feat: mark Delhi/NCR on maps; reframe to Delhi; polish README

- visualizer: overlay Delhi star + NCR bounding box on full WRF-India domain
- app: enable mark_delhi, retitle subtitle/footer to Delhi/NCR (IIT-Delhi)
- README: restructured with badges, tables, run-local, PM2.5 guide

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

README.md

@@ -9,7 +9,7 @@ python_version: "3.12"
 app_file: app.py
 pinned: false
 license: mit
-short_description: PM2.5 forecast demo with saved predictions
+short_description: Delhi / NCR PM2.5 forecast demo with saved predictions
 datasets:
   - sumit1703/pm25-forecasting-data
 tags:
@@ -17,72 +17,131 @@ tags:
   - air-quality
   - pm25
   - pollution-forecasting
+  - delhi
   - deep-learning
   - data-visualization
 suggested_hardware: cpu-basic
 ---
 
-# 🌬️ PM2.5 Pollution Forecasting Demo
+<div align="center">
 
-**ANRF AISEHack Phase 2 — Theme 2 — Pollution Forecasting (IIT Delhi)**
+# 🌬️ PM2.5 Pollution Forecasting — Delhi / NCR
 
-This demo visualizes predictions from a **ConvLSTM + Fourier Neural Operator (FNO)** hybrid model
-trained to forecast PM2.5 air pollution levels across a 140×124 spatial grid over Northern India.
+**Deep-learning air-quality forecasting for Delhi and the National Capital Region**
 
-## Live Links
+_ANRF · AISEHack Phase 2 · Theme 2 — Pollution Forecasting · **IIT Delhi**_
 
-- Live Demo: https://huggingface.co/spaces/sumit1703/pm25-forecasting
-- Dataset: https://huggingface.co/datasets/sumit1703/pm25-forecasting-data
-- GitHub: https://github.com/sumitjadhav1703/pm25-forecasting-demo
+[![Live Demo](https://img.shields.io/badge/🤗_Live_Demo-HF_Spaces-blue)](https://huggingface.co/spaces/sumit1703/pm25-forecasting)
+[![Dataset](https://img.shields.io/badge/🤗_Dataset-pm25--forecasting--data-yellow)](https://huggingface.co/datasets/sumit1703/pm25-forecasting-data)
+[![GitHub](https://img.shields.io/badge/GitHub-source-black?logo=github)](https://github.com/sumitjadhav1703/pm25-forecasting-demo)
+[![Rank](https://img.shields.io/badge/Phase_2_Rank-🥈_2nd-silver)](https://www.kaggle.com/competitions/anrf-aise-hack-phase-2-theme-2-pollution-forecasting-iitd)
 
-## Important Note
+</div>
 
-This Space visualizes precomputed PM2.5 predictions saved from the Kaggle GPU run. It does not run live model inference, training, or torch at runtime.
+---
+
+## Overview
+
+A **ConvLSTM + Fourier Neural Operator (FNO)** hybrid model forecasts PM2.5 up to **16 hours ahead**.
+Predictions run on a WRF simulation grid (140 × 124, ~25 km) covering India, with **Delhi and the
+surrounding NCR highlighted** on every map — the focus region of the IIT-Delhi competition.
+
+> **Note:** This Space visualizes precomputed predictions saved from the Kaggle GPU run.
+> It does **not** run live model inference, training, or `torch` at runtime.
+
+---
+
+## 🖱️ How to Use
 
-## How to Use
+| Step | Action |
+|------|--------|
+| 1 | Drag the **Test Window** slider to pick a time period from the test set |
+| 2 | Drag the **Forecast Hour** slider to choose how far ahead (+1h → +16h) |
+| 3 | Compare the last-known map (left) with the model's forecast (right) — **Delhi / NCR is boxed and starred** |
+| 4 | Read the forecast statistics below the maps |
 
-1. Use the **Test Window** slider to select a time period from the test dataset
-2. Use the **Forecast Hour** slider to select how far ahead (+1h to +16h)
-3. Compare the last known PM2.5 map (left) with the model's forecast (right)
-4. Read the statistics below the maps
+---
 
-## Model Architecture
+## 🧠 Model Architecture
 
 | Component | Details |
 |-----------|---------|
-| Encoder   | Stacked ConvLSTM (2 layers) |
-| Spatial   | Fourier Neural Operator (FNO) |
-| Decoder   | UNet with SE blocks |
-| Input     | 10 hours × 20 atmospheric features × 140×124 grid |
-| Output    | 16-hour PM2.5 forecast |
-| Training  | Kaggle T4 GPU, ~8 hours |
+| **Encoder** | Stacked ConvLSTM (2 layers) |
+| **Spatial operator** | Fourier Neural Operator (FNO) |
+| **Decoder** | UNet with SE blocks |
+| **Input** | 10 hours × 20 atmospheric features × 140 × 124 grid |
+| **Output** | 16-hour PM2.5 forecast |
+| **Training** | Kaggle T4 GPU · ~8 hours |
+
+---
 
-## Competition Results
+## 🏆 Competition Results
 
-- **Competition:** ANRF AISEHack Phase 2 — Theme 2 (IIT Delhi)  
-- **Team:** Binary Bombers  
-- **Phase 2 Rank:** 2  
-- **Final Score:** 0.8795 (sMAPE-based)
+| | |
+|---|---|
+| **Competition** | ANRF · AISEHack Phase 2 · Theme 2 — Pollution Forecasting (IIT Delhi) |
+| **Team** | Binary Bombers |
+| **Phase 2 Rank** | 🥈 **2** of 21 |
+| **Final Score** | **0.8795** (sMAPE-based) |
 
-## Kaggle Leaderboard Proof
+<details>
+<summary><strong>📊 Kaggle leaderboard proof</strong></summary>
 
-The final private leaderboard for **ANRF - AISEHack - Phase 2 - Theme 2 - Pollution Forecasting (IITD)** shows:
+<br>
 
-* **Team:** Binary Bombers
-* **Final Rank:** 2
-* **Final Score:** 0.8795
-* **Entries:** 21
+Final private leaderboard for **ANRF – AISEHack – Phase 2 – Theme 2 – Pollution Forecasting (IITD)**:
 
-> Note: Kaggle competition pages may require login to view the leaderboard.
+- **Team:** Binary Bombers
+- **Final Rank:** 2
+- **Final Score:** 0.8795
+- **Entries:** 21
 
 ![Kaggle Phase 2 Rank 2 Proof](assets/proof/kaggle-phase2-rank2.png)
 
-Competition link: https://www.kaggle.com/competitions/anrf-aise-hack-phase-2-theme-2-pollution-forecasting-iitd
+_Kaggle competition pages may require login to view the leaderboard._
+[View competition →](https://www.kaggle.com/competitions/anrf-aise-hack-phase-2-theme-2-pollution-forecasting-iitd)
+
+</details>
+
+---
+
+## 📦 Dataset
+
+4 months of WRF-simulated atmospheric data — **APRIL_16, JULY_16, OCT_16, DEC_16** —
+covering the India domain with a focus on Delhi / NCR air quality. Features include
+PM2.5, wind components, temperature, PBLH, and various emission tracers.
+
+Hosted at 🤗 [`sumit1703/pm25-forecasting-data`](https://huggingface.co/datasets/sumit1703/pm25-forecasting-data).
+
+---
+
+## 🚀 Run Locally
+
+```bash
+git clone https://github.com/sumitjadhav1703/pm25-forecasting-demo
+cd pm25-forecasting-demo
+pip install -r requirements.txt
+python app.py          # serves on http://0.0.0.0:7860
+```
+
+Predictions download automatically from the HF dataset on first launch (cached to `/tmp`).
+
+---
+
+## 🗺️ PM2.5 Guide
+
+| Range (μg/m³) | Level |
+|---|---|
+| 0 – 15 | 🟢 Good |
+| 15 – 35 | 🟡 Moderate |
+| 35 – 55 | 🟠 Sensitive |
+| 55 – 150 | 🔴 Unhealthy |
+| 150+ | 🟣 Hazardous |
+
+---
 
-## Dataset
+<div align="center">
 
-The competition dataset contains 4 months of WRF-simulated atmospheric data:
-APRIL_16, JULY_16, OCT_16, DEC_16. Features include PM2.5, wind components,
-temperature, PBLH, and various emission tracers.
+Built by **Sumit** — B.Tech AI &amp; Data Science, JNEC MGM University
 
-> Built by Sumit — B.Tech AI & Data Science, JNEC MGM University
+</div>

app.py

@@ -50,6 +50,7 @@ def update(window_slider: int, hour_slider: int):
         title=f"Input PM2.5 — Last Known Hour\n(Test window {original_window})",
         vmin=vmin, vmax=vmax,
         lat=LAT, lon=LON,
+        mark_delhi=True,
     )
 
     pred_img = make_heatmap(
@@ -57,6 +58,7 @@ def update(window_slider: int, hour_slider: int):
         title=f"Predicted PM2.5 — +{h + 1}h Forecast\n(Test window {original_window})",
         vmin=vmin, vmax=vmax,
         lat=LAT, lon=LON,
+        mark_delhi=True,
     )
 
     stats = compute_stats(pred_frame, input_frame)
@@ -71,8 +73,8 @@ with gr.Blocks(
 ) as demo:
 
     gr.Markdown("# 🌬️ PM2.5 Pollution Forecasting")
-    gr.HTML('<p class="subtitle">ANRF AISEHack Phase 2 — Deep Learning Air Quality Forecast over India</p>')
-    gr.Markdown("This demo visualizes precomputed predictions from saved `.npy` files. It does not run live model inference.")
+    gr.HTML('<p class="subtitle">ANRF AISEHack Phase 2 (IIT Delhi) — Deep Learning PM2.5 Forecast · Delhi / NCR</p>')
+    gr.Markdown("This demo visualizes precomputed predictions from saved `.npy` files. It does not run live model inference. **Delhi and the surrounding NCR are marked on each map.**")
 
     with gr.Row():
         with gr.Column(scale=1):
@@ -112,7 +114,7 @@ with gr.Blocks(
 ---
 **Model:** ConvLSTM encoder + Fourier Neural Operator (FNO) hybrid  
 **Training:** Kaggle T4 GPU · ANRF competition dataset · 4 months × 16 atmospheric features  
-**Grid:** 140 × 124 spatial points · Northern India  
+**Grid:** 140 × 124 spatial points · WRF domain over India · **Delhi / NCR highlighted**
 **Input:** 10 hours of atmospheric data → **Output:** 16-hour PM2.5 forecast  
 **Competition Rank:** 2 · Final Score: 0.8795 (sMAPE-based)
 """)

visualizer.py

@@ -5,8 +5,14 @@ import matplotlib
 matplotlib.use("Agg")   # non-interactive backend — required for server/Gradio
 import matplotlib.pyplot as plt
 import matplotlib.colors as mcolors
+from matplotlib.patches import Rectangle
 from PIL import Image
 
+# Delhi / NCR reference geometry (used to annotate the full-domain map)
+DELHI_LON, DELHI_LAT = 77.21, 28.61          # Delhi city centre
+NCR_LON = (74.0, 80.0)                        # National Capital Region bounds
+NCR_LAT = (26.0, 31.0)
+
 # WHO/AEQ PM2.5 breakpoints (μg/m³) for color bands annotation
 PM25_LEVELS = [
     (0,   15,  "Good",      "#00e400"),
@@ -26,6 +32,7 @@ def make_heatmap(
     lon:      np.ndarray = None,
     figsize:  tuple = (6, 5),
     dpi:      int   = 110,
+    mark_delhi: bool = False,
 ) -> Image.Image:
     """
     Render a PM2.5 spatial heatmap.
@@ -83,6 +90,31 @@ def make_heatmap(
     for spine in ax.spines.values():
         spine.set_edgecolor("#444444")
 
+    # Delhi / NCR annotation on the full geographic domain
+    if mark_delhi and lat is not None and lon is not None:
+        ax.add_patch(Rectangle(
+            (NCR_LON[0], NCR_LAT[0]),
+            NCR_LON[1] - NCR_LON[0],
+            NCR_LAT[1] - NCR_LAT[0],
+            fill=False, edgecolor="#00e5ff", linewidth=1.4,
+            linestyle="--", zorder=5,
+        ))
+        ax.plot(
+            DELHI_LON, DELHI_LAT,
+            marker="*", markersize=12,
+            markerfacecolor="#00e5ff", markeredgecolor="black",
+            markeredgewidth=0.6, zorder=6, label="Delhi (NCR)",
+        )
+        ax.annotate(
+            "Delhi",
+            xy=(DELHI_LON, DELHI_LAT),
+            xytext=(DELHI_LON + 1.6, DELHI_LAT + 1.4),
+            color="#00e5ff", fontsize=8, fontweight="bold", zorder=6,
+        )
+        leg = ax.legend(loc="lower left", fontsize=7, framealpha=0.35)
+        for txt in leg.get_texts():
+            txt.set_color("white")
+
     ax.set_title(title, color="white", fontsize=9, pad=8, fontweight="bold")
 
     plt.tight_layout()