Add small_commune_v3 (Var + Bouches-du-Rhône)

README.md

@@ -18,6 +18,8 @@ tags:
 
 # firepredict: daily wildfire-start risk for the communes of the Var (France)
 
+> **Two versions in this repository.** The files at the root are `small_commune_v2` (the Var only, 5 inputs), described below. The folder [`v3/`](v3/) holds `small_commune_v3`, one model shared by the Var (83) and the Bouches-du-Rhône (13); see the section "Version 3" below. Both are frozen and both have a 2026 test.
+
 A small logistic regression that gives each of the 153 communes of the Var (83), France, a daily chance that a fire of 0.1 ha or more starts there. It is built only from open data. Code, data pipeline and full report: [github.com/guillaumegay13/firepredict](https://github.com/guillaumegay13/firepredict).
 
 > **Research preview.** This is not an official warning tool. Fire danger warnings in France are issued by Météo-France and the prefectures. The every-year results below are exploratory (see [Limitations](#limitations)); the clean test is the 2026 one.
@@ -141,6 +143,26 @@ The inputs were chosen after looking at results across all years, so the scores
 - **Retrospective 2026 test, blind to the outcomes.** On 9 November 2026, before any 2026 fire record was published, the model was frozen and its forecasts for 1 January – 9 November 2026 were saved in `forecasts_2026.csv` (SHA-256 `2be6091ab25539dc47bb751822d2a9b2b48f21218f8238bf0be1e1653dc8c6d6`). They were all computed on that day, from the weather available then. This is a retrospective test, not a live trial. When BDIFF publishes the 2026 fires, the forecasts will be scored as they are.
 - **Live archive.** From 9 November 2026, each day's forecast from the frozen model is saved in the [`forecasts/`](https://github.com/guillaumegay13/firepredict/tree/main/forecasts) folder of the GitHub repository and committed the same day, so the git history dates it.
 
+
+## Version 3 (folder `v3/`)
+
+`small_commune_v3`, frozen on 9 November 2026, is a single logistic regression trained on the Var and the Bouches-du-Rhône together (2,140 positive commune-days). It uses the five inputs above plus:
+
+- `rate_x_summer`: log usual rate × June–September. Busy communes concentrate their fires in summer more than quiet ones; without this term the model underestimated the Var's busiest communes in summer by half and overestimated Aix-en-Provence in winter.
+- `dept_13`: one level per department (1 for the Bouches-du-Rhône), so the chances are calibrated in each department. In a transfer test, rankings carried over between departments but probabilities did not.
+
+`v3/usual_rates.csv` has a `department` column; `v3/metadata.json` lists the inputs, weights and fill values.
+
+Every-year test, each year hidden in both departments at once (positive commune-days):
+
+| | Var (83) | Bouches-du-Rhône (13) |
+|---|---|---|
+| ROC AUC, v3 | 0.779 (v2: 0.782) | 0.864 (single-department model: 0.865) |
+| FWI × commune history | 0.757; v3 better in 17 / 19 years | 0.858; v3 better in 13 / 20 years |
+| 15 communes a day: caught | 219 (reference 220) | 630 (reference 605) |
+
+Calibration holds in both departments up to 1%; above 1% in the Var the model predicts 1.2% where 0.8% happened. The same limits apply as for v2 (exploratory results, approximate FWI, two neighbouring departments only). The 2026 test of v3 covers both departments: `v3/forecasts_2026.csv`, SHA-256 `783debd1cd6848fb0cd667ef2abc418a2221379c9698366e1e6ead8eae0a856c`. Daily v3 forecasts are archived in the GitHub repository under `forecasts/v3/`.
+
 ## Limitations
 
 - **Exploratory results.** The inputs and the 30-day window were chosen after looking at the same years used for evaluation, so p-values and ranges are optimistic. The ranges are conditional on the chosen model.

v3/metadata.json

@@ -0,0 +1,94 @@
+{
+  "version": "small_commune_v3",
+  "frozen_at": "2026-11-09T18:22:06+00:00",
+  "git_commit": "2c6273ccec739a86e8ee47ff234dfb9a6fc00d60",
+  "departments": [
+    "83",
+    "13"
+  ],
+  "question": "chance that at least one fire >= 0.1 ha starts in the commune on the target day; issued the morning before, from data up to two days before the target day",
+  "inputs": [
+    "log_usual_rate",
+    "log_local_fwi",
+    "local_wb_30d",
+    "orange_period",
+    "red_period",
+    "rate_x_summer",
+    "dept_13"
+  ],
+  "weights_standardised": {
+    "log_usual_rate": 0.4967,
+    "log_local_fwi": 0.6093,
+    "local_wb_30d": -0.5668,
+    "orange_period": 0.2342,
+    "red_period": 0.2297,
+    "rate_x_summer": 0.6662,
+    "dept_13": 0.0118
+  },
+  "fill_values": {
+    "log_usual_rate": -1.297705,
+    "log_local_fwi": 2.753336,
+    "local_wb_30d": -39.099998,
+    "orange_period": 0.0,
+    "red_period": 0.0,
+    "rate_x_summer": -0.0,
+    "dept_13": 0.0
+  },
+  "training_positive_commune_days": {
+    "83": 902,
+    "13": 1238
+  },
+  "training_years": {
+    "83": [
+      2006,
+      2007,
+      2008,
+      2009,
+      2010,
+      2011,
+      2012,
+      2013,
+      2014,
+      2015,
+      2016,
+      2017,
+      2018,
+      2019,
+      2020,
+      2021,
+      2022,
+      2023,
+      2025
+    ],
+    "13": [
+      2006,
+      2007,
+      2008,
+      2009,
+      2010,
+      2011,
+      2012,
+      2013,
+      2014,
+      2015,
+      2016,
+      2017,
+      2018,
+      2019,
+      2020,
+      2021,
+      2022,
+      2023,
+      2024,
+      2025
+    ]
+  },
+  "model_sha256": "3e4e928b44492b0eedcaca8ad90d3042bc1826b4c0bce34761adaf9beefb69d6",
+  "forecasts_2026": {
+    "kind": "outcome-blinded retrospective test",
+    "rows": 85136,
+    "first_day": "2026-01-01",
+    "last_day": "2026-11-09",
+    "sha256": "783debd1cd6848fb0cd667ef2abc418a2221379c9698366e1e6ead8eae0a856c"
+  }
+}

v3/model.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e4e928b44492b0eedcaca8ad90d3042bc1826b4c0bce34761adaf9beefb69d6
+size 1873

v3/usual_rates.csv

@@ -0,0 +1,273 @@
+department,insee,usual_rate
+83,83001,0.21052631578947367
+83,83002,0.3684210526315789
+83,83003,0.5263157894736842
+83,83004,0.7368421052631579
+83,83005,0.05263157894736842
+83,83006,0.15789473684210525
+83,83007,0.5789473684210527
+83,83008,0.0
+83,83009,0.2631578947368421
+83,83010,0.21052631578947367
+83,83011,0.3157894736842105
+83,83012,0.631578947368421
+83,83013,0.0
+83,83014,0.0
+83,83015,0.2631578947368421
+83,83016,0.2631578947368421
+83,83017,0.2631578947368421
+83,83018,0.21052631578947367
+83,83019,0.47368421052631576
+83,83020,0.10526315789473684
+83,83021,0.21052631578947367
+83,83022,0.10526315789473684
+83,83023,0.5263157894736842
+83,83025,0.10526315789473684
+83,83026,0.47368421052631576
+83,83027,0.3157894736842105
+83,83028,0.42105263157894735
+83,83029,0.2631578947368421
+83,83030,0.21052631578947367
+83,83031,0.5263157894736842
+83,83032,0.21052631578947367
+83,83033,0.47368421052631576
+83,83034,0.47368421052631576
+83,83035,0.3157894736842105
+83,83036,0.15789473684210525
+83,83037,0.10526315789473684
+83,83038,0.05263157894736842
+83,83039,0.05263157894736842
+83,83040,0.10526315789473684
+83,83041,0.42105263157894735
+83,83042,0.10526315789473684
+83,83043,0.5263157894736842
+83,83044,0.15789473684210525
+83,83045,0.3684210526315789
+83,83046,0.42105263157894735
+83,83047,0.21052631578947367
+83,83048,0.21052631578947367
+83,83049,0.21052631578947367
+83,83050,1.5263157894736843
+83,83051,0.8421052631578947
+83,83052,0.15789473684210525
+83,83053,0.21052631578947367
+83,83054,0.10526315789473684
+83,83055,0.3157894736842105
+83,83056,0.5789473684210527
+83,83057,0.3157894736842105
+83,83058,0.6842105263157895
+83,83059,0.10526315789473684
+83,83060,0.15789473684210525
+83,83061,1.0
+83,83062,0.21052631578947367
+83,83063,0.5263157894736842
+83,83064,0.15789473684210525
+83,83065,0.42105263157894735
+83,83066,0.42105263157894735
+83,83067,0.2631578947368421
+83,83068,0.5789473684210527
+83,83069,0.9473684210526315
+83,83070,0.3157894736842105
+83,83071,0.3684210526315789
+83,83072,0.9473684210526315
+83,83073,0.6842105263157895
+83,83074,0.05263157894736842
+83,83075,0.21052631578947367
+83,83076,0.15789473684210525
+83,83077,0.15789473684210525
+83,83078,0.05263157894736842
+83,83079,0.10526315789473684
+83,83080,0.2631578947368421
+83,83081,0.2631578947368421
+83,83082,0.2631578947368421
+83,83083,0.3157894736842105
+83,83084,0.05263157894736842
+83,83085,0.42105263157894735
+83,83086,0.6842105263157895
+83,83087,0.42105263157894735
+83,83088,0.21052631578947367
+83,83089,0.10526315789473684
+83,83090,0.47368421052631576
+83,83091,0.2631578947368421
+83,83092,0.3157894736842105
+83,83093,0.0
+83,83094,0.15789473684210525
+83,83095,0.3684210526315789
+83,83096,0.3157894736842105
+83,83097,0.21052631578947367
+83,83098,0.05263157894736842
+83,83099,0.42105263157894735
+83,83100,0.3157894736842105
+83,83101,0.631578947368421
+83,83102,0.10526315789473684
+83,83103,0.15789473684210525
+83,83104,0.21052631578947367
+83,83105,0.0
+83,83106,0.15789473684210525
+83,83107,0.7368421052631579
+83,83108,0.10526315789473684
+83,83109,0.15789473684210525
+83,83110,0.05263157894736842
+83,83111,0.10526315789473684
+83,83112,0.15789473684210525
+83,83113,0.15789473684210525
+83,83114,0.21052631578947367
+83,83115,0.15789473684210525
+83,83116,0.2631578947368421
+83,83117,0.10526315789473684
+83,83118,0.3157894736842105
+83,83119,0.10526315789473684
+83,83120,0.3684210526315789
+83,83121,0.47368421052631576
+83,83122,0.0
+83,83123,0.2631578947368421
+83,83124,1.0526315789473684
+83,83125,0.10526315789473684
+83,83126,0.42105263157894735
+83,83127,0.42105263157894735
+83,83128,0.0
+83,83129,0.21052631578947367
+83,83130,0.0
+83,83131,0.05263157894736842
+83,83132,0.05263157894736842
+83,83133,0.5263157894736842
+83,83134,0.3684210526315789
+83,83135,0.0
+83,83136,0.15789473684210525
+83,83137,1.631578947368421
+83,83138,0.2631578947368421
+83,83139,0.3684210526315789
+83,83140,0.8421052631578947
+83,83141,0.21052631578947367
+83,83142,0.15789473684210525
+83,83143,0.5263157894736842
+83,83144,0.15789473684210525
+83,83145,0.47368421052631576
+83,83146,0.3684210526315789
+83,83147,0.10526315789473684
+83,83148,0.42105263157894735
+83,83149,0.3157894736842105
+83,83150,0.3684210526315789
+83,83151,0.0
+83,83152,0.10526315789473684
+83,83153,0.47368421052631576
+83,83154,0.47368421052631576
+13,13001,3.75
+13,13002,1.0
+13,13003,0.2
+13,13004,0.4
+13,13005,1.2
+13,13006,0.2
+13,13007,0.8
+13,13008,0.15
+13,13009,0.0
+13,13010,0.3
+13,13011,0.15
+13,13012,0.0
+13,13013,0.05
+13,13014,0.35
+13,13015,0.2
+13,13016,0.25
+13,13017,0.05
+13,13018,0.0
+13,13019,0.95
+13,13020,0.1
+13,13021,0.1
+13,13022,0.45
+13,13023,0.1
+13,13024,0.05
+13,13025,0.35
+13,13026,0.6
+13,13027,0.1
+13,13028,0.8
+13,13029,0.05
+13,13030,0.2
+13,13031,0.05
+13,13032,0.45
+13,13033,0.5
+13,13034,0.2
+13,13035,0.15
+13,13036,0.15
+13,13037,0.5
+13,13038,0.65
+13,13039,1.4
+13,13040,0.85
+13,13041,0.65
+13,13042,0.1
+13,13043,0.35
+13,13044,0.2
+13,13045,0.2
+13,13046,0.1
+13,13047,3.15
+13,13048,0.6
+13,13049,0.35
+13,13050,0.7
+13,13051,1.4
+13,13052,0.0
+13,13053,0.05
+13,13054,0.15
+13,13055,5.85
+13,13056,4.4
+13,13057,0.05
+13,13058,0.6
+13,13059,0.35
+13,13060,0.5
+13,13061,0.0
+13,13062,0.1
+13,13063,1.6
+13,13064,0.1
+13,13065,0.65
+13,13066,0.25
+13,13067,0.45
+13,13068,0.15
+13,13069,0.4
+13,13070,0.05
+13,13071,1.95
+13,13072,0.2
+13,13073,0.35
+13,13074,0.1
+13,13075,0.25
+13,13076,0.25
+13,13077,1.55
+13,13078,0.05
+13,13079,0.3
+13,13080,0.55
+13,13081,1.6
+13,13082,0.4
+13,13083,0.0
+13,13084,0.35
+13,13085,0.2
+13,13086,0.5
+13,13087,0.5
+13,13088,1.0
+13,13089,0.0
+13,13090,0.0
+13,13091,0.5
+13,13092,0.65
+13,13093,0.15
+13,13094,0.05
+13,13095,0.1
+13,13096,0.05
+13,13097,0.4
+13,13098,1.0
+13,13099,0.15
+13,13100,0.25
+13,13101,0.2
+13,13102,0.25
+13,13103,1.1
+13,13104,0.15
+13,13105,0.0
+13,13106,0.95
+13,13107,0.25
+13,13108,0.4
+13,13109,0.2
+13,13110,0.55
+13,13111,0.05
+13,13112,1.0
+13,13113,0.15
+13,13114,0.6
+13,13115,0.1
+13,13116,0.0
+13,13117,0.75
+13,13118,0.5
+13,13119,0.0