wuff-mann commited on
Commit
1cd194f
·
verified ·
1 Parent(s): d4ae980

Update ASR reproducibility dataset card

Browse files
Files changed (1) hide show
  1. README.md +50 -0
README.md ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ pretty_name: ASR CYGNSS-SMAP H-GMM reproducibility data
4
+ task_categories:
5
+ - tabular-regression
6
+ tags:
7
+ - cygnss
8
+ - smap
9
+ - gnss-r
10
+ - ocean-wind
11
+ - remote-sensing
12
+ - gaussian-mixture-model
13
+ ---
14
+
15
+ # ASR CYGNSS-SMAP H-GMM reproducibility repository
16
+
17
+ This public dataset repository stores **derived CYGNSS-SMAP collocations, model
18
+ checkpoints, evaluation statistics, and manuscript figures** for the manuscript
19
+ on year-adaptive CYGNSS sea-surface wind retrieval.
20
+
21
+ ## Upstream public data
22
+
23
+ - CYGNSS Level-2 Science Data Record v3.2 (`CYGNSS_L2_V3.2`), NASA PO.DAAC,
24
+ DOI: 10.5067/CYGNS-L2X32.
25
+ - JPL SMAP Level-2B CAP Sea Surface Salinity and extreme-wind product v5.0
26
+ (`SMAP_JPL_L2B_SSS_CAP_V5`), NASA PO.DAAC, DOI: 10.5067/SMP50-2TOCS.
27
+
28
+ The upstream raw HDF5/netCDF granules are **not mirrored** here. They are
29
+ retrieved temporarily in Google Colab, transformed into quality-controlled
30
+ collocation shards, and deleted after the derived shard has been uploaded.
31
+
32
+ ## Repository structure
33
+
34
+ - `data/collocations/`: monthly Parquet shards for 2021-2025.
35
+ - `training/`: balanced historical training sample.
36
+ - `models/`: hard-gated H-GMM and chronological calibrators.
37
+ - `predictions/`: 2025 raw and strict out-of-time test predictions.
38
+ - `metrics/`: JSON/CSV statistical outputs and bootstrap intervals.
39
+ - `figures/`: vector PDF and 600-dpi PNG manuscript figures.
40
+ - `manifests/`: provenance, checksums, split definitions, and processing logs.
41
+ - `code/`: exact Colab notebook and support package.
42
+
43
+ ## Important limitations
44
+
45
+ SMAP is treated as an external reference product rather than error-free ground
46
+ truth. The strict primary protocol uses the earliest 5% of 2025 observations as
47
+ chronological anchors and evaluates all later observations without anchor/test
48
+ overlap.
49
+
50
+ Repository: `wuff-mann/ASR-CYGNSS-HGMM-Reproducibility`