打造数据驱动的城市：利用 Python 整合住房、治安与社会剥夺指标

import pgeocode
import pandas as pd

# 初始化英国的邮编查询对象
nomi = pgeocode.Nominatim("GB")

def batch_geocode(df, postcode_col):
    # 清洗邮编格式
    clean_postcodes = df[postcode_col].astype(str).str.strip()
    unique_pcs = clean_postcodes.unique()
    
    # 批量查询，提高效率
    geo_results = nomi.query_postal_code(unique_pcs)
    
    # 建立映射字典
    pc_to_lat = dict(zip(unique_pcs, geo_results.latitude))
    pc_to_lon = dict(zip(unique_pcs, geo_results.longitude))
    
    df["lat"] = clean_postcodes.map(pc_to_lat)
    df["lon"] = clean_postcodes.map(pc_to_lon)
    
    return df.dropna(subset=["lat", "lon"])

from pathlib import Path

# 递归查找所有 CSV 文件
all_csvs = list(Path("police_data").rglob("*.csv"))
stop_search_points = []

for csv_file in all_csvs:
    try:
        # 只读取必要的经纬度列，节省内存
        df = pd.read_csv(csv_file, usecols=["Latitude", "Longitude"])
        df = df.dropna()
        # 粗略过滤掉不在利物浦范围内的点，加速后续处理
        mask = (df['Latitude'].between(53.0, 54.0)) & \
               (df['Longitude'].between(-3.5, -2.0))
        stop_search_points.append(df[mask])
    except:
        continue

# 合并所有数据帧
all_stops = pd.concat(stop_search_points, ignore_index=True)

import geopandas as gpd

# 将警方数据转换为 GeoDataFrame
gdf_stops = gpd.GeoDataFrame(
    all_stops, 
    geometry=gpd.points_from_xy(all_stops.Longitude, all_stops.Latitude),
    crs="EPSG:4326"
)

# 核心操作：判断点在哪个多边形内 (predicate='within')
# lsoa_lcr 是我们的行政区划多边形数据
stops_with_lsoa = gpd.sjoin(gdf_stops, lsoa_lcr[['LSOA21CD', 'geometry']], predicate='within')

# 聚合统计：计算每个 LSOA 的犯罪数量
stop_counts = stops_with_lsoa['LSOA21CD'].value_counts().reset_index()
stop_counts.columns = ['LSOA21CD', 'StopSearch_Count']

# 归一化燃料贫困率
f_min, f_max = master_gdf["Fuel_Poverty_Rate"].min(), master_gdf["Fuel_Poverty_Rate"].max()
master_gdf["norm_fuel"] = (master_gdf["Fuel_Poverty_Rate"] - f_min) / (f_max - f_min)

# 归一化 IMD (反转，10为最不贫困，1为最贫困 -> 映射后 1.0 为高风险)
master_gdf["norm_imd"] = (11 - master_gdf["IMD_Decile"]) / 10.0

# 归一化犯罪数据 (使用 95 分位值作为最大值，防止极端异常值拉低整体区分度)
c_max = master_gdf["Crime_Count"].quantile(0.95)
master_gdf["norm_crime"] = (master_gdf["Crime_Count"] / c_max).clip(upper=1.0)

import folium

m = folium.Map(location=[centre_lat, centre_lon], zoom_start=11, tiles="CartoDB positron")

# 创建自定义图层面板
folium.map.CustomPane("poly_pane", z_index=400).add_to(m)
folium.map.CustomPane("housing_pane", z_index=620).add_to(m)
folium.map.CustomPane("brownfield_pane", z_index=650).add_to(m)

from branca.element import Element

# 将 Python 处理好的数据转为 JSON 字符串注入前端
geojson_str = master_gdf.to_json()

custom_ui = f"""
<div id="controls">
    <h4>Layer Weights (LSOA Scoring)</h4>
    <!-- 滑块控件 -->
    <div><label>Fuel Poverty</label><input type="range" id="w_fuel" oninput="updateWeights()"></div>
    <div><label>Crime Density</label><input type="range" id="w_crime" oninput="updateWeights()"></div>
    <!-- ... 其他滑块 ... -->
</div>

<script>
    var lsoaData = {geojson_str}; // 注入数据
    
    // 动态计算分数的函数
    function updateWeights() {{
        // 获取滑块的值
        var w_fuel = parseFloat(document.getElementById('w_fuel').value);
        var w_crime = parseFloat(document.getElementById('w_crime').value);
        
        // 遍历每个地理单元重新计算颜色
        geojsonLayer.eachLayer(function(layer) {{
            var p = layer.feature.properties;
            // 加权平均公式
            var score = (p.norm_fuel * w_fuel + p.norm_crime * w_crime) / (w_fuel + w_crime);
            
            // 动态设置样式
            layer.setStyle({{ fillColor: getColor(score) }});
        }});
    }}
</script>
"""
m.get_root().html.add_child(Element(custom_ui))