Aggregating continuous rows within a SQL table - sql

Question

I am trying to run an aggregate function on the following SQL table to sum up all the "LengthOfRecord" grouped by "Long+Lat" and only rows that are contiguous (i.e. "RowNumber" that is in running sequence).
+-----------+-----------+---------------+----------------+
| RowNumber | Vessel ID | Long+Lat | LengthOfRecord |
+-----------+-----------+---------------+----------------+
| 102313179 | Vessel 01 | 123.751 1.196 | 181 |
| 102313180 | Vessel 01 | 123.751 1.196 | 179 |
| 102313181 | Vessel 01 | 123.751 1.196 | 361 |
| 102313182 | Vessel 01 | 123.751 1.196 | 359 |
| 102313183 | Vessel 01 | 123.751 1.196 | 180 |
| 102313184 | Vessel 01 | 123.751 1.196 | 181 |
| 102313185 | Vessel 01 | 123.751 1.196 | 179 |
| 102313186 | Vessel 01 | 123.751 1.196 | 180 |
| 102313187 | Vessel 01 | 123.751 1.196 | 360 |
| 102313188 | Vessel 01 | 123.751 1.196 | 360 |
| 102313189 | Vessel 01 | 123.751 1.196 | 180 |
| 102313191 | Vessel 01 | 123.751 1.196 | 181 |
| 102313298 | Vessel 01 | 123.750 1.197 | 180 |
| 102313375 | Vessel 01 | 123.742 1.196 | 179 |
| 102313376 | Vessel 01 | 123.742 1.196 | 359 |
| 102313377 | Vessel 01 | 123.742 1.196 | 180 |
| 102313379 | Vessel 01 | 123.742 1.196 | 181 |
| 102313380 | Vessel 01 | 123.742 1.196 | 178 |
+-----------+-----------+---------------+----------------+
The following is the result that I am trying to achieve through SQL statements. Is there anyway that I can do this through an SQL query?
+-----------+---------------+----------------+
| Vessel ID | Long+Lat | LengthOfRecord |
+-----------+---------------+----------------+
| Vessel 01 | 123.751 1.196 | 2881 |
| Vessel 01 | 123.750 1.197 | 180 |
| Vessel 01 | 123.742 1.196 | 1077 |
+-----------+---------------+----------------+

Answer 1

You can do this using a difference in row numbers approach:
select vesselId, latLong, sum(lengthOfRecord)
from (select t.*,
row_number() over (partition by vesselId order by rowNumber) as seqnum,
row_number() over (partition by vesselId, latlong order by rowNumber) as seqnum_latlong
from table t
) t
group by (seqnum - seqnum_latlong), latLong, vesselId;
The difference of row number approach is a bit tricky to explain. It identifies adjacent rows with the same values. If you run the subquery, you will see how the calculation works.

Answer 2

This may be long but hopefully covers your requirements in a relatively readable manner:
declare #t table (RowNumber int not null, VesselID varchar(17) not null,
LatLong varchar(19),LengthOfRecord int not null)
insert into #t(RowNumber,VesselID,LatLong,LengthOfRecord) values
(102313179,'Vessel 01','123.751 1.196',181),
(102313180,'Vessel 01','123.751 1.196',179),
(102313181,'Vessel 01','123.751 1.196',361),
(102313182,'Vessel 01','123.751 1.196',359),
(102313183,'Vessel 01','123.751 1.196',180),
(102313184,'Vessel 01','123.751 1.196',181),
(102313185,'Vessel 01','123.751 1.196',179),
(102313186,'Vessel 01','123.751 1.196',180),
(102313187,'Vessel 01','123.751 1.196',360),
(102313188,'Vessel 01','123.751 1.196',360),
(102313189,'Vessel 01','123.751 1.196',180),
(102313191,'Vessel 01','123.751 1.196',181),
(102313298,'Vessel 01','123.750 1.197',180),
(102313375,'Vessel 01','123.742 1.195',179),
(102313376,'Vessel 01','123.742 1.195',359),
(102313377,'Vessel 01','123.742 1.195',180),
(102313379,'Vessel 01','123.742 1.195',181),
(102313380,'Vessel 01','123.742 1.195',178)
;With ContiguousRN as (
select
*,
ROW_NUMBER() OVER (PARTITION BY VesselID ORDER BY RowNumber) as rn
from
#t
), Starts as (
select
r1.VesselID,
r1.rn,
r1.LatLong,
ROW_NUMBER() OVER (PARTITION BY r1.VesselID ORDER BY r1.rn) as srn
from
ContiguousRN r1
left join
ContiguousRN r2
on
r1.rn = r2.rn + 1 and
r1.VesselID = r2.VesselID and
r1.LatLong = r2.LatLong
where
r2.rn is null
), Ends as (
select
r1.VesselID,
r1.rn,
r1.LatLong,
ROW_NUMBER() OVER (PARTITION BY r1.VesselID ORDER BY r1.rn) as srn
from
ContiguousRN r1
left join
ContiguousRN r2
on
r1.rn = r2.rn - 1 and
r1.VesselID = r2.VesselID and
r1.LatLong = r2.LatLong
where
r2.rn is null
), Sequences as (
select
s.VesselID,
s.LatLong,
s.rn as StartRow,e.rn as EndRow
from
Starts s
inner join
Ends e
on
s.VesselID = e.VesselID and
s.srn = e.srn
)
select
seq.VesselID,
seq.LatLong,
(select SUM(LengthOfRecord) from ContiguousRN r
where r.VesselID = seq.VesselID and
r.rn between seq.StartRow and seq.EndRow) as LengthOfRecord
from Sequences seq
I've changed some of the column names so that I don't have to keep quoting them because they contain spaces or punctuation. I'd also recommend you either store the position in a genuine geography-typed column or you store lat and long in separate columns.
So, the above query. The first CTE (ContiguousRN) just arranges for us to have row numbers (rn) that don't have gaps, unlike RowNumber. The second and third queries locate the rows within the table that are the start and end of each run - basically, locating rows where the immediate preceding or succeeding rows have different LatLong values. We also generate a separate series of row numbers for just these rows, so that, in Sequences, we can combine each start row with its corresponding end row.
Finally, in the last select, we bring this together and we total up all of the rows that sit between each start and end marker.
I've assumed throughout that VesselID should be used as some form of partitioning value and that your actual data may contain details for more than one vessel and this process shouldn't mingle the data together. If that's not so, you can remove most of the conditions around VesselID in the above.
Results:
VesselID LatLong LengthOfRecord
----------------- ------------------- --------------
Vessel 01 123.751 1.196 2881
Vessel 01 123.750 1.197 180
Vessel 01 123.742 1.195 1077