+- `D:\orbit_pools\envi\LT1A`
+- `D:\orbit_pools\envi\LT1B`
+
+High-level inventory:
+
+- LT1 scene directories found: `1500`.
+- LT1A orbit TXT files found: `896`, spanning `20230508` to `20251219`.
+- LT1B orbit TXT files found: `858`, spanning `20230510` to `20251219`.
+
+Metadata caveat:
+
+- `*.meta.xml` and `*_Check.xml` are not safe to parse as whole XML documents in PowerShell because many files contain malformed Chinese text near the tail, for example bad `usePreciseOrbit` closing text.
+- The needed `productInfo` block is structurally valid. For stack discovery, parse only `<productInfo>...</productInfo>`.
+- Gamma `par_LT1_SLC` should still use the original `.meta.xml`; do not rewrite source metadata unless a Gamma run proves the malformed tail is a blocker.
+
+## Candidate Stack
+
+First trial should use a narrow single-center stack, not the broad system time-series grouping.
+
+Recommended trial stack:
+
+- Satellite: `LT1B`
+- Relative orbit: `114`
+- Direction: `DESCENDING`
+- Imaging mode: `STRIP1`
+- Polarization: `HH`
+- Approximate center: `E131.2 / N43.8`
+- Scene count at this center: `7`
+- Scenes with precise orbit TXT currently present: `5`
+
+Use the 5 scenes with available precise orbit first:
+
+| Date | Orbit TXT | Scene |
+| --- | --- | --- |
+| 20240516 | yes | `LT1B_MONO_SYC_STRIP1_012047_E131.2_N43.8_20240516_SLC_HH_S2A_0000399289` |
+| 20240711 | yes | `LT1B_MONO_SYC_STRIP1_012880_E131.2_N43.8_20240711_SLC_HH_S2A_0000450956` |
+| 20240905 | yes | `LT1B_MONO_SYC_STRIP1_013713_E131.2_N43.8_20240905_SLC_HH_S2A_0000501650` |
+| 20250417 | yes | `LT1B_MONO_SYC_STRIP1_017045_E131.2_N43.8_20250417_SLC_HH_S2A_0000713375` |
+| 20251002 | yes | `LT1B_MONO_SYC_STRIP1_019544_E131.2_N43.8_20251002_SLC_HH_S2A_0000891257` |
+
+Do not include these two in the first run unless the missing orbits are added:
+
+| Date | Orbit TXT | Scene |
+| --- | --- | --- |
+| 20250612 | no | `LT1B_MONO_SYC_STRIP1_017878_E131.2_N43.8_20250612_SLC_HH_S2A_0000772122` |
+| 20250807 | no | `LT1B_MONO_SYC_STRIP1_018711_E131.2_N43.8_20250807_SLC_HH_S2A_0000831367` |
+
+Common bounding box across the broader 13-scene `LT1B relOrbit 114 / E131-N44` candidate:
+
+- lon: `130.8615 .. 131.1638`
+- lat: `43.7127 .. 44.0987`
+
+For the narrow `E131.2/N43.8` trial stack, overlap is visually/metadata-wise much tighter:
+
+- Each scene center is around `131.20E, 43.79N`.
+- Each scene bbox is roughly `130.81..131.62E`, `43.48..44.10N`.
+
+Secondary candidate if the first stack fails:
+
+- `LT1B relOrbit 114 DESCENDING STRIP1 HH`, center `E130.8/N43.9`.
+- 5 dates, 4 with orbit: `20250425`, `20250620`, `20250815`, `20251010`.
+- One scene is `MONO_MH1` while the others are `MONO_SYC`; keep it as secondary, not first choice.
+
+## Current System Pairing Limitations
+
+The current time-series stack selection is useful for broad discovery but is too coarse for Gamma IPTA production.
+
+Observed code behavior:
+
+- LT1A/LT1B are normalized into the same satellite family `LT1`.
+- The compatibility key only uses direction, satellite family, imaging mode, and polarization.
+- Relative orbit, absolute track family, scene center/strip identity, receiving station, and detailed LT1 product variant are not hard grouping keys.
+- The stable-stack selector tries to recover by common AOI overlap and pairwise network connectivity.
+- The SBAS network selector uses time-baseline, center-distance and overlap thresholds, but does not use Gamma-derived perpendicular baseline at planning time.
+- Time-series processors currently accepted by the service are only `isce2_stack_mintpy` and `sarscape_sbas`; there is no `gamma_ipta_sbas` processor code yet.
+
+For the Gamma IPTA trial, do not rely on the current automatic PS stack plan as the source of truth. Use a manually audited stack manifest first.
+
+## Trial Run Checklist
+
+### 1. Create Isolated Work Directory
+
+Actual root:
+
+```text
+D:\Code\Insar_management_system_v2\backend\runtime\gamma_ipta_trials\lt1b_r114_e1312_n438_20240516_20251002
+```
+
+Keep these subdirectories:
+
+```text
+input\scenes
+input\orbits
+gamma\slc
+gamma\mli
+gamma\diff
+gamma\ipta
+logs
+publish
+```
+
+For the first trial, prefer symlinks or a manifest that points to source scenes. Avoid duplicating large TIFF files unless Gamma scripts require local flat layout.
+
+### 2. Build Scene Manifest
+
+For each selected scene, record:
+
+- scene directory
+- `.tiff`
+- `.meta.xml`
+- precise orbit TXT
+- date
+- relative orbit
+- direction
+- mode
+- polarization
+- center lon/lat
+- bbox
+
+This manifest becomes the hand-audited truth for the trial.
+
+### 3. Convert LT1 Products To Gamma SLC
+
+For each selected scene:
+
+```bash
+source /mnt/d/Code/Insar_management_system_v2/deploy/wsl/profiles/gamma_env.sh
+par_LT1_SLC <scene.tiff> <scene.meta.xml> <yyyymmdd>.slc.par <yyyymmdd>.slc
+```
+
+Then apply precise orbit:
+
+```bash
+/home/administrator/miniconda3/envs/insar_wsl_v1/bin/python \
+  /usr/local/GAMMA_SOFTWARE-20240627/ISP/scripts/LT1_precision_orbit.py \
+  <yyyymmdd>.slc.par <orbit_txt_or_orbit_folder>
+```
+
+### 4. Build SLC/MLI Tables
+
+Create Gamma tables for the stack:
+
+```text
+SLC_tab
+RMLI_tab
+```
+
+Use multilook settings conservative enough for a first run. The goal is robustness and fast feedback, not final product resolution.
+
+### 5. Baseline And Pair Network
+
+Use Gamma baseline tools first, not the system's center-distance approximation:
+
+- `base_init`
+- `base_perp`
+- IPTA baseline tools such as `base_orbit_pt`, `base_par_pt`, `base_ls_pt` as needed by the official IPTA path.
+
+For the first 5-scene stack, start with a simple connected small-baseline network:
+
+- adjacent pairs by time
+- add one or two skip pairs only if coherence and baseline look acceptable
+
+Expected adjacent temporal intervals:
+
+- `20240516 -> 20240711`: 56 days
+- `20240711 -> 20240905`: 56 days
+- `20240905 -> 20250417`: 224 days
+- `20250417 -> 20251002`: 168 days
+
+The large seasonal gaps are acceptable for a trial only if perpendicular baseline and coherence are reasonable. If they are poor, switch to a denser 2025-only or 2024-only local test, even with fewer dates.
+
+### 6. Differential Interferograms
+
+Use official Gamma DIFF commands/scripts for:
+
+- coregistration
+- interferogram generation
+- simulated topographic phase
+- differential phase
+- filtering
+- coherence
+- unwrapping if needed by the chosen IPTA path
+
+Do not implement custom SBAS inversion in the management system.
+
+### 7. IPTA Processing
+
+Use Gamma IPTA commands for point/stack time-series processing. Confirm exact command sequence against the installed:
+
+```text
+/usr/local/GAMMA_SOFTWARE-20240627/IPTA/html/IPTA_users_guide.pdf
+```
+
+Commands observed in the local IPTA module include:
+
+- `multi_def_pt`
+- `ts_rate`
+- `ts_rate_pt`
+- `base_ls_pt`
+- `base_par_pt`
+- `ph_base_pt`
+- `atm_mod_pt`
+- `pt2geo`
+- `dis_ipta`
+
+### 8. Review Outputs
+
+Minimum acceptance checks for the first run:
+
+- every selected scene converts to SLC
+- precise orbit update succeeds for every SLC
+- all intended pairs generate interferograms
+- coherence is not uniformly poor
+- unwrapping or IPTA point solution is not globally unstable
+- one geocoded velocity or displacement-rate raster/vector product can be inspected
+- logs and command manifests are complete enough to reproduce the run
+
+## Trial Progress On 2026-05-18
+
+### Files Created For This Trial
+
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/input/scene_manifest.json`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/01_prepare_slc.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/02_mli_and_baseline.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/03_coreg_one_pair.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/04_cc_stats.py`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/05_coreg_adjacent_pairs.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/06_prepare_rdc_dem.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/07_coreg_common_ref_stack.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/08_diff_unwrap_common_ref.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/09_mb_ts_rate.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/10_float_stats.py`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/11_make_timeseries_previews.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/12_geocode_export_timeseries.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/13_phase_to_los.py`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/14_build_trial_summary.py`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/15_make_los_velocity_maps.sh`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/16_plot_monitor_point_timeseries.py`
+- `backend/runtime/gamma_ipta_trials/lt1b_r114_e1312_n438_20240516_20251002/scripts/17_make_geocoded_web_previews.sh`
+
+### Completed
+
+- Created a hand-audited 5-scene LT1B manifest for `relOrbit 114 / DESCENDING / STRIP1 / HH / E131.2 N43.8`.
+- Converted all 5 LT1 TIFF products with Gamma `par_LT1_SLC`.
+- Applied precise orbit updates with Gamma `LT1_precision_orbit.py`.
+- Built `gamma/slc/SLC_tab`.
+- Built 8x8 multilooked MLI products with Gamma `multi_look`.
+- Built `gamma/mli/RMLI_tab`.
+- Generated all-pair and adjacent-pair baseline tables with Gamma `base_calc`.
+- Ran adjacent-pair coregistration for all 4 selected adjacent pairs using Gamma `SLC_coreg.py`.
+- Generated 4 interferograms and coherence products using Gamma `create_offset`, `SLC_intf`, and `cc_wave`.
+- Generated browse previews using Gamma `rasmph` and `raspwr`.
+- Computed coherence statistics for each adjacent pair.
+- Reused the existing Copernicus30 DEM cache covering `E131.2/N43.8`.
+- Generated a reference-geometry RDC DEM for `20240905` using Gamma `gc_map1`, `geocode`, and `gc_map_fine`.
+- Built a common-reference RSLC/RMLI stack in `20240905` geometry.
+- Generated common-reference differential interferograms with Gamma `phase_sim_orb` and `SLC_diff_intf`.
+- Filtered common-reference differential interferograms with Gamma `adf`.
+- Unwrapped common-reference differential interferograms with Gamma `mcf`.
+- Ran Gamma IPTA `mb` to solve the image-based phase time-series.
+- Ran Gamma IPTA `ts_rate` to estimate a linear phase-rate map.
+- Generated Gamma preview rasters for `ts_rate`, `sigma_rate`, and `hgt_correction`.
+- Exported phase-rate, sigma, height-correction, and LOS-rate rasters to EPSG:4326 GeoTIFF.
+- Generated explicit LOS velocity maps in `mm/year` for both away-from-radar-positive and toward-radar-positive conventions.
+- Generated one example monitoring-point LOS displacement curve as PNG, CSV, and metadata JSON.
+- Generated north-up WGS84 web preview PNGs from the geocoded LOS GeoTIFF products.
+
+### Key Outputs
+
+- SLC stack: `gamma/slc/SLC_tab`
+- MLI stack: `gamma/mli/RMLI_tab`
+- All-pair baseline table: `gamma/diff/bperp_all_pairs.txt`
+- Adjacent-pair table: `gamma/diff/itab_adjacent`
+- Pair-specific RSLC products: `gamma/rslc/*_to_*.rslc`
+- Pair-specific quality reports: `gamma/rslc/*_to_*.rslc.coreg_quality`
+- Adjacent interferograms: `gamma/int/*.int`
+- Adjacent coherence rasters: `gamma/int/*.cc`
+- Adjacent interferogram previews: `gamma/int/*.int.bmp`
+- Adjacent coherence previews: `gamma/int/*.cc.bmp`
+- Coherence statistics: `logs/*_cc_stats.txt`
+- Reference RDC DEM: `gamma/dem/20240905_8rlks.rdc.dem`
+- Common-reference stack tables: `gamma/common_20240905/SLC_tab`, `gamma/common_20240905/RMLI_tab`
+- Common-reference differential stack: `gamma/common_20240905/diff/*/*_8rlks.diff_filt.unw`
+- Gamma `mb` time-series list: `gamma/common_20240905/timeseries/diff_ts.tab`
+- Gamma `mb` phase time-series: `gamma/common_20240905/timeseries/diff_ts_*.diff`
+- Gamma `mb` residual sigma: `gamma/common_20240905/timeseries/sigma_ts`
+- Gamma `mb` height correction: `gamma/common_20240905/timeseries/hgt_correction`
+- Gamma `ts_rate` output: `gamma/common_20240905/timeseries/ts_rate`
+- Gamma `ts_rate` sigma output: `gamma/common_20240905/timeseries/sigma_rate`
+- Preview rasters: `gamma/common_20240905/timeseries/*.bmp`
+- Geocoded GeoTIFF exports: `publish/geotiff/*.tif`
+- Explicit LOS velocity previews: `publish/geotiff/los_rate_toward_mm_per_year.bmp`, `publish/geotiff/los_rate_away_mm_per_year.bmp`
+- Geocoded web previews: `publish/geotiff/los_rate_toward_mm_per_year.geo_preview.png`, `publish/geotiff/los_sigma_mm_per_year.geo_preview.png`
+- Monitoring-point curve: `publish/monitor_points/auto_low_sigma_high_rate_timeseries.png`
+- Monitoring-point values: `publish/monitor_points/auto_low_sigma_high_rate_timeseries.csv`
+- Trial summary: `publish/trial_summary.json`
+
+### Baseline Notes
+
+`base_calc` generated 10 all-pair entries. The adjacent temporal network is:
+
+| Pair | Delta days | Bperp from `base_calc` |
+| --- | ---: | ---: |
+| 20240516 -> 20240711 | 56 | 646.09670 |
+| 20240711 -> 20240905 | 56 | -236.76960 |
+| 20240905 -> 20250417 | 224 | 249.35880 |
+| 20250417 -> 20251002 | 168 | 59.18150 |
+
+The first validated pair was `20240711 -> 20240905` because it has a 56-day interval and moderate perpendicular baseline in this stack.
+
+### Adjacent-Pair Quality
+
+Final `SLC_coreg.py` quality-test summaries:
+
+| Pair | Accepted offsets | Final std range | Final std azimuth |
+| --- | ---: | ---: | ---: |
+| 20240516 -> 20240711 | 2476 / 2688 | 0.0861 | 0.1120 |
+| 20240711 -> 20240905 | 2483 / 2656 | 0.0370 | 0.1245 |
+| 20240905 -> 20250417 | 2322 / 2688 | 0.0487 | 0.0617 |
+| 20250417 -> 20251002 | 2274 / 2688 | 0.0304 | 0.0533 |
+
+Coherence statistics from Gamma big-endian float files:
+
+| Pair | valid `[0,1]` | p25 | median | p75 | p99 | Comment |
+| --- | ---: | ---: | ---: | ---: | ---: | --- |
+| 20240516 -> 20240711 | 100% | 0.127489 | 0.212285 | 0.323520 | 0.649398 | Weakest pair; high Bperp and low coherence |
+| 20240711 -> 20240905 | 100% | 0.629307 | 0.733220 | 0.802884 | 0.909973 | Strong pair |
+| 20240905 -> 20250417 | 100% | 0.339094 | 0.485239 | 0.612798 | 0.833928 | Usable trial pair |
+| 20250417 -> 20251002 | 100% | 0.394165 | 0.582921 | 0.734878 | 0.932300 | Usable trial pair |
+
+All adjacent pairs generated expected Gamma products. The full 5-scene chain can continue, but the first pair is a quality risk for unwrapping/IPTA. A more conservative follow-up is to run the 4-scene sub-stack from `20240711` to `20251002`, or keep the 5-scene stack but down-weight or exclude `20240516 -> 20240711` if later unwrapping/IPTA residuals are poor.
+
+Implementation note:
+
+- `SLC_coreg.py` writes the refined offset parameter file to a secondary-date path such as `gamma/slc/20240711.slc.off`.
+- Trial scripts copy that file to pair-specific paths such as `gamma/rslc/20240711_to_20240905.rslc.off`.
+- Future skip-pair or non-adjacent networks must use pair-specific offset files to avoid accidental reuse after the same secondary scene is coregistered to another reference.
+
+### Common-Reference Time-Series Trial
+
+The image-based Gamma IPTA path was also run using a common `20240905` reference geometry. This is closer to the official `mb -> ts_rate` time-series chain than the first adjacent-pair wrapped interferogram check.
+
+Common-reference inputs and products:
+
+- reference geometry: `20240905`
+- DEM source: existing Copernicus30 cache under `backend/runtime/pyint_dem_cache`
+- RDC DEM: `gamma/dem/20240905_8rlks.rdc.dem`
+- common-reference SLC table: `gamma/common_20240905/SLC_tab`
+- common-reference MLI table: `gamma/common_20240905/RMLI_tab`
+- common-reference differential ITAB:
+
+```text
+1 3 1 1
+2 3 2 1
+3 4 3 1
+3 5 4 1
+```
+
+Filtered differential coherence statistics:
+
+| Pair | valid `[0,1]` | p25 | median | p75 | p99 |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| 20240516 -> 20240905 | 100% | 0.497369 | 0.884586 | 0.945308 | 0.980583 |
+| 20240711 -> 20240905 | 100% | 0.919149 | 0.968188 | 0.979449 | 0.990185 |
+| 20240905 -> 20250417 | 100% | 0.366279 | 0.790519 | 0.909991 | 0.978094 |
+| 20240905 -> 20251002 | 100% | 0.193104 | 0.815562 | 0.917209 | 0.972355 |
+
+Gamma `mb` outputs:
+
+- `gamma/common_20240905/timeseries/diff_ts_001.diff` through `diff_ts_005.diff`
+- `gamma/common_20240905/timeseries/diff_ts.tab`
+- `gamma/common_20240905/timeseries/itab_ts`
+- `gamma/common_20240905/timeseries/sigma_ts`
+- `gamma/common_20240905/timeseries/hgt_correction`
+
+Gamma `ts_rate` outputs:
+
+- `gamma/common_20240905/timeseries/ts_rate`
+- `gamma/common_20240905/timeseries/ts_const`
+- `gamma/common_20240905/timeseries/sigma_rate`
+- `gamma/common_20240905/timeseries/ts_rate.bmp`
+- `gamma/common_20240905/timeseries/sigma_rate.bmp`
+- `gamma/common_20240905/timeseries/hgt_correction.bmp`
+
+GeoTIFF exports:
+
+- `publish/geotiff/ts_rate_rad_per_year.tif`
+- `publish/geotiff/sigma_rate_rad_per_year.tif`
+- `publish/geotiff/sigma_ts_rad.tif`
+- `publish/geotiff/hgt_correction_m.tif`
+- `publish/geotiff/los_rate_m_per_year.tif`
+- `publish/geotiff/los_sigma_m_per_year.tif`
+- `publish/geotiff/los_rate_away_mm_per_year.tif`
+- `publish/geotiff/los_rate_toward_mm_per_year.tif`
+- `publish/geotiff/los_sigma_mm_per_year.tif`
+- `publish/geotiff/los_rate_toward_mm_per_year.geo_preview.png`
+- `publish/geotiff/los_sigma_mm_per_year.geo_preview.png`
+- `publish/trial_summary.json`
+
+Float output statistics using Gamma big-endian float:
+
+| File | Non-zero pixels | p25 | median | p75 | p99 |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| `ts_rate` | 8,868,956 | -1.663257 | -0.271621 | 0.869116 | 2.787572 |
+| `sigma_rate` | 8,868,956 | 0.396037 | 0.677578 | 1.010539 | 2.298520 |
+| `sigma_ts` | 8,892,237 | 0.000704 | 0.001557 | 0.002535 | 0.518188 |
+| `hgt_correction` | 8,892,250 | -16.989384 | 7.596325 | 36.411520 | 98.636933 |
+
+Explicit LOS velocity output statistics in `mm/year`:
+
+| File | Non-zero pixels | p01 | p25 | median | p75 | p99 |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| `los_rate_toward_mm_per_year.rdc` | 8,868,956 | -52.779640 | -16.455760 | 5.142855 | 31.491957 | 84.590944 |
+| `los_sigma_mm_per_year.rdc` | 8,868,956 | 0.530715 | 7.498536 | 12.829198 | 19.133450 | 43.519972 |
+
+This is a successful first local Gamma official-chain time-series trial. It is still a technical validation run, not a production-grade SBAS product: the stack has only 5 dates, the network is minimal, and reference region and unwrapping masks were conservative defaults.
+
+The exported GeoTIFFs are EPSG:4326, `2222 x 2237`, Float32, LZW-compressed Cloud Optimized GeoTIFFs with `NoData=0`.
+
+The `*.bmp` files generated by Gamma `rasdt_pwr` are RDC processing-geometry browse images. They are useful for quick processing QA but are not map products and should not be used as the default UI map preview. The UI/default web preview should use the `*.geo_preview.png` files generated from the EPSG:4326 GeoTIFF products.
+
+### LOS Sign Convention
+
+Gamma `ts_rate` is a phase-rate raster in `rad/year`. Converting phase rate to LOS displacement rate requires a sign convention:
+
+- `los_rate_away_mm_per_year = phase_rate * wavelength / (4*pi) * 1000`
+- `los_rate_toward_mm_per_year = -phase_rate * wavelength / (4*pi) * 1000`
+
+Gamma `dispmap` documents two conventions:
+
+- `sflg=0`, the default: motion away from radar is negative, so motion toward radar is positive; deformation and unwrapped phase have opposite signs.
+- `sflg=1`: motion away from radar is positive; deformation and unwrapped phase have the same sign.
+
+For system productization, use explicit names and prefer `los_rate_toward_mm_per_year` as the default display product because it matches Gamma `dispmap` default `sflg=0`. Keep the away-positive version available when another downstream convention requires direct phase-sign products.
+
+### Monitoring Point Curve
+
+An example monitoring point was selected automatically from low-sigma, high-rate, non-edge pixels:
+
+- radar pixel: range `336`, azimuth `2290`
+- approximate lon/lat: `131.4340324903`, `43.8008322757`
+- reference date in the plotted time series: `20240711`
+- LOS convention: toward radar positive, away from radar negative
+- fitted LOS velocity: `50.1085 mm/year`
+- fitted LOS velocity sigma: `0.0063 mm/year`
+
+Generated outputs:
+
+- `publish/monitor_points/auto_low_sigma_high_rate_timeseries.png`
+- `publish/monitor_points/auto_low_sigma_high_rate_timeseries.csv`
+- `publish/monitor_points/auto_low_sigma_high_rate_metadata.json`
+
+The plotted values are:
+
+| Date | LOS displacement, toward-positive mm |
+| --- | ---: |
+| 20240516 | -7.673957 |
+| 20240711 | 0.000000 |
+| 20240905 | 7.673922 |
+| 20250417 | 38.412669 |
+| 20251002 | 61.464305 |
+
+This is a single example point only. It is not a validated monitoring-point network and should not be interpreted as a representative area-wide deformation curve. The automatic selection favors a non-edge pixel with relatively high absolute velocity and low fitted sigma so the curve is visually inspectable. Production monitoring points need one of these inputs:
+
+- user-clicked map lon/lat
+- imported engineering monitoring-point layer
+- a configured regular grid or point-of-interest set
+- a quality-filtered automatic point sampler with spacing, coherence/sigma thresholds, and manual review
+
+Until that is implemented, the single curve is a capability demonstration and should be labeled as such in the UI.
+
+### Current Open Items
+
+- Review the `ts_rate.bmp`, `sigma_rate.bmp`, `hgt_correction.bmp`, LOS velocity BMPs, monitoring-point PNG, and exported GeoTIFFs visually in GIS.
+- Tune reference region, coherence thresholds, and pair network before treating the result as production.
+- Decide whether to keep the weak/long 2025 pair, add skip-pairs, or use a denser data sequence when more LT1 precise orbits are available.
+- Keep system integration as orchestration around Gamma commands; do not implement custom SBAS inversion in application code.
+
+## Productization Decisions After Trial
+
+If the 5-scene trial succeeds, add a new managed processor instead of bending existing ISCE/SARscape flows:
+
+```text
+processor_code = gamma_ipta_sbas
+engine_code = gamma
+workflow = gamma_ipta_sbas
+```
+
+Required system changes:
+
+- Add a Gamma IPTA stack manifest builder.
+- Add LT1-specific hard grouping keys:
+  - satellite platform, not only family, unless cross-satellite LT1A/LT1B is explicitly validated
+  - relative orbit
+  - orbit direction
+  - imaging mode
+  - polarization
+  - scene strip/center bucket
+  - product variant or station/submode where it affects compatibility
+- Add a Gamma baseline audit step before final pair network selection.
+- Persist selected Gamma pair network separately from the coarse planning graph.
+- Keep system code as orchestration only; Gamma remains the processing authority.
+
+## Current Recommendation
+
+Use the manual audited `LT1B relOrbit 114 DESCENDING STRIP1 HH / E131.2 N43.8` stack and the trial scripts as the reference path for productization.
+
+Do not start Gamma IPTA production from the current automatic time-series plan. It can be used for discovery, but production stack selection needs the hard grouping keys and Gamma baseline audit described above.
+
+For the next engineering step, add a managed `gamma_ipta_sbas` processor that orchestrates the Gamma commands rather than reimplementing SBAS inversion in application code.