geocoding/flagging iso2 centroid locations on GBIF

[jhnwllr]

background

Sometimes data publishers will not know the exact lat-lon location of a record and will enter the lat-long center of the locality instead. This is a data issue because users might be unaware that an observation is pinned to a locality center and assume it is a precise location.

From previous work it is known that most centroids are coming from museum collections (basisOfRecord=PRESERVED_SPECIMEN).

false positives problem

@MattBlissett has pointed out that in some cases, if we were to flag records, we would end up flagging many "non-centroid" false positives.

In the figure, the UK centroid where many non-centroid human observations are mixed with fewer "real centroids" from likely retrospectively geo-coded records. Many museum records are sitting directly on the centroid, but as a user you are probably also concerned with the few museum records somewhat further away from the centroid. (rings are 2km and 5km buffers)

Publishers would probably not like to have many records flagged that just happen to be near centroids.

Of course, for some centroids this isn't a problem at all.

users vs publishers

Users who want to filter for centroid locations, are more interested in making sure no outliers make it into their models, than false positives. So most users would rather over flag centroids.

Publishers would rather we be more judicious with flagging, so their datasets don't get littered with false positives.

This is why I recommend treating centroids more neutrally as a geocoded location rather than a data quality flag.

data quality flag vs location filter

Fake potential UI below.

In my view, thinking about centroids as useful, interesting neutral locations rather than as data quality problems/flags makes centroids easier to work with. Since we are never going to eliminate all false positives, it makes more sense to treat centroids as locations. This becomes even more apparent when we start talking about province and state centroids, which are also useful locations, but will produce even more "false positives".

There are some really small provinces but it would still be useful to geocode the centroids.

One disadvantage to treating centroids as simply geocoded locations would be that we might need to include an additional column in downloads to make it useful for users. Also it is difficult to filter out unwanted records with the current interface.

Below you can review 30 sampled centroids for iso2 places over 30K sqkm. It is usually impossible to tell if a point on a centroid is a "real" centroid but usually if a preserved specimen is somewhat close to known centroid, it is highly likely to be a "real centroid".

iso2	name	n_preserved_specimen	n_human_observation	source	centroid_2km
AO	Angola	154	0	geolocate	link%20)
AU	Australia	17	2	CoordinateCleaner	link%20)
BE	Belgium	1009	12011	geolocate	link%20)
CM	Cameroon	20	0	geolocate	link%20)
CA	Canada	1326	0	geolocate	link%20)
CN	China	1979	0	geolocate	link%20)
CI	Cote d'Ivoire	368	0	CoordinateCleaner	link%20)
GL	Greenland	122	0	CoordinateCleaner	link%20)
GT	Guatemala	1287	80	geolocate	link%20)
HU	Hungary	54	136	CoordinateCleaner	link%20)
IE	Ireland	94	158	geolocate	link%20)
IT	Italy	2005	1	geolocate	link%20)
LR	Liberia	278	0	geolocate	link%20)
MG	Madagascar	2024	0	geolocate	link%20)
MR	Mauritania	3	0	geolocate	link%20)
MA	Morocco	1	0	CoordinateCleaner	link%20)
NL	Netherlands	476	22423	geolocate	link%20)
NZ	New Zealand	24	225	CoordinateCleaner	link%20)
NI	Nicaragua	152	128	geolocate	link%20)
NE	Niger	68	0	CoordinateCleaner	link%20)
NG	Nigeria	64	74	CoordinateCleaner	link%20)
SA	Saudi Arabia	0	0	CoordinateCleaner	link%20)
RS	Serbia	92	5	CoordinateCleaner	link%20)
SE	Sweden	45	866	CoordinateCleaner	link%20)
CH	Switzerland	167	72	CoordinateCleaner	link%20)
TW	Taiwan	1694	2099	CoordinateCleaner	link%20)
AE	United Arab Emirates	5	0	CoordinateCleaner	link%20)
GB	United Kingdom	76	1134	CoordinateCleaner	link%20)
VN	Vietnam	304	4	geolocate	link%20)
ZW	Zimbabwe	5	29	geolocate	link%20)

This brain dump are my current thoughts. Open to any divergent opinions or discussion.

@timrobertson100 @ahahn-gbif @MattBlissett

[jhnwllr]

@MortenHofft has raised the point that the data quality flag for publishers could be separated from the neutrally geocoded records. I agree with this perspective.

The data quality flag could have the following properties:

• Small buffer around the centroid ~500m
• Only flag preserved specimens
• Only flag large countries

[jhnwllr]

We probably want the action/menu to be "exclude within 2/5/10km of centroid".

[tucotuco]

@jhnwllr This is great work. However, I want to challenge a premise.

Users who want to filter for centroid locations, are more interested in making sure no outliers make it into their models, than false positives. So most users would rather over flag centroids.

Scientists who do not want inappropriate data in their models need to verify those data before using them. Without the explicit measure of uncertainty they can't do that, centroid or not, outlier or not. That doesn't mean there is anything wrong with your attempts to help highlight records that require review before use, but there is a MUCH simpler and all-encompassing test for that. Does the location have either a geospatial footprint (an actual geometry in the data) or uncertainty for the coordinates (as a distance)? Every record that gets used in modeling should have one of those. If the records were properly georeferenced, you wouldn't have to worry about geocoding flagging you would already have it in coordinateUncertaintyInMeters. Another focus of effort could be to provide the coordinates and uncertainty for records that do not have coordinates, but do have unambiguous administrative geography, to the highest specificity you can.

Independently, keep in mind that there are also many different kinds of centroids (e.g., Australia has at least five), and your proposal covers just one of those.

[ArthurChapman]

There are a number of issues with (country) centroids not discussed above. I agree largely with what @tucotuco has said.

However, if we take Australia. Many early records for Australia just say "Nova Hollandia" or "Australiia" and a default has been placed at the "center" of Australia [but see below]. When these observations were made, the only European discovery was around the Australian coast - and definitely not in the desert in central Australia. A better representation would be a footprint derived by buffering the coast (or in many cases, just the east coast around Sydney/Botany Bay, etc.) One ends up with a record as the centroid which is in the Australian desert for an observation from the wet coast.

Just to illustrate this point, there is a record of a marine starfish on GBIF at -25.274, 133.775 (middle of Australia's desert) - which in itself is an interesting centroid that I think comes from the CIA database and for which I have not worked out the scientific basis for.

The determination of the centroid of a country can be carried out using a number of methods - none are wrong, just determined using different methods. In the case of Australia, this can result in differences of hundreds of kilometers.

Nova Hollandia" or "Australia" on ALA
-25.274, 133.775
(24 records on ALA within 1 km)

Centre of Gravity Method
23° 07' South, 132° 08' (-23.1166667, 132.33333)
(0 records on ALA within 1 km)

Lambert Gravitational Centre
25° 36' 36.4"S, 134° 21' 17.3"E (-25.610111, 134.3548056)
(Lots of records within 1km - also it comes up on the map on ALA about 300 m from where Geosciences Australia places it!) – Datum problem?

Furthest Point from Coastline
23° 02'S, 132° 10'E (-23.033333, 132.1666667)
(0 records on ALA within 1 km)

Geodetic Median Point
23° 33' 09.89"S, 133° 23' 46.00"E (-23.5527472, 133.396111)
(8 records on ALA within 1 km)

Johnston Geodetic Station
25° 56' 49.3"S, 133° 12' 34.7"E (-25.9470278, 133.2096389)
(lots of records on ALA within 1 km)

Notethat a lot of records from both Lamberts Centre and from Johnson's Geodetic Centre are actual observations from those places (I have made recent observations there myself with a determined uncertainty) - i.e. high precision records. Many, however, are defaults of "Australia"

So, in summary, there are several issues.

What is "Australia"

1. Mainland Australia
2. Mainland Australia + Tasmania
3. Mainland Australia + Tasmania + main islands (Lord Howe, Norfolk, Christmas
4. EEZ
5. Continental Shelf
6. Australia + territories including Australia Antarctica Area.

How was the center (centroid) determined

1. Centre of Gravity Method
2. Lambert Gravitational Centre
3. Furthest Point from Coastline
4. Geodetic Median Point
5. Johnston Geodetic Station

There are similar issues with the Australian States and determining the centroids there.

[MortenHofft]

Scientists who do not want inappropriate data in their models need to verify those data before using them. Without the explicit measure of uncertainty they can't do that, centroid or not, outlier or not

Records without an uncertainty stated: 1,373,034,036 (out of 2,104,374,546 with coordinates)

A flag that the record is missing uncertainty might be useful in that case since ≈2/3 are missing it.

[tucotuco]

I think it would be useful, but also make sure not to flag it if there is a footprintWKT provided that is not just a POINT.

[jhnwllr]

I have taken the time to extract centroids for iso2 places from the Getty Thesaurus of Geographic Names (TGN). I have pasted them all here for reference.

The TGN is a source of many centroids on GBIF.

Interestingly, we are still missing the centroids that @ArthurChapman points to for Australia. Extracting these centroids is already a bit dodgy, so source="Arthur Chapman" might a good solution for AU centroids.

The point about dwc:coordinateUncertaintyInMeters that @tucotuco raises is totally correct. But I think we can still geocode centroids while also encouraging publishers to fill in this very important field.

iso2	tgn_name	lat	lon	n_specimen	n_hobservation	source	centroid_2km
MX	Mexico	23	-102	18648	16	TGN	link%20)
PH	Philippines	13	122	7604	6	TGN	link%20)
SG	Singapore	1.3667	103.8	7026	35108	TGN	link%20)
MU	Mauritius	-20.3	57.5833	5947	28	TGN	link%20)
NF	Norfolk Island	-29.033333	167.95	5553	6872	TGN	link%20)
SE	Sweden	62	15	5000	255	TGN	link%20)
CU	Cuba	21.5	-80	4748	0	TGN	link%20)
CH	Switzerland	47	8	4743	23677	TGN	link%20)
FR	France	46	2	4573	1984	TGN	link%20)
JP	Japan	36	138	4313	81	TGN	link%20)
GY	Guyana	5	-59	4180	0	TGN	link%20)
DE	Germany	51.5	10.5	3769	18	TGN	link%20)
FJ	Fiji	-18	178	3162	4	TGN	link%20)
NC	New Caledonia	-21.5	165.5	3101	9	TGN	link%20)
BR	Brazil	-10	-55	2929	2	TGN	link%20)
AU	Australia	-25	135	2874	9	TGN	link%20)
BM	Bermuda	32.3333	-64.75	2676	9249	TGN	link%20)
CX	Christmas Island	-10.5	105.6667	2645	226	TGN	link%20)
DK	Denmark	56	10	2633	4452	TGN	link%20)
CD	Democratic Republic of the Congo	-.0167	25	2585	0	TGN	link%20)
CR	Costa Rica	10	-84	2495	1154	TGN	link%20)
US	United States	38	-98	2370	3714	TGN	link%20)
BS	Bahamas	24	-76	2276	0	TGN	link%20)
JM	Jamaica	18.25	-77.5	2206	24	TGN	link%20)
BN	Brunei Darussalam	4.5	114.6667	2130	0	TGN	link%20)
PA	Panama	9	-80	2037	0	TGN	link%20)
MG	Madagascar	-20	47	2025	0	TGN	link%20)
LU	Luxembourg	49.75	6.1667	2019	4737	TGN	link%20)
IT	Italy	42.8333	12.8333	2005	1	TGN	link%20)
CN	China	35	105	1978	0	TGN	link%20)
FO	Faeroe Islands	62	-7	1939	81	TGN	link%20)
AW	Aruba	12.5	-69.9667	1837	3045	TGN	link%20)
CW	Curacao	12.166	-69	1799	11517	TGN	link%20)
NZ	New Zealand	-42	174	1685	910	TGN	link%20)
PG	Papua New Guinea	-6	147	1636	0	TGN	link%20)
HU	Hungary	47	20	1623	123	TGN	link%20)
LK	Sri Lanka	7	81	1620	7	TGN	link%20)
ZA	South Africa	-30	26	1579	0	TGN	link%20)
BB	Barbados	13.1667	-59.5333	1559	1598	TGN	link%20)
SB	Solomon Islands	-8	159	1500	0	TGN	link%20)
TN	Tunisia	34	9	1397	4	TGN	link%20)
DO	Dominican Republic	19	-70.6667	1338	0	TGN	link%20)
CA	Canada	60	-96	1326	0	TGN	link%20)
AD	Andorra	42.55	1.583	1308	3238	TGN	link%20)
GT	Guatemala	15.5	-90.25	1287	80	TGN	link%20)
MQ	Martinique	14.6667	-61	1284	103	TGN	link%20)
EC	Ecuador	-2	-77.5	1256	0	TGN	link%20)
AT	Austria	47.3333	13.3333	1206	324	TGN	link%20)
CZ	Czech Republic	49.75	15	1206	38	TGN	link%20)
GP	Guadeloupe	16.25	-61.5833	1203	3547	TGN	link%20)
BZ	Belize	17.25	-88.75	1201	18302	TGN	link%20)
IN	India	20	77	1190	2	TGN	link%20)
MS	Montserrat	16.75	-62.2	1141	3798	TGN	link%20)
TZ	Tanzania	-6	35	1138	663	TGN	link%20)
KE	Kenya	1	38	1132	11	TGN	link%20)
FI	Finland	64	26	1092	525	TGN	link%20)
RU	Russia	60	47	1070	0	TGN	link%20)
HK	Hong Kong	22.25	114.1667	1066	6839	TGN	link%20)
BE	Belgium	50.8333	4	1009	12011	TGN	link%20)
SC	Seychelles	-4.5833	55.6667	977	4	TGN	link%20)
CO	Colombia	4	-72	963	2	TGN	link%20)
ID	Indonesia	-5	120	952	0	TGN	link%20)
RE	Reunion	-21.1	55.6	927	2401	TGN	link%20)
CL	Chile	-30	-71	890	4852	TGN	link%20)
GL	Greenland	72	-40	865	0	TGN	link%20)
PE	Peru	-10	-76	857	3	TGN	link%20)
TT	Trinidad and Tobago	11	-61	850	0	TGN	link%20)
SN	Senegal	14	-14	806	12	TGN	link%20)
IS	Iceland	65	-18	772	189	TGN	link%20)
HT	Haiti	19	-72.4167	761	0	TGN	link%20)
GD	Grenada	12.1167	-61.6667	757	701	TGN	link%20)
ES	Spain	40	-4	753	184	TGN	link%20)
TH	Thailand	15	100	751	23	TGN	link%20)
KN	Saint Kitts and Nevis	17.3333	-62.75	743	400	TGN	link%20)
MT	Malta	35.9167	14.4167	731	1603	TGN	link%20)
PY	Paraguay	-23	-58	721	0	TGN	link%20)
PN	Pitcairn Islands	-25.0667	-130.1	721	105	TGN	link%20)
EE	Estonia	59	26	720	1147	TGN	link%20)
MC	Monaco	43.7333	7.4167	713	4454	TGN	link%20)
JE	Jersey	49.2167	-2.1167	696	2053	TGN	link%20)
DZ	Algeria	28	3	674	0	TGN	link%20)
NO	Norway	62	10	620	535	TGN	link%20)
GR	Greece	39	22	588	19	TGN	link%20)
RO	Romania	46	25	586	136	TGN	link%20)
MN	Mongolia	46	105	576	1	TGN	link%20)
SR	Suriname	4	-56	566	0	TGN	link%20)
DM	Dominica	15.5	-61.3333	558	3	TGN	link%20)
TW	Taiwan	24	121	547	20602	TGN	link%20)
CV	Cape Verde	16	-24	542	0	TGN	link%20)
CM	Cameroon	6	12	539	0	TGN	link%20)
VE	Venezuela	8	-66	523	0	TGN	link%20)
SK	Slovakia	48.6667	19.5	507	3	TGN	link%20)
EG	Egypt	27	30	499	0	TGN	link%20)
LI	Liechtenstein	47.1667	9.5333	496	8749	TGN	link%20)
GI	Gibraltar	36.1333	-5.35	469	8711	TGN	link%20)
BL	Saint BarthÚlemy	17.9	-62.833	459	5430	TGN	link%20)
AG	Antigua and Barbuda	17.05	-61.8	444	138	TGN	link%20)
CI	C¶te d'Ivoire	8	-5	423	0	TGN	link%20)
MP	Northern Mariana Islands	15.213	145.755	422	1290	TGN	link%20)
SH	Saint Helena, Ascension and Tristan da Cunha	-15.95	-5.7	395	82	TGN	link%20)
PT	Portugal	39.5	-8	391	2132	TGN	link%20)
YT	Mayotte	-12.8333	45.1667	386	309	TGN	link%20)
PL	Poland	52	20	380	242	TGN	link%20)
MH	Marshall Islands	10	167	380	0	TGN	link%20)
AM	Armenia	40	45	373	9	TGN	link%20)
BO	Bolivia	-17	-65	367	1	TGN	link%20)
AI	Anguilla	18.2167	-63.05	361	1668	TGN	link%20)
IL	Israel	31.5	34.75	359	125	TGN	link%20)
SY	Syria	35	38	359	0	TGN	link%20)
BW	Botswana	-22	24	348	0	TGN	link%20)
MW	Malawi	-13.5	34	346	0	TGN	link%20)
KR	South Korea	37	127.5	341	90	TGN	link%20)
AR	Argentina	-34	-64	336	10	TGN	link%20)
MZ	Mozambique	-18.25	35	330	0	TGN	link%20)
LC	Saint Lucia	13.8833	-60.9667	330	190	TGN	link%20)
MM	Myanmar	22	98	326	0	TGN	link%20)
ZW	Zimbabwe	-19	29	288	0	TGN	link%20)
SL	Sierra Leone	8.5	-11.5	286	0	TGN	link%20)
VN	Viet Nam	16	106	280	0	TGN	link%20)
LR	Liberia	6.5	-9.5	278	0	TGN	link%20)
GU	Guam	13.4667	144.8333	268	385	TGN	link%20)
KM	Comoros	-12.1667	44.25	255	1	TGN	link%20)
UG	Uganda	2	33	253	1	TGN	link%20)
KY	Cayman Islands	19.5	-80.6667	245	0	TGN	link%20)
VC	Saint Vincent and the Grenadines	13.0833	-61.2	243	10	TGN	link%20)
NR	Nauru	-.5333	166.9167	243	1003	TGN	link%20)
IR	Iran	32	53	240	53	TGN	link%20)
GA	Gabon	-1	11.75	225	2	TGN	link%20)
GE	Georgia	42	43.5	223	32	TGN	link%20)
UY	Uruguay	-33	-56	222	0	TGN	link%20)
NL	Netherlands	52.5	5.75	219	1808	TGN	link%20)
MY	Malaysia	2.5	112.5	217	0	TGN	link%20)
VU	Vanuatu	-16	167	212	1	TGN	link%20)
VG	British Virgin Islands	18.5	-64.5	212	1	TGN	link%20)
TR	Turkey	39	35	210	7	TGN	link%20)
NG	Nigeria	10	8	210	0	TGN	link%20)
NI	Nicaragua	13	-85	205	0	TGN	link%20)
PW	Palau	6	134	200	13	TGN	link%20)
NP	Nepal	28	84	198	0	TGN	link%20)
BJ	Benin	9.5	2.25	197	33	TGN	link%20)
PK	Pakistan	30	70	192	0	TGN	link%20)
BD	Bangladesh	24	90	191	2	TGN	link%20)
HR	Croatia	45.1667	15.5	191	4	TGN	link%20)
HN	Honduras	15	-86.5	188	0	TGN	link%20)
NA	Namibia	-22	17	187	7	TGN	link%20)
YE	Yemen	15.5	47.5	174	0	TGN	link%20)
WF	Wallis and Futuna Islands	-13.3	-176.2	168	84	TGN	link%20)
MA	Morocco	32	-5	167	8	TGN	link%20)
GH	Ghana	8	-2	163	6	TGN	link%20)
SA	Saudi Arabia	25	45	162	0	TGN	link%20)
AO	Angola	-12.5	18.5	154	0	TGN	link%20)
TO	Tonga	-20	-175	150	19	TGN	link%20)
CY	Cyprus	35	33	133	124	TGN	link%20)
KH	Cambodia	13	105	127	0	TGN	link%20)
SD	Sudan	16	30	126	0	TGN	link%20)
ET	Ethiopia	8	39	122	0	TGN	link%20)
VA	Holy See	41.903	12.453	119	9301	TGN	link%20)
GF	French Guiana	4	-53	116	365	TGN	link%20)
IQ	Iraq	33	44	115	0	TGN	link%20)
NU	Niue	-19.0333	-169.8667	114	11	TGN	link%20)
MO	Macau	22.1667	113.55	106	2072	TGN	link%20)
KP	North Korea	40	127	105	0	TGN	link%20)
PF	French Polynesia	-15	-140	105	0	TGN	link%20)
GM	The Gambia	13.5	-15.5	104	45	TGN	link%20)
OM	Oman	21	57	102	0	TGN	link%20)
UA	Ukraine	49	32	102	0	TGN	link%20)
IE	Ireland	53	-8	94	158	TGN	link%20)
GN	Guinea	11	-10	87	0	TGN	link%20)
ZM	Zambia	-15	30	85	0	TGN	link%20)
FK	Falkland Islands	-51.75	-59	83	0	TGN	link%20)
MV	Maldives	3.2	73	77	0	TGN	link%20)
BA	Bosnia and Herzegovina	44.25	17.8333	73	0	TGN	link%20)
TC	Turks and Caicos Islands	21.7333	-71.5833	69	0	TGN	link%20)
KW	Kuwait	29.5	47.75	65	0	TGN	link%20)
SZ	Swaziland	-26.5	31.5	65	2	TGN	link%20)
SV	El Salvador	13.8333	-88.9167	65	1	TGN	link%20)
SM	San Marino	43.9333	12.4167	64	109	TGN	link%20)
BT	Bhutan	27.5	90.5	64	4922	TGN	link%20)
LB	Lebanon	33.8333	35.8333	60	3	TGN	link%20)
BV	Bouvet Island	-54.4333	3.4	57	17	TGN	link%20)
PS	Gaza Strip	31.4167	34.3333	56	75	TGN	link%20)
AF	Afghanistan	33	65	55	0	TGN	link%20)
WS	Samoa	-13.8	-172.133333	53	35	TGN	link%20)
LV	Latvia	57	25	53	6	TGN	link%20)
GW	Guinea-Bissau	12	-15	52	22	TGN	link%20)
LA	Laos	18	105	49	0	TGN	link%20)
GQ	Equatorial Guinea	2	10	49	0	TGN	link%20)
DJ	Djibouti	11.5	42.5	42	0	TGN	link%20)
CF	Central African Republic	7	21	41	0	TGN	link%20)
CG	Congo	-1	15	41	0	TGN	link%20)
KI	Kiribati	-5	-170	40	0	TGN	link%20)
NE	Niger	16	8	38	1	TGN	link%20)
ER	Eritrea	15	39	38	0	TGN	link%20)
AL	Albania	41	20	38	4	TGN	link%20)
PS	State of Palestine	31.92157	35.20329	38	640	TGN	link%20)
SO	Somalia	6	48	35	0	TGN	link%20)
PM	Saint Pierre and Miquelon	46.8333	-56.3333	31	2	TGN	link%20)
IM	Isle of Man	54.25	-4.5	29	3438	TGN	link%20)
UZ	Uzbekistan	41	64	29	0	TGN	link%20)
TJ	Tajikistan	39	71	28	0	TGN	link%20)
MK	North Macedonia	41.666	21.75	26	61	TGN	link%20)
KZ	Kazakhstan	48	68	24	0	TGN	link%20)
LY	Libya	25	17	22	0	TGN	link%20)
RW	Rwanda	-2	30	21	26	TGN	link%20)
TM	Turkmenistan	40	60	21	0	TGN	link%20)
ML	Mali	17	-4	20	1	TGN	link%20)
CC	Cocos Islands	-12	96.8333	17	4	TGN	link%20)
LT	Lithuania	56	24	17	3	TGN	link%20)
KG	Kyrgyzstan	41	75	17	0	TGN	link%20)
BF	Burkina Faso	13	-2	16	0	TGN	link%20)
SJ	Svalbard	78	20	16	404	TGN	link%20)
ME	Montenegro	42.5	19.3333	16	23	TGN	link%20)
ST	Sao Tome and Principe	1	7	15	0	TGN	link%20)
TL	Timor-Leste	-8.5833	126	14	1	TGN	link%20)
AZ	Azerbaijan	40.5	47.5	14	0	TGN	link%20)
SI	Slovenia	46.083	15	14	15	TGN	link%20)
JO	Jordan	31	36	13	1	TGN	link%20)
FM	Federated States of Micronesia	5	152	12	0	TGN	link%20)
AS	American Samoa	-14.3167	-170.5	12	0	TGN	link%20)
TG	Togo	8	1.1667	12	0	TGN	link%20)
GS	South Georgia and South Sandwich Islands	-56	-33	11	0	TGN	link%20)
TV	Tuvalu	-8	178	10	0	TGN	link%20)
LS	Lesotho	-29.5	28.25	10	0	TGN	link%20)
PS	West Bank	32	35.25	10	118	TGN	link%20)
BI	Burundi	-3.5	30	9	0	TGN	link%20)
GB	United Kingdom	54	-4.5	9	162	TGN	link%20)
TF	French Southern and Antarctic Lands	-43	67	8	1	TGN	link%20)
TD	Chad	15	19	7	0	TGN	link%20)
BG	Bulgaria	42.666	25.25	6	37	TGN	link%20)
BH	Bahrain	26	50.5	6	267	TGN	link%20)
AE	United Arab Emirates	24	54	5	0	TGN	link%20)
TK	Tokelau	-9	-171.75	4	0	TGN	link%20)
MD	Moldova	47.25	28.583	4	18	TGN	link%20)
BY	Belarus	53	28	4	0	TGN	link%20)
GG	Guernsey	49.5833	-2.333	3	0	TGN	link%20)
IO	British Indian Ocean Territory	-7	72.0167	3	0	TGN	link%20)
SS	South Sudan	7.5	30	3	0	TGN	link%20)
QA	Qatar	25.5	51.25	2	1	TGN	link%20)
MR	Mauritania	20	-12	2	1	TGN	link%20)
HM	Heard Island and McDonald Islands	-53	73	0	0	TGN	link%20)
IN	Bassas da India	-21.4167	39.7	0	0	TGN	link%20)
CK	Cook Islands	-16.083	-161.583	0	0	TGN	link%20)
RS	Serbia	44.166	20.833	0	0	TGN	link%20)

TGN source

other centroid sources already imported into the geocoder

https://github.com/gbif/geocode/blob/e1609c922f840939d9ccecf0ce8b1ef9a473f019/database/geolocate_centroids.sql https://github.com/gbif/geocode/blob/e1609c922f840939d9ccecf0ce8b1ef9a473f019/database/coordinatecleaner_centroids.sql

[ArthurChapman]

Further to my earlier post, and comments by @jhnwllr above - of the five centroids for Australia, there are two common ones that have been used for specimens and observations in the past.

The first is Lamberts Gravitational Centre (see reference at https://www.atlasobscura.com/places/lambert-centre-of-australia).

The second is Johnston's Geodetic Centre (see reference at https://www.xnatmap.org/adnm/docs/2013/1965%20JGS2.htm which also discusses how this and other "centres" were calculated). This latter paper shows the complexities in determining country/continental centroids.

For those interested - there is a paper here on the five Australian Centroids plus centroids for each of the Australian States and Territories (https://www.ga.gov.au/scientific-topics/national-location-information/dimensions/centre-of-australia-states-territories)

[jhnwllr]

I decided that collecting these centroids needed more organization, so I made repo to aggregate different centroid sources into one source. https://github.com/jhnwllr/catalogue-of-centroids

@MattBlissett

[ArthurChapman]

Great job @jhnwllr. Wouldn't be great to have the detailed methodology for each of the centroids. I know, from looking at Australia's, detailed methodologies are very difficult to find. Looking at the Australian ones, interesting that you found 9. I guess some of the more southern ones include Tasmania, whereas many of the others are for mainland Australia. Good job.

[MattBlissett]

To implement this, I think we should have a table of all reasonable centroids (Lambert's or Johnston's or geolocate or TCN or anyone else's method) for countries and country-like things (Australia with and without Tasmania, the UK with and without Shetland, USA with and without Alaska and Hawaii etc). (Some are already on the debug map -- NB some layers may crash a browser, but the centroid layer is fine.)

During interpretation, either for all records, or records without an uncertainty, or specimen records without an uncertainty, we can calculate the distance to the nearest centroid in metres and store the number, at least if it's below some maximum distance. Is 5km a reasonable cut-off? (The cut-off has implications for interpretation speed.)

The API can then allow filtering for distanceFromCentroid > X, where X could be any value, but the portal UI can have preset values if we like.

[ArthurChapman]

@MattBlissett - I agree such a table is a good idea. Thinking on why we are wanting this - does anyone use a centroid for the "USA" for collections where the centroid used includes Hawaii? Not sure that they do. Australia + Tasmania - maybe - but I don't think it is common - and then you also add long outliers (Macquarie Island, Christmas Island, Norfolk Island, etc.) I don't think anyone uses a centroid for recording "Australia" that would include any of those. By putting them in a table - it might encourage people to use them and I don't think that is a good idea. [It may be a fun exercise for a geographer - but that is not our motivation]. If we get too politically correct - what about France and all their Pacific Island "territories" I think that trying to include outlying islands for all countries could be a minefield - both politically and otherwise (South China Sea). I would avoid those.

Perhaps, the only way to determine what should and should not be included is to look at what people have used - for example looking at all collections that say "Australia", "Nova Hollandia" etc. and see what has been used and include those - ignoring the many other options that no one has actually used for biological collections. Note that you have countries whose boundaries, and thus extent and centroid,have changed over time, and thus the centroid will vary with year of georeferencing.

I guess, the second use is that we would want to encourage people who are retrospectively georeferencing and wish to use a centroid , to use a consistent centroid - i.e that we provide guidance - e.g. if we have several for Australia - one may be asterisked with a recommendation that this is the recommended centroid.

[jhnwllr]

@MattBlissett

For now I think we should use only centroids from "countries".

type == PCLI (places with an iso-code) in this file https://github.com/jhnwllr/catalogue-of-centroids/blob/master/centroids.tsv

I believe that 5km is a great cutoff.

[MattBlissett]

Does distanceFromCentroidInMeters (in the GBIF term namespace) seem reasonable for this?

[MattBlissett]

@jhnwllr and others, would you recommend calculating this value for all occurrences, or a subset (e.g. exclude observations)?

[jhnwllr]

I would say that we have to calculate distanceFromCentroidInMeters for everything even if "true" centroids are usually PRESERVED_SPECIMEN.

[ArthurChapman]

I'd be careful excluding all observation within a proscribed distanceFromCentroidInMeters from all centroids. I know that in Australia, only one or two of the 7 or so centroids have ever been used as a default with respect to collections/observations. This would mean excluding many good records that have little or nothing to do with the centroid other than coincidence of location. Also, uncritically excluding records from centroids is a problem. I think you would have to ignore records that already have a dwc:uncertantyInMeters that is a smallish number, because, I know in my own case for example, I have deliberately collected at the centroid locations and they will have an Uncertainty of less than a 100 meters or so. Perhaps excluding records from centroids should also take into account the location or verbatimLocation. If it says "Nova Hollandia" or "Australia" and nothing else, then the centroid will likely be an artificial location, but if the location says "near Lambert's Centre of Australia" then the location is likely to be a good location and should not be excluded.

[jhnwllr]

@ArthurChapman Right now the centroids are merely being tagged in our backed system and there isn't a final decision on exclusion or inclusion of centroids in downloads, maps ect.

My preference would be for centroids to be treated as neutrally as possible as "interesting locations", rather than immediately assume there is a problem. The fact that a point lies near a centroid should be surfaced to users (and publishers) probably as a somewhat neutral data quality flag.

[MattBlissett]

@MortenHofft, the API is now deployed, so the new filter can be added to the portal.

https://api.gbif.org/v1/occurrence/search?distance_from_centroid_in_meters=*,1000&basis_of_record=PRESERVED_SPECIMEN (query for searching for data at centroids)

https://api.gbif.org/v1/occurrence/search?distance_from_centroid_in_meters=2000,*&basis_of_record=PRESERVED_SPECIMEN (more common for searching for data not at centroids)

An exact distance match on a floating point number probably isn't much use, so I suggest we default to an open range (a minimum) like the 2000,* above.

[peggynewman]

This is a great discussion. I'm interested in whether we can provide guidance to data publishers on how to represent these kinds of geocoding so that it's quite obvious to users how the location has been determined, for example in the dataGeneralisations field - or does coordinateUncertainty give us enough? I've had an issue pop up with iNaturalist lately where an obscured (random placement in a .2 *.2 grid) platypus record ended up in a dam that doesn't have platypus, but it took a bit of investigation to work out that it had been obscured. The record states 28km uncertainty, but the fact that it's a random placement distinguishes it from a centroid.

[jhnwllr]

@peggynewman I think coordinateUncertainty and footprintWKT are probably the best fields for publishers to use. The fact that it is a centroid is sort of secondary to the uncertainty problem. I have seen publishers using various fields to indicate that a point is centroid.

A free text search for centroid gives many examples: https://www.gbif.org/occurrence/search?q=centroid

For example, this record says that it is a "centroid of Sweden" but gives an uncertainty of 50m :( https://www.gbif.org/occurrence/3431110088

[jhnwllr]

BTW this is now implemented, so I am closing this issue. https://www.gbif.org/occurrence/map?advanced=1&distance_from_centroid_in_meters=0,0