Precipitation - Exploring Rainfall Across the UK

Goal: Using the Imago precipitation product to explore rainfall across the UK, discovering the wettest and driest MSOAs and the regional precipitation characteristics.

Data:

NoteInstalling the data

The Imago precipitation data is hosted on the data catalogue. To download it:

  1. Sign up for a free account on the data catalogue and log in.
  2. Search for precipitation MSOA (or browse the precipitation datasets).
  3. Open the data link, Precipitation per MSOA in 2024, and download the .gpkg file.
  4. Save the files into a ../data/ folder in your project, keeping the original filenames (precipitation_indicators_MSOA_2024.gpkg).

For this notebook, you’ll need the 2024 MSOA file, and then repeat for the 2024 LAD file.

A question to motivate the data

A local authority believes that 2024 was an exceptionally wet year. Before investigating whether rainfall was unusual, we first need to understand how rainfall was distributed across the UK and whether wet conditions were concentrated in specific regions or spread more evenly across the country.

Installing Libraries

We install core geospatial libraries needed for this work.

# Geospatial Libraries
import pandas as pd
import geopandas as gpd
import numpy as np

# Plotting
import matplotlib.pyplot as plt

Loading the Data

Using geopandas, we can read the gpkg. As it’s a gpkg it already has the geometric shape of the areas (in the geometry column) and will not need a separate boundary file.

precip_msoa = gpd.read_file("data/precipitation_indicators_MSOA_2024.gpkg")
precip_msoa.head()
dt_zn_c dt_zn_n rainfall_annual_mm winter_rainfall spring_rainfall summer_rainfall autumn_rainfall annual_anomaly_abs annual_anomaly_std extreme_days_4sd year geometry
0 E02000001 City of London 001 696.647095 223.671677 189.220276 107.580292 218.248856 158.290131 2.148560 26.304880 2024 MULTIPOLYGON (((532135.138 182198.131, 532158....
1 E02000002 Barking and Dagenham 001 695.338257 236.243835 192.781601 106.833694 205.600983 131.554367 1.715453 23.520077 2024 MULTIPOLYGON (((548881.563 190845.265, 548881....
2 E02000003 Barking and Dagenham 002 667.498779 230.828064 185.345505 100.507462 195.728653 120.716499 1.595210 22.027685 2024 MULTIPOLYGON (((549102.438 189324.625, 548954....
3 E02000004 Barking and Dagenham 003 611.439758 218.056824 171.161316 89.372581 175.921600 93.010658 1.238803 23.146259 2024 MULTIPOLYGON (((551550.056 187364.705, 551478 ...
4 E02000005 Barking and Dagenham 004 632.301819 224.054337 176.073090 93.059616 183.112747 102.117798 1.371893 22.568747 2024 MULTIPOLYGON (((549099.634 187656.076, 549161....

This file contains multiple rainfall indicators. For this notebook we use the following two:

  • rainfall_annual_mm: Average yearly rainfall depth in mm. It measures the average depth of rain that fell across a specific local area (MSOA) during the year, rather than a total volume.
  • geometry: spatial boundaries for each MSOA.

Understanding the Distribution of Rainfall

Before mapping the data locally, it is useful to look at the national summary statistics. These provide a high-level overview of the rainfall conditions observed across the whole of the UK during 2024.

(precip_msoa[["rainfall_annual_mm"]].agg(["mean", "std", "min", "max"]))
rainfall_annual_mm
mean 972.689669
std 274.112145
min 516.364502
max 3140.853516

Rainfall is rarely distributed evenly. Some areas receive substantially more rainfall than others, and understanding this variation helps provide context for local claims about exceptionally wet conditions.

Calculating the bins

To calculate bins for the histograms Freedman-Diaconis rule is used because it adapts bin width based on the variability of the data while limiting the influence of outliers. This is suitable for rainfall related variables, where distributions can be skewed due to regional differences and extreme events.

# Freedman-Diaconis Calculations
data = precip_msoa["rainfall_annual_mm"]
data_points = len(data)
q1 = data.quantile(0.25)
q3 = data.quantile(0.75)

# Calculate interquartile range
iqr = q3 - q1

bin_width = (2 * iqr) / (data_points ** (1 / 3))

ann_rainfall_bins = int(np.ceil((data.max() - data.min()) / bin_width))
ann_rainfall_bins = round(ann_rainfall_bins, 0)
print(f"Amount of rainfall bins: {ann_rainfall_bins}")
Amount of rainfall bins: 92
fig, ax = plt.subplots(1, figsize=(7, 4))

ax.hist(
    precip_msoa["rainfall_annual_mm"].dropna(),
    bins=ann_rainfall_bins,
    edgecolor="black",
    color="darkslategrey",
)
ax.set_title("Annual Rainfall (mm)", fontweight="bold")
ax.set_ylabel("Count")
ax.grid(axis="y", linestyle="--", alpha=0.3)
ax.grid(axis="x", visible=False)

fig.suptitle("Distribution of rainfall exposure across MSOAs, 2024", fontweight="bold")
plt.tight_layout()
plt.show()

NoteInterpretation

The distribution is positively skewed, with most MSOAs clustered around moderate rainfall totals and a smaller number of particularly wet locations extending the upper tail of the distribution. This is highlighting how the majority of MSOAs experience a relatively low amount of rainfall however the rainfall countrywide varies with a tail of progressively higher means.

Finding the extreme MSOAs

One simple way to understand rainfall concentration is to rank areas by total annual rainfall.

wettest = precip_msoa.nlargest(10, "rainfall_annual_mm")[
    ["dt_zn_c", "dt_zn_n", "rainfall_annual_mm"]
]
driest = precip_msoa.nsmallest(10, "rainfall_annual_mm")[
    ["dt_zn_c", "dt_zn_n", "rainfall_annual_mm"]
]

print("Wettest MSOAs, 2024:")
print(wettest.to_string(index=False))

print("\nDriest MSOAs, 2024:")
print(driest.to_string(index=False))
Wettest MSOAs, 2024:
  dt_zn_c               dt_zn_n  rainfall_annual_mm
E02003976         Allerdale 012         3140.853516
E02004015    South Lakeland 001         3128.609863
W02000258 Rhondda Cynon Taf 007         3121.053467
S02003309              Lochalsh         2909.670654
W02000261 Rhondda Cynon Taf 010         2838.054199
S02003281    Fort William South         2818.538086
W02000262 Rhondda Cynon Taf 011         2793.097412
S02003310            Skye South         2776.712891
W02000260 Rhondda Cynon Taf 009         2737.283691
S02003279         Lochaber West         2706.211182

Driest MSOAs, 2024:
  dt_zn_c             dt_zn_n  rainfall_annual_mm
E02003291 Southend-on-Sea 013          516.364502
E02003311        Thurrock 016          517.631042
E02003304        Thurrock 009          517.999329
E02005010      Canterbury 001          519.244873
E02006926        Thurrock 020          523.255981
E02005011      Canterbury 002          523.720581
E02007005        Thurrock 022          527.915466
E02006859        Thurrock 019          530.438965
E02003312        Thurrock 017          530.513855
E02004560          Maldon 006          530.607605
NoteOutliers

A single year’s ranking is sensitive to one unusual season. An MSOA at the top of this list in 2024 or the bottom of this list is solely for this year, and it does not reveal if it is part of a bigger regional pattern, a specific seasonal spike or just an isolated high rainfall year.

Mapping rainfall by local area

Mapping rainfall reveals the spatial structure hidden within the rankings. Neighbouring MSOAs often experience similar conditions, producing regional rainfall patterns that are not obvious from tables alone.

fig, ax = plt.subplots(1, figsize=(8, 6))

precip_msoa.plot(
    column="rainfall_annual_mm",
    cmap="Blues",
    legend=True,
    ax=ax,
    linewidth=0,
    legend_kwds={"shrink": 0.5, "label": "mm"},
)
ax.set_title("Annual rainfall", fontweight="bold")
ax.set_axis_off()

fig.suptitle("Rainfall exposure across MSOAs, 2024", fontweight="bold")
plt.tight_layout()
plt.show()

Patterns

Beyond serving as a useful sanity check, mapping the data reveals spatial patterns that highlight regional variations in rainfall. We can clearly see that certain parts of the country get lots of rainfall (Scotland, Lancashire, and Wales) while other parts such as the south of England do not. We can see a clear southeast-to-northwest gradient of increasing rainfall.

Regional differences

So far we’ve looked at MSOAs. Imago also publishes a LAD level version of the same precipitation product with rainfall computed directly at the local authority level, rather than something we would have to derive ourselves by averaging MSOAs up. Using this pre-aggregated product is much simpler than trying to average MSOA values ourselves.

Loading LAD Data

This data is loaded in the same way as the MSOA data, and since it is in GeoPackage (.gpkg) format, it already includes the boundaries. We print the first observations of the dataset to inspect its structure.

precip_lad = gpd.read_file("data/precipitation_indicators_LAD_2024.gpkg")
precip_lad.head()
LAD21CD LAD21NM rainfall_annual_mm winter_rainfall spring_rainfall summer_rainfall autumn_rainfall annual_anomaly_abs annual_anomaly_std extreme_days_4sd year geometry
0 E06000001 Hartlepool 686.761658 217.900101 194.122314 140.255630 175.145996 110.656914 1.173922 21.315712 2024 MULTIPOLYGON (((447213.9 537036.104, 447206.00...
1 E06000002 Middlesbrough 741.491333 192.090210 209.281387 154.297791 201.528397 119.056717 1.304255 19.883654 2024 MULTIPOLYGON (((448609.9 521982.6, 448586.35 5...
2 E06000003 Redcar and Cleveland 865.258606 225.683945 230.743408 182.541229 234.379837 159.905365 1.732852 20.556686 2024 MULTIPOLYGON (((455932.335 527880.697, 455939....
3 E06000004 Stockton-on-Tees 693.715271 203.911972 199.658813 135.914932 192.368607 104.323410 1.147774 20.970232 2024 MULTIPOLYGON (((444157.002 527956.304, 444139....
4 E06000005 Darlington 716.726807 240.145233 214.126801 129.527863 199.460587 90.094154 0.950165 20.299484 2024 MULTIPOLYGON (((423496.602 524724.299, 423475....

To examine the extremes of the distribution, we look at the wettest and driest local authorities in 2024.

wettest_lad = precip_lad.nlargest(10, "rainfall_annual_mm")[
    ["LAD21CD", "LAD21NM", "rainfall_annual_mm"]
]

driest_lad = precip_lad.nsmallest(10, "rainfall_annual_mm")[
    ["LAD21CD", "LAD21NM", "rainfall_annual_mm"]
]

print("Wettest local authorities, 2024:")
print(wettest_lad.to_string(index=False))
print("Driest local authorities, 2024:")
print(driest_lad.to_string(index=False))
Wettest local authorities, 2024:
  LAD21CD           LAD21NM  rainfall_annual_mm
W06000016 Rhondda Cynon Taf         2264.987793
E07000031    South Lakeland         2228.302002
W06000012 Neath Port Talbot         2155.255127
S12000035   Argyll and Bute         2122.896973
E07000029          Copeland         2113.462891
W06000002           Gwynedd         2104.406250
W06000013          Bridgend         2029.671753
S12000030          Stirling         2024.052734
W06000024    Merthyr Tydfil         1982.400146
S12000017          Highland         1925.752808
Driest local authorities, 2024:
  LAD21CD              LAD21NM  rainfall_annual_mm
E07000075             Rochford          565.752625
E06000033      Southend-on-Sea          568.152588
E07000074               Maldon          569.419800
E07000114               Thanet          579.956970
E06000034             Thurrock          584.442139
E07000069         Castle Point          584.549500
E07000076             Tendring          598.346802
E09000002 Barking and Dagenham          619.939941
E07000071           Colchester          624.941345
E09000016             Havering          633.256348

We can create a horizontal bar chart comparing the 15 wettest and 15 driest local authorities to highlight the absolute differences between the extremes. We then generate a choropleth map of the UK to show how this annual rainfall is distributed geographically across all local authorities.

fig, ax = plt.subplots(figsize=(7, 10))

lad_sorted = precip_lad.sort_values("rainfall_annual_mm", ascending=False)
top_bottom = pd.concat([lad_sorted.head(15), lad_sorted.tail(15)])

ax.barh(top_bottom["LAD21NM"], top_bottom["rainfall_annual_mm"], color="darkslategrey")
ax.set_xlabel("Annual rainfall (mm)")
ax.set_title("Local authorities: wettest and driest 15, 2024", fontweight="bold")
ax.invert_yaxis()
plt.tight_layout()
plt.show()

fig, ax = plt.subplots(figsize=(7, 8))

precip_lad.plot(
    column="rainfall_annual_mm",
    cmap="Blues",
    legend=True,
    ax=ax,
    linewidth=0.2,
    edgecolor="white",
    legend_kwds={"shrink": 0.6, "label": "mm"},
)
ax.set_title("Annual rainfall by local authority, 2024", fontweight="bold")
ax.set_axis_off()

plt.tight_layout()
plt.show()

Comparing the two scales

Next, we examine whether the trends observed at the LAD level align with the finer-grained MSOA patterns. If a LAD ranks among the wettest, do its individual MSOAs from earlier in the notebook also sit toward the top of the MSOA ranking? Alignment between these levels confirms that regional trends are representative of the area, rather than being driven by isolated outliers that distort the local authority average. Divergence between the two scales is also valuable; a local authority with near-average overall rainfall might still contain an individual MSOA with extreme exposure, which would be obscured by the aggregate LAD average alone

Mapping the gap between scales

We can visualise this comparison by mapping the deviation of each MSOA’s rainfall from its local authority’s average. A value close to zero means the MSOA is typical of its local authority district, while large positive or negative values highlight localised areas with unusually high or low rainfall compared to the regional average.

# Spatial overlay in order to compare
msoa_lad_overlap = gpd.overlay(
    precip_msoa[["dt_zn_c", "geometry"]],
    precip_lad[["LAD21CD", "LAD21NM", "rainfall_annual_mm", "geometry"]].rename(
        columns={"rainfall_annual_mm": "lad_rainfall_annual_mm"}
    ),
    how="intersection",
    keep_geom_type=False,
)

# Keep the LAD each MSOA overlaps with most, by area
msoa_lad_overlap["overlap_area"] = msoa_lad_overlap.geometry.area
best_match = msoa_lad_overlap.sort_values(
    "overlap_area", ascending=False
).drop_duplicates(subset="dt_zn_c", keep="first")[
    ["dt_zn_c", "LAD21CD", "LAD21NM", "lad_rainfall_annual_mm"]
]

msoa_with_lad = precip_msoa.merge(best_match, on="dt_zn_c", how="left")

n_unmatched = msoa_with_lad["LAD21CD"].isna().sum()
print(
    f"{n_unmatched} MSOAs have no matching LAD (The North of Ireland is not covered by the LAD precipitation product)"
)

msoa_with_lad = msoa_with_lad.dropna(subset=["LAD21CD"])

msoa_with_lad["rainfall_diff"] = (
    msoa_with_lad["rainfall_annual_mm"] - msoa_with_lad["lad_rainfall_annual_mm"]
)

# Clip the colour scale to the 1st/99th percentile rather than true min/max
# A small number of genuine outliers (like large, internally varied Highland LADs) would compress the entire typical range into a narrow band near white.
# Outliers beyond this range are still shown, just fully saturated.
p01 = msoa_with_lad["rainfall_diff"].quantile(0.01)
p99 = msoa_with_lad["rainfall_diff"].quantile(0.99)
diff_limit = max(abs(p01), abs(p99))

fig, ax = plt.subplots(figsize=(9, 9))
fig.patch.set_facecolor("#dfdddd")
ax.set_facecolor("#dfdddd")

msoa_with_lad.plot(
    column="rainfall_diff",
    cmap="RdBu",
    legend=True,
    ax=ax,
    linewidth=0,
    vmin=-diff_limit,
    vmax=diff_limit,
    legend_kwds={
        "shrink": 0.6,
        "label": "mm (MSOA - parent LAD, clipped at 1st/99th pct)",
    },
)
ax.set_title(
    "MSOA rainfall relative to its own local authority, 2024", fontweight="bold"
)
ax.set_axis_off()
plt.tight_layout()
plt.show()

extremes = msoa_with_lad.nlargest(3, "rainfall_diff")[["dt_zn_n", "rainfall_diff"]]
print(extremes.to_string(index=False))

extremes_low = msoa_with_lad.nsmallest(3, "rainfall_diff")[["dt_zn_n", "rainfall_diff"]]
print(extremes_low.to_string(index=False))
850 MSOAs have no matching LAD (The North of Ireland is not covered by the LAD precipitation product)

           dt_zn_n  rainfall_diff
     Allerdale 012    1412.004883
          Lochalsh     983.917847
South Lakeland 001     900.307861
           dt_zn_n  rainfall_diff
          Seaboard   -1134.862122
Inverness Merkinch   -1107.408691
Inverness Muirtown   -1092.341003
Note

Most MSOAs sit close to their LAD’s average, appearing pale across much of England. Scotland and Wales are the exception with Lochalsh sitting 984 mm above its LAD average while Inverness sits 1107 mm below. These larger, more topographically varied local authorities can contain substantial rainfall differences within a single administrative boundary. England’s generally smaller LADs show less internal variation.

This highlights an important point when interpreting regional rainfall statistics. A local authority average can provide a useful summary of overall conditions, but it may also hide variation between individual communities. Looking at both MSOA and LAD scales helps reveal whether rainfall patterns are widespread across an authority or concentrated in particular areas.

What have we learnt?

This first case study explored how rainfall was distributed across the UK during 2024.

Understanding where rainfall was concentrated is the first step in investigating claims about exceptionally wet conditions. However, annual rainfall totals alone do not explain why an area was wet.

Two places can record similar annual rainfall totals while experiencing very different seasonal patterns. One may receive rainfall steadily throughout the year, while another may be heavily influenced by a particularly wet season.

We found that:

  • Rainfall is not evenly distributed across the country.
  • Clear regional patterns emerge when rainfall is mapped.
  • There are multiple ways to identify rainfall extremes, such as sorting rankings or visualising patterns on a map.
  • Local authority averages capture broad regional trends, but can sometimes mask variation between individual MSOAs.

In the next case study, we break annual rainfall into spring, summer, autumn, and winter totals to investigate what drove rainfall patterns across the UK during 2024.