Lesson 3 · 11:30–12:30

Who is at risk?

A hot surface is a risk where people are packed in without shade: risk = surface temperature × crowding × lack of shade.

Two maps from space, one headcount

Surface temperature and tree cover come from satellites on the same 30 m squares (lessons 1 and 2). People come from the Census 2011 headcount of each ward, carried in the GBA ward file: the latest official count, as India's next census counts people in February 2027. The ward map with every headcount is on OpenCity (KML: opens in QGIS or Google Earth). Download them:
day_surface_temp = datapack.layer("lst_day")  # °C, Landsat at 10:40 am (lesson 1)
night_surface_temp = datapack.layer("lst_night")  # °C, MODIS at night (lesson 1)
trees = datapack.layer("canopy_frac")  # 0-1 tree cover, Meta/WRI tree heights (lesson 2)
wards = load_wards()  # the 369 GBA wards, with their Census 2011 headcount
wards[["label", "census2011_pop", "area_km2"]].head()  # area in km²
label census2011_pop area_km2
Central 1 · Ramaswamy Palya 26,124 0.7
Central 2 · Jayamahal 25,834 1.423
Central 3 · Vasanth Nagar 28,272 1.946
Central 4 · Sampangirama Nagar 27,874 4.676
Central 5 · Shivajinagar 25,484 1.163

One row per ward

Average each map inside every ward with zonal_stats (lesson 1), and divide each ward's headcount by its area (1 km² = 100 hectares):
labels = ward_labels(wards, datapack.grid30())  # each square -> its ward

f = zonal_stats({
    "day_surface_temp": day_surface_temp,
    "night_surface_temp": night_surface_temp,
    "trees": trees,
}, labels).rename(columns=lambda c: c.removesuffix("_mean"))
w = wards.set_index("ward_uid")
f.insert(0, "ward", w["label"])
f["people_per_ha"] = w["census2011_pop"] / (w["area_km2"] * 100)  # Census 2011 people ÷ hectares
f["surface_temp"] = (f.day_surface_temp + f.night_surface_temp) / 2  # one surface temperature: morning and night
f = f.round({"day_surface_temp": 2, "night_surface_temp": 2, "surface_temp": 2, "trees": 3, "people_per_ha": 0})
f.head()
ward day_surface_temp night_surface_temp trees people_per_ha surface_temp
Central 1 · Ramaswamy Palya 38.48 24.13 0.213 373 31.3
Central 2 · Jayamahal 37.66 24.07 0.32 182 30.86
Central 3 · Vasanth Nagar 37.15 24.12 0.27 145 30.64
Central 4 · Sampangirama Nagar 36.7 24.45 0.402 60 30.57
Central 5 · Shivajinagar 39.47 24.15 0.185 219 31.81

From three numbers to one score

The three numbers are in different units, so we rescale each from 0 for the city's lowest ward to 1 for its highest: surface temperature from 29.20 °C (0) to 33.33 °C (1), people from 12 (0) to 1,337 per hectare (1), and lack of shade from 55.5 % tree cover (0) to 1.0 % (1). Multiplying the three gives a high score only to a ward that is hot, packed and treeless at once: the top 30 % of scores are High risk, the next 30 % Medium and the rest Low. The score compares wards with each other, so it says where to act first.
def scale(x):  # 0 = the city's lowest ward, 1 = its highest
    return (x - x.min()) / (x.max() - x.min())

hot = scale(f.surface_temp)
crowded = scale(f.people_per_ha)
no_shade = scale(-f.trees)  # 1 = the ward with the fewest trees

risk = hot * crowded * no_shade  # high only if all three are high
Central 52 · Padarayanapura: 33.05 °C surface temperature → (33.05 − 29.20) ÷ (33.33 − 29.20) = 0.932; 1,113 people per hectare → 0.831; 2.7% tree cover → 0.969. Score: 0.932 × 0.831 × 0.969 = 0.750, the city's highest.

Your weights

How much should each count? Each slider is a power: 1 counts it once, 2 counts it twice, 0 leaves it out. The top 30 % of wards are High risk.

Try it

  1. Set People to 0.
  2. Set Lack of shade to 3.
  3. A plain average instead of multiplying:
plain = (hot + crowded + no_shade) / 3  # a plain average instead of multiplying
compare = f[["ward"]].assign(multiplied=risk.rank(ascending=False), plain=plain.rank(ascending=False))
compare.assign(moved=compare.multiplied - compare.plain).nlargest(10, "moved")
ward multiplied plain moved
ward_uid
225 South 40 · Anjanapura 355 191 164
162 East 27 · Vibhootipura 315 155 160
141 East 6 · Kalkere 296 150 146
273 West 16 · Ullal 305 162 143
269 West 12 · Andrahalli 288 157 131
219 South 34 · Chikkathoguru 329 209 120
270 West 13 · Nada prabhu Kempegowda Nagara 268 159 109
237 South 52 · Talagattapura 366 261 105
220 South 35 · Vishwapriya Nagara 238 134 104
82 North 19 · Thanisandra 207 106 101
moved = how many places a ward climbs when the three are averaged instead of multiplied. A risk map shows where to look first; check it on the ground.
Made with open data and open-source Python · marimo · Altair