Bewerken

Delen via


geo_polygon_to_s2cells()

Applies to: ✅ Microsoft FabricAzure Data ExplorerAzure MonitorMicrosoft Sentinel

Calculates S2 cell tokens that cover a polygon or multipolygon on Earth. This function is a useful geospatial join tool.

Read more about S2 cell hierarchy.

Syntax

geo_polygon_to_s2cells(polygon [, level[, radius]])

Learn more about syntax conventions.

Parameters

Name Type Required Description
polygon dynamic ✔️ Polygon or multipolygon in the GeoJSON format.
level int Defines the requested cell level. Supported values are in the range [0, 30]. If unspecified, the default value 11 is used.
radius real Buffer radius in meters. If unspecified, the default value 0 is used.

Returns

Array of S2 cell token strings that cover a polygon or a multipolygon. If radius is set to a positive value, then the covering will be, in addition to input shape, of all points within the radius of the input geometry. If polygon, level, radius is invalid, or the cell count exceeds the limit, the query will produce a null result.

Note

  • Covering the polygon with S2 cell tokens can be useful in matching coordinates to polygons that might include these coordinates and matching polygons to polygons.
  • The polygon covering tokens are of the same S2 cell level.
  • The maximum count of tokens per polygon is 65536.
  • The geodetic datum used for measurements on Earth is a sphere. Polygon edges are Geodesics on the sphere.
  • If input polygon edges are straight cartesian lines, consider using geo_polygon_densify() in order to convert planar edges to geodesics.

Motivation for covering polygons with S2 cell tokens

Without this function, here's one approach we could take in order to classify coordinates into polygons containing these coordinates.

let Polygons = 
    datatable(description:string, polygon:dynamic)
    [  
      "New York",  dynamic({"type":"Polygon","coordinates":[[[-73.85009765625,40.85744791303121],[-74.16046142578125,40.84290487729676],[-74.190673828125,40.59935608796518],[-73.83087158203125,40.61812224225511],[-73.85009765625,40.85744791303121]]]}),
      "Seattle",   dynamic({"type":"Polygon","coordinates":[[[-122.200927734375,47.68573021131587],[-122.4591064453125,47.68573021131587],[-122.4755859375,47.468949677672484],[-122.17620849609374,47.47266286861342],[-122.200927734375,47.68573021131587]]]}),
      "Las Vegas", dynamic({"type":"Polygon","coordinates":[[[-114.9,36.36],[-115.4498291015625,36.33282808737917],[-115.4498291015625,35.84453450421662],[-114.949951171875,35.902399875143615],[-114.9,36.36]]]}),
    ];
let Coordinates = 
    datatable(longitude:real, latitude:real)
    [
      real(-73.95),  real(40.75), // New York
      real(-122.3),  real(47.6),  // Seattle
      real(-115.18), real(36.16)  // Las Vegas
    ];
Polygons | extend dummy=1
| join kind=inner (Coordinates | extend dummy=1) on dummy
| where geo_point_in_polygon(longitude, latitude, polygon)
| project longitude, latitude, description

Output

longitude latitude description
-73.95 40.75 New York city
-122.3 47.6 Seattle
-115.18 36.16 Las Vegas

While this method works in some cases, it's inefficient. This method does a cross-join, meaning that it tries to match every polygon to every point. This process consumes a large amount of memory and compute resources. Instead, we would like to match every polygon to a point with a high probability of containment success, and filter out other points.

This match can be achieved by the following process:

  1. Converting polygons to S2 cells of level k,
  2. Converting points to the same S2 cells level k,
  3. Joining on S2 cells,
  4. Filtering by geo_point_in_polygon(). This phase can be omitted if some amount of false positives is ok. The maximum error will be the area of s2 cells at level k beyond the boundary of the polygon.

Choosing the S2 cell level

  • Ideally we would want to cover every polygon with one or just a few unique cells such that no two polygons share the same cell.
  • If the polygons are close to each other, choose the S2 cell level such that its cell edge will be smaller (4, 8, 12 times smaller) than the edge of the average polygon.
  • If the polygons are far from each other, choose the S2 cell level such that its cell edge will be similar or bigger than the edge of the average polygon.
  • In practice, covering a polygon with more than 10,000 cells might not yield good performance.
  • Sample use cases:
  • S2 cell level 5 might prove to be good for covering countries/regions.
  • S2 cell level 16 can cover dense and relatively small Manhattan (New York) neighborhoods.
  • S2 cell level 11 can be used for covering suburbs of Australia.
  • Query run time and memory consumption might differ greatly because of different S2 cell level values.

Warning

Covering a large-area polygon with small-area cells can lead to a huge amount of covering cells. As a result, the query might return null.

Note

Performance improvement suggestions:

  • If possible, reduce coordinates table size before join, by grouping coordinates that are very close to each other by using geospatial clustering or by filtering out unnesessary coordinates due to nature of the data or business needs.
  • If possible, reduce polygons count due to nature of the data or business needs. Filter out unnecessary polygons before join, scope to the area of interest or unify polygons.
  • In case of very big polygons, reduce their size using geo_polygon_simplify().
  • Changing S2 cell level may improve performance and memory consumption.
  • Changing join kind and hint may improve performance and memory consumption.
  • If a positive radius is set, you can try to improve performance by reverting to radius 0 on buffered shape using geo_polygon_buffer().

Examples

The following example classifies coordinates into polygons.

let Polygons = 
    datatable(description:string, polygon:dynamic)
    [
        'Greenwich Village', dynamic({"type":"Polygon","coordinates":[[[-73.991460000000131,40.731738000000206],[-73.992854491775518,40.730082566051351],[-73.996772,40.725432000000154],[-73.997634685522883,40.725786309886963],[-74.002855946639244,40.728346630056791],[-74.001413,40.731065000000207],[-73.996796995070824,40.73736378205173],[-73.991724524037934,40.735245208931886],[-73.990703782359589,40.734781896080477],[-73.991460000000131,40.731738000000206]]]}),
        'Upper West Side',   dynamic({"type":"Polygon","coordinates":[[[-73.958357552055688,40.800369095633819],[-73.98143901556422,40.768762584141953],[-73.981548752788598,40.7685590292784],[-73.981565335901905,40.768307084720796],[-73.981754418060945,40.768399727738668],[-73.982038573548124,40.768387823012056],[-73.982268248204349,40.768298621883247],[-73.982384797518051,40.768097213086911],[-73.982320919746599,40.767894461792181],[-73.982155532845766,40.767756204474757],[-73.98238873834039,40.767411004834273],[-73.993650353659021,40.772145571634361],[-73.99415893763998,40.772493009137818],[-73.993831082030937,40.772931787850908],[-73.993891252437052,40.772955194876722],[-73.993962585514595,40.772944653908901],[-73.99401262480508,40.772882846631894],[-73.994122058082397,40.77292405902601],[-73.994136652588594,40.772901870174394],[-73.994301342391154,40.772970028663913],[-73.994281535134448,40.77299380206933],[-73.994376552751078,40.77303955110149],[-73.994294029824005,40.773156243992048],[-73.995023275860802,40.773481196576356],[-73.99508939189289,40.773388475039134],[-73.995013963716758,40.773358035426909],[-73.995050284699261,40.773297153189958],[-73.996240651898916,40.773789791397689],[-73.996195837470992,40.773852356184044],[-73.996098807369748,40.773951805299085],[-73.996179459973888,40.773986954351571],[-73.996095245226442,40.774086186437756],[-73.995572265161172,40.773870731394297],[-73.994017424135961,40.77321375261053],[-73.993935876811335,40.773179512586211],[-73.993861942928888,40.773269531698837],[-73.993822393527211,40.773381758622882],[-73.993767019318497,40.773483981224835],[-73.993698463744295,40.773562141052594],[-73.993358326468751,40.773926888327956],[-73.992622663865575,40.774974056037109],[-73.992577842766124,40.774956016359418],[-73.992527743951555,40.775002110439829],[-73.992469745815342,40.775024159551755],[-73.992403837191887,40.775018140390664],[-73.99226708903538,40.775116033858794],[-73.99217809026365,40.775279293897171],[-73.992059084937338,40.775497598192516],[-73.992125372394938,40.775509075053385],[-73.992226867797001,40.775482211026116],[-73.992329346608813,40.775468900958522],[-73.992361756801131,40.775501899766638],[-73.992386042960277,40.775557180424634],[-73.992087684712729,40.775983970821372],[-73.990927174149746,40.777566878763238],[-73.99039616003671,40.777585065679204],[-73.989461267506471,40.778875124584417],[-73.989175778438053,40.779287524015778],[-73.988868617400072,40.779692922911607],[-73.988871874499793,40.779713738253008],[-73.989219022880576,40.779697895209402],[-73.98927785904425,40.779723439271038],[-73.989409054180143,40.779737706471963],[-73.989498614927044,40.779725044389757],[-73.989596493388234,40.779698146683387],[-73.989679812902509,40.779677568658038],[-73.989752702937935,40.779671244211556],[-73.989842247806507,40.779680752670664],[-73.990040102120489,40.779707677698219],[-73.990137977524839,40.779699769704784],[-73.99033584033225,40.779661794394983],[-73.990430598697046,40.779664973055503],[-73.990622199396725,40.779676064914298],[-73.990745069505479,40.779671328184051],[-73.990872114282197,40.779646007643876],[-73.990961672224358,40.779639683751753],[-73.991057472829539,40.779652352625774],[-73.991157429497036,40.779669775606465],[-73.991242817404469,40.779671367084504],[-73.991255318289745,40.779650782516491],[-73.991294887120119,40.779630209208889],[-73.991321967649895,40.779631796041372],[-73.991359455569423,40.779585883337383],[-73.991551059227476,40.779574821437407],[-73.99141982585985,40.779755280287233],[-73.988886144117032,40.779878898532999],[-73.988939656706265,40.779956178440393],[-73.988926103530844,40.780059292013632],[-73.988911680264692,40.780096037146606],[-73.988919261468567,40.780226094343945],[-73.988381050202634,40.780981074045783],[-73.988232413846987,40.781233144215555],[-73.988210420831663,40.781225482542055],[-73.988140000000143,40.781409000000224],[-73.988041288067166,40.781585961353777],[-73.98810029382463,40.781602878305286],[-73.988076449145055,40.781650935001608],[-73.988018059972219,40.781634188810422],[-73.987960792842145,40.781770987031535],[-73.985465811970457,40.785360700575431],[-73.986172704965611,40.786068452258647],[-73.986455862401996,40.785919219081421],[-73.987072345615601,40.785189638820121],[-73.98711901394276,40.785210319004058],[-73.986497781023601,40.785951202887254],[-73.986164628806279,40.786121882448327],[-73.986128422486075,40.786239001331111],[-73.986071135219746,40.786240706026611],[-73.986027274789123,40.786228964236727],[-73.986097637849426,40.78605822569795],[-73.985429321269592,40.785413942184597],[-73.985081137732209,40.785921935110366],[-73.985198833254501,40.785966552197777],[-73.985170502389906,40.78601333415817],[-73.985216218673656,40.786030501816427],[-73.98525509797993,40.785976205511588],[-73.98524273937646,40.785972572653328],[-73.98524962933017,40.785963139855845],[-73.985281779186749,40.785978620950075],[-73.985240032884533,40.786035858136792],[-73.985683885242182,40.786222123919686],[-73.985717529004575,40.786175994668795],[-73.985765660297687,40.786196274858618],[-73.985682871922691,40.786309786213067],[-73.985636270930442,40.786290150649279],[-73.985670722564691,40.786242911993817],[-73.98520511880038,40.786047669212785],[-73.985211035607492,40.786039554883686],[-73.985162639946992,40.786020999769754],[-73.985131636312062,40.786060297019972],[-73.985016964065125,40.78601423719563],[-73.984655078830457,40.786534741807841],[-73.985743787901043,40.786570082854738],[-73.98589227228328,40.786426529019593],[-73.985942854994988,40.786452847880334],[-73.985949561556794,40.78648711396653],[-73.985812373526713,40.786616865357047],[-73.985135209703174,40.78658761889551],[-73.984619428584324,40.786586016349787],[-73.981952458164173,40.790393724337193],[-73.972823037363767,40.803428052816756],[-73.971036786332192,40.805918478839672],[-73.966701,40.804169000000186],[-73.959647,40.801156000000113],[-73.958508540159471,40.800682279767472],[-73.95853274080838,40.800491362464697],[-73.958357552055688,40.800369095633819]]]}),
        'Upper East Side',   dynamic({"type":"Polygon","coordinates":[[[-73.943592454622546,40.782747908206574],[-73.943648235390199,40.782656161333449],[-73.943870759887162,40.781273026571704],[-73.94345932494096,40.780048275653243],[-73.943213862652243,40.779317588660199],[-73.943004239504688,40.779639495474292],[-73.942716005450905,40.779544169476175],[-73.942712374762181,40.779214856940001],[-73.942535563208608,40.779090956062532],[-73.942893408188027,40.778614093246276],[-73.942438481745029,40.777315235766039],[-73.942244919522594,40.777104088947254],[-73.942074188038887,40.776917846977142],[-73.942002667222781,40.776185317382648],[-73.942620205199006,40.775180871576474],[-73.94285645694552,40.774796600349191],[-73.94293043781397,40.774676268036011],[-73.945870899588215,40.771692257932997],[-73.946618690150586,40.77093339256956],[-73.948664164778933,40.768857624399587],[-73.950069793030679,40.767025088383498],[-73.954418260786071,40.762184104951245],[-73.95650786241211,40.760285256574043],[-73.958787773424007,40.758213471309809],[-73.973015157270069,40.764278692864671],[-73.955760332998182,40.787906554459667],[-73.944023,40.782960000000301],[-73.943592454622546,40.782747908206574]]]}),
    ];
let Coordinates = 
    datatable(longitude:real, latitude:real)
    [
        real(-73.9741), 40.7914, // Upper West Side
        real(-73.9950), 40.7340, // Greenwich Village
        real(-73.9584), 40.7688, // Upper East Side
    ];
let Level = 16;
Polygons
| extend covering = geo_polygon_to_s2cells(polygon, Level) // cover every polygon with s2 cell token array
| mv-expand covering to typeof(string)                     // expand cells array such that every row will have one cell mapped to its polygon
| join kind=inner hint.strategy=broadcast                  // assume that Polygons count is small (In some specific case)
(
    Coordinates
    | extend covering = geo_point_to_s2cell(longitude, latitude, Level) // cover point with cell
) on covering // join on the cell, this filters out rows of point and polygons where the point definitely does not belong to the polygon
| where geo_point_in_polygon(longitude, latitude, polygon) // final filtering for exact result
| project longitude, latitude, description

Output

longitude latitude description
-73.9741 40.7914 Upper West Side
-73.995 40.734 Greenwich Village
-73.9584 40.7688 Upper East Side

Here is even more improvement on the above query. Count storm events per US state. The below query performs a very efficient join because it doesn't carry polygons through the join and uses lookup operator

let Level = 6;
let polygons = materialize(
    US_States
    | project StateName = tostring(features.properties.NAME), polygon = features.geometry, id = new_guid());
let tmp = 
    polygons
    | project id, covering = geo_polygon_to_s2cells(polygon, Level) 
    | mv-expand covering to typeof(string)
    | join kind=inner hint.strategy=broadcast
            (
                StormEvents
                | project lng = BeginLon, lat = BeginLat
                | project lng, lat, covering = geo_point_to_s2cell(lng, lat, Level)
            ) on covering
    | project-away covering, covering1;
tmp | lookup polygons on id
| project-away id
| where geo_point_in_polygon(lng, lat, polygon)
| summarize StormEventsCountByState = count() by StateName

Output

StateName StormEventsCountByState
Florida 960
Georgia 1085
... ...

The following example filters out polygons that don't intersect with the area of the polygon of interest. The maximum error is diagonal of s2cell length. This example is based on a polygonized earth at night raster file.

let intersection_level_hint = 7;
let area_of_interest = dynamic({"type": "Polygon","coordinates": [[[-73.94966125488281,40.79698248639272],[-73.95841598510742,40.800426144169315],[-73.98124694824219,40.76806170936614],[-73.97283554077148,40.7645513650551],[-73.94966125488281,40.79698248639272]]]});
let area_of_interest_covering = geo_polygon_to_s2cells(area_of_interest, intersection_level_hint);
EarthAtNight
| project value = features.properties.DN, polygon = features.geometry
| extend covering = geo_polygon_to_s2cells(polygon, intersection_level_hint)
| mv-apply c = covering to typeof(string) on
(
    summarize is_intersects = take_anyif(1, array_index_of(area_of_interest_covering, c) != -1)
)
| where is_intersects == 1
| count

Output

Count
83

Count of cells that will be needed in order to cover some polygon with S2 cells of level 5.

let polygon = dynamic({"type":"Polygon","coordinates":[[[0,0],[0,50],[100,50],[0,0]]]});
print s2_cell_token_count = array_length(geo_polygon_to_s2cells(polygon, 5));

Output

s2_cell_token_count
286

Covering a large-area polygon with small-area cells returns null.

let polygon = dynamic({"type":"Polygon","coordinates":[[[0,0],[0,50],[100,50],[0,0]]]});
print geo_polygon_to_s2cells(polygon, 30);

Output

print_0

Covering a large-area polygon with small-area cells returns null.

let polygon = dynamic({"type":"Polygon","coordinates":[[[0,0],[0,50],[100,50],[0,0]]]});
print isnull(geo_polygon_to_s2cells(polygon, 30));

Output

print_0
1