-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path2-clean.R
1010 lines (851 loc) · 48.1 KB
/
2-clean.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
######################################################################################################################################################
# 1. Remove known historical inaccuracies
# https://www.cl.cam.ac.uk/research/dtg/weather/inaccuracies.html
# On 31st July 2008, the readings for humidity pressure and wind speed stuck
# at about 6pm BST. The logger was reset on 1st August at about 9:00am
weather.08.08.01 <- weather.raw[ds > '2008-08-01 00:00:00']
weather.08.08.01 <- data.table(weather.08.08.01)
weather.08.08.01 <- weather.08.08.01[, .(ds,
temperature,
humidity,
dew.point,
pressure,
wind.speed.mean,
wind.bearing.mean,
wind.speed.max,
rainfall,
sunshine,
rainy,
sunny,
sunrise,
sunset,
secs)]
setkey(weather.08.08.01, ds)
summary(weather.08.08.01)
# Add missing observations to weather.08.08.01 as NAs
min.ds <- min(weather.08.08.01$ds)
max.ds <- max(weather.08.08.01$ds)
all.time.stamps <- seq(min.ds, max.ds, by='30 mins')
weather.08.08.01 <- merge(weather.08.08.01, data.table(ds=all.time.stamps), by='ds', all=TRUE, roll=NA)
weather.08.08.01[, HMS:=strftime(ds, format = "%H:%M:%S")]
weather.08.08.01[, secs:=as.numeric(as.difftime(HMS))]
weather.08.08.01$HMS <- NULL
summary(weather.08.08.01)
weather.08.08.01$missing <- 0
weather.08.08.01$known_inaccuracy <- 0
# rainfall and sunshine measurements are known to be very problematic
# Not using for now
# Better not have -ve rainfall
weather.08.08.01[rainfall < 0, 'known_inaccuracy'] <- 1
weather.08.08.01[rainfall < 0, 'rainfall'] <- NA # TODO or 0?
# Low humidity problem from 2015-11-24 08:30:00 to 2015-11-27 13:00:00
# all 3 or less (very influential)
# 20 here is a bit arbitrary but agrees with ISD Cam data
#weather.08.08.01[humidity < 20, 'humidity'] <- NA
# Exclude (individually)
weather.08.08.01[sunny==1 & ds < sunrise, 'sunshine'] <- NA
weather.08.08.01[sunny==1 & ds > sunset, 'sunshine'] <- NA
weather.08.08.01[sunny==1 & ds < sunrise, 'sunshine'] <- NA
weather.08.08.01[sunny==1 & ds > sunset, 'sunshine'] <- NA
weather.08.08.01[sunny==1 & ds < sunrise, 'known_inaccuracy'] <- 1
weather.08.08.01[sunny==1 & ds > sunset, 'known_inaccuracy'] <- 1
weather.08.08.01[sunny==1 & ds < sunrise, 'known_inaccuracy'] <- 1
weather.08.08.01[sunny==1 & ds > sunset, 'known_inaccuracy'] <- 1
#weather.08.08.01 <- weather.08.08.01[!(sunny==1 & ds < sunrise)]
#weather.08.08.01 <- weather.08.08.01[!(sunny==1 & ds > sunset)]
#dim(weather.08.08.01)
#weather.08.08.01 <- weather.08.08.01[, !c("rainfall", "sunshine")]
# Find max UK and Cambridge recorded wind speeds?
# Remove anything greater than 350
# Based on ISD Cam data and boxplot
#weather.08.08.01 <- weather.08.08.01[wind.speed.max <= 600]
#weather.08.08.01 <- weather.08.08.01[wind.speed.mean <= 600]
#weather.08.08.01[wind.speed.max >= 350, 'wind.speed.max'] <- NA
#weather.08.08.01[wind.speed.mean >= 350, 'wind.speed.mean'] <- NA
# Remove unusual wind bearings fix
# Don't remove - replace with NAs?!
#weather.08.08.01 <- weather.08.08.01[wind.bearing.mean %in% seq(0, 360, 45), 'wind.bearing.mean'] <- NA
# Unrealistic low temperature fix
#weather.08.08.01 <- weather.08.08.01[temperature != -400, ]
# Remove unusual secs values
weather.08.08.01 <- weather.08.08.01[secs %in% seq(0, 86400, 1800), ]
# On 12th August 2008, rainfall was not recorded, despite heavy rain falling over
# Cambridge in the morning.
#weather.08.08.01 <- weather.08.08.01[ds != '2008-08-12', rainfall] <- NA
# From 19th August 2008 to 27th August 2008 inclusive, rainfall was not recorded
#weather.08.08.01 <- weather.08.08.01[ds < '2008-08-19' | ds > '2008-08-27', ]
# From 3rd September 2008 to 4th September 2008 at about 1pm, rainfall was again not
# recorded due to a blocked sensor
#weather.08.08.01 <- weather.08.08.01[ds != '2008-09-03' & ds != '2008-09-04', ]
# From 25th October 2008 to 4th November 2008, rainfall was again not correctly recorded
# due to a partially blocked sensor.
# Some rainfall was recorded, but the amounts recorded are far too small to be correct.
#weather.08.08.01 <- weather.08.08.01[ds < '2008-10-25' | ds > '2008-11-04', ]
# On 3rd April 2009 the Sunshine and Humidity sensors became stuck. Readings for Temperature
# and Humidity until 17:30 are known to be wrong
weather.08.08.01[ds >= '2009-04-03 00:00' & ds < '2009-04-03 17:30', 'temperature'] <- NA
weather.08.08.01[ds >= '2009-04-03 00:00' & ds < '2009-04-03 17:30', 'humidity'] <- NA
weather.08.08.01[ds >= '2009-04-03 00:00' & ds < '2009-04-03 17:30', 'known_inaccuracy'] <- 1
weather.08.08.01[ds >= '2009-04-03 00:00' & ds < '2009-04-03 17:30', 'known_inaccuracy'] <- 1
# The Sunshine sensor appears to have become a daylight sensor - the threshold for
# determining 'Sunny' has changed, possibly from 3rd April when it was first noticed
# that it had become stuck.
# Over the Easter weekend of 10th April 2009 to 13th April 2009, the rainfall was not
# recorded (blocked sensor again).
# It was unblocked on the Tuesday once we could gain access.
#weather.08.08.01 <- weather.08.08.01[ds < '2009-04-10' | ds > '2009-04-13', ]
# Between 6th and 28th February, the rainfall sensor has not been recording any rainfall,
# due to water damage (!) to one of the cables.
#weather.08.08.01 <- weather.08.08.01[ds < '2010-02-06' | ds > '2010-02-28', ]
# No results were recorded for 14th August 2010, and there was an interruption on 16th
# August 2010 due to a power failure
weather.08.08.01[ds >= '2010-08-14 00:00' & ds < '2010-08-14 21:00', ]
# Insert missing NAs further below
weather.08.08.01[ds >= '2010-08-16 00:00' & ds < '2010-08-16 23:30', ]
# temperature may be stuck between 2010-08-16 00:30:00 and 2010-08-16 06:30:00
# otherwise looks OK - No changes
# A blocked rain sensor on 23rd August 2010 may have given high readings for rainfall on
# that day.
#weather.08.08.01 <- weather.08.08.01[ds != '2010-08-23', ]
# A blocked rain sensor on 26th February 2011 gave extreme readings for rainfall on that
# day of >170mm. A few mm would be a more realistic amount.
#weather.08.08.01 <- weather.08.08.01[ds != '2011-02-26', ]
# A blocked rain sensor from 19th August 2011 to 23rd August 2011 gave gave extreme readings
# for rainfall whenever it rained.
# It really isn't clear why a blocked sensor registers an excess of rainfall, rather than
# just none. Perhaps 10mm for 19th,
# and 4mm up to 1pm on the 23rd would be a better estimate.
#weather.08.08.01 <- weather.08.08.01[ds < '2011-08-19' | ds > '2011-08-23', ]
# The rain sensor failed completely on 6th September. We will endeavour to find a
# replacement as soon as possible.
# On 30 October 2011, the data logging system experienced a failure at 1:00 AM. No data was
# recorded until 11:28 AM on 31 October 2011.
# We will investigate the issues that led to the interruption. A temporary fix was found for
# the rain sensor, but until replaced the
# recorded precipitation data should not be trusted.
#weather.08.08.01[ds >= '2011-10-30 01:00' & ds < '2011-10-31 11:30', ]
# Insert missing NAs further below
# On Tuesday 10 January 2012, no data has been recorded between 8:30 AM and 3:30 PM. This
# was due to a scheduled power outage in the Computer Laboratory building, followed by some
# issues on cold-starting the weather logging system.
weather.08.08.01[ds >= '2012-01-10 08:30' & ds < '2012-01-10 15:30', ]
# Insert missing NAs further below
# On 30 January 2012, a planned interruption occured between 4:30 PM and 6:30 PM. On this
# occasion, we have updated our logging software to process data from the new precipitation
# sensor (Thies Clima). As a result of a data processing error, some precipitation readings
# later that day and early morning on 31 January are negative. A fix is in place as of 31
# Jan 2012, 4:30 PM.
#weather.08.08.01 <- weather.08.08.01[ds < '2012-01-30' | ds > '2012-01-31', ]
# 3 readings which look reasonable - no changes
# On 20 August 2012, a cottonwood seed stuck in the sensor area (protective caps) and moved
# around by the wind caused erroneous precipitation readings before 2:26 PM.
#weather.08.08.01 <- weather.08.08.01[ds != '2012-08-20', ]
# On the 28th January 2015 around 11:00 AM, the humidity sensor connection became
# intermittant due to corrosion.
# Humidity and Dew Point readings were unreliable and occasionally erratic until
# the issue was resolved on the 2nd February 2015 at 6:00PM
weather.08.08.01[ds >= '2015-01-28 11:00' & ds < '2015-02-02 18:00', 'humidity'] <- NA
weather.08.08.01[ds >= '2015-01-28 11:00' & ds < '2015-02-02 18:00', 'dew.point'] <- NA
weather.08.08.01[ds >= '2015-01-28 11:00' & ds < '2015-02-02 18:00', 'known_inaccuracy'] <- 1
weather.08.08.01[ds >= '2015-01-28 11:00' & ds < '2015-02-02 18:00', 'known_inaccuracy'] <- 1
# From late Sept 2020 until 15 Jan 2021, no precipitation data was recorded due to a failed
# power supply for the Thies sensor, and the inaccessibility of the building due to Covid.
# dew.point should never be greater than temperature!
weather.08.08.01[dew.point > temperature, 'known_inaccuracy'] <- 1
weather.08.08.01[dew.point > temperature, 'dew.point'] <- NA
# wind.speed.mean should never be greater than wind.speed.max!
weather.08.08.01[wind.speed.mean > wind.speed.max, 'known_inaccuracy'] <- 1
weather.08.08.01[wind.speed.mean > wind.speed.max, 'wind.speed.max'] <- weather.08.08.01[wind.speed.mean > wind.speed.max, 'wind.speed.mean']
######################################################################################################################################################
# 2. Check equation approimations for calculating temperature, dew.point and humidity
# Temperature and dew.point approimations https://earthscience.stackexchange.com/q/14899/18379
# Relative humidity approimation https://earthscience.stackexchange.com/q/16570/18379
# Currently no fixable cases :-(
# TODO Check any existing values contradict the equation approximations
weather.08.08.01[is.na(temperature) & !is.na(dew.point) & !is.na(humidity), ]
# Empty data.table (0 rows and 18 cols): ds,temperature,humidity,dew.pointressure,wind.speed.mean...
weather.08.08.01[is.na(humidity) & !is.na(dew.point) & !is.na(temperature), ]
# Empty data.table (0 rows and 18 cols): ds,temperature,humidity,dew.pointressure,wind.speed.mean...
weather.08.08.01[is.na(dew.point) & !is.na(humidity) & !is.na(temperature), ]
# Empty data.table (0 rows and 18 cols): ds,temperature,humidity,dew.pointressure,wind.speed.mean...
isd[is.na(temp) & !is.na(dew_point) & !is.na(rh), ]
# Empty data.table (0 rows and 8 cols): time,temp,wd,ws,dew_point,rh...
isd[is.na(dew_point) & !is.na(temp) & !is.na(rh), ]
# Empty data.table (0 rows and 8 cols): time,temp,wd,ws,dew_point,rh...
isd[is.na(rh) & !is.na(temp) & !is.na(dew_point), ]
# Empty data.table (0 rows and 8 cols): time,temp,wd,ws,dew_point,rh...
######################################################################################################################################################
# 3. Compare Computer Lab and Cambridge Airport (ISD) measurements
# Calculate correlations and plot observation distributions
# Remove obvious outliers from both data sets
# Annecdotally, Cambridge Airport appears to be higher quality data source
setkey(isd.filled, time)
setkey(weather.08.08.01, ds)
weather.08.08.01$isd_outlier <- 0
weather.cors <- weather.08.08.01[isd.filled, .(ds,
temperature,
temp,
humidity,
rh,
dew.point,
dew_point,
wind.speed.mean,
wind.speed.max,
ws,
wind.bearing.mean,
wd,
rainfall,
sunshine,
ceil_hgt,
visibility)]
cor(weather.cors[complete.cases(weather.cors), .(temperature, temp*10)])
# temperature V2
# temperature 1.0000000 0.9745562
# V2 0.9745562 1.0000000
cor(weather.cors[complete.cases(weather.cors), .(dew.point, dew_point*10)])
# dew.point V2
# dew.point 1.0000000 0.9117574
# V2 0.9117574 1.0000000
cor(weather.cors[complete.cases(weather.cors), .(humidity, rh)])
# humidity rh
# humidity 1.0000000 0.8962506
# rh 0.8962506 1.0000000
cor(weather.cors[complete.cases(weather.cors), .(wind.speed.mean/10, ws)])
# V1 ws
# V1 1.0000000 0.7881784
# ws 0.7881784 1.0000000
# Usable or not?
# What's the harm in using it and what's the harm in not using it?
cor(weather.cors[complete.cases(weather.cors), .(wind.speed.max/10, ws)])
cor(weather.cors[complete.cases(weather.cors), .(wind.bearing.mean, wd)])
# wind.bearing.mean wd
# wind.bearing.mean 1.0000000 0.6770643
# wd 0.6770643 1.0000000
# Usable or not?
# What's the harm in using it and what's the harm in not using it?
plot(density(weather.cors[complete.cases(weather.cors), temperature]), col='blue', ylim=c(0,0.006))
lines(density(weather.cors[complete.cases(weather.cors), temp*10]), col='green')
boxplot(weather.cors[complete.cases(weather.cors), .(temperature, temp*10)])
# Restrict to -70 to 380
weather.08.08.01[temperature < -70, 'isd_outlier'] <- 1
weather.08.08.01[temperature > 380, 'isd_outlier'] <- 1
weather.08.08.01[temperature < -70, 'temperature'] <- NA
weather.08.08.01[temperature > 380, 'temperature'] <- NA
isd.filled[temp < -7, 'temp'] <- NA
isd.filled[temp > 38, 'temp'] <- NA
plot(density(weather.cors[complete.cases(weather.cors), dew.point]), col='blue')
lines(density(weather.cors[complete.cases(weather.cors), dew_point*10]), col='green')
boxplot(weather.cors[complete.cases(weather.cors), .(dew.point, dew_point*10)])
# Restrict to -100 to 210
weather.08.08.01[dew.point < -100, 'isd_outlier'] <- 1
weather.08.08.01[dew.point > 210, 'isd_outlier'] <- 1
weather.08.08.01[dew.point < -100, 'dew.point'] <- NA
weather.08.08.01[dew.point > 210, 'dew.point'] <- NA
isd.filled[dew_point < -10, 'dew_point'] <- NA
isd.filled[dew_point > 21, 'dew_point'] <- NA
plot(density(weather.cors[complete.cases(weather.cors), humidity]), col='blue', ylim=c(0,0.035))
lines(density(weather.cors[complete.cases(weather.cors), rh]), col='green')
boxplot(weather.cors[complete.cases(weather.cors), .(humidity, rh)])
# Restrict to 20 to 100
#weather.08.08.01[humidity < 20, 'isd_outlier'] <- 1
weather.08.08.01[humidity > 100, 'isd_outlier'] <- 1
#weather.08.08.01[humidity < 20, 'humidity'] <- NA
weather.08.08.01[humidity > 100, 'humidity'] <- NA
#isd.filled[rh < 20, 'rh'] <- NA
isd.filled[rh > 100, 'rh'] <- NA
# NOTE Setting these NAs to 20 or 100 is not necessarily the right thing to do
plot(density(weather.cors[complete.cases(weather.cors), wind.speed.mean/10]), col='blue', ylim=c(0, 0.2))
lines(density(weather.cors[complete.cases(weather.cors), ws]), col='green')
boxplot(weather.cors[complete.cases(weather.cors), .(wind.speed.mean/10, ws)])
plot(density(weather.cors[complete.cases(weather.cors), wind.speed.max/10]), col='blue', ylim=c(0, 0.2))
lines(density(weather.cors[complete.cases(weather.cors), ws]), col='green')
boxplot(weather.cors[complete.cases(weather.cors), .(wind.speed.max/10, ws)])
# Restrict to 0 to 350
weather.08.08.01[wind.speed.mean < 0, 'isd_outlier'] <- 1
weather.08.08.01[wind.speed.max < 0, 'isd_outlier'] <- 1
weather.08.08.01[wind.speed.mean > 350, 'isd_outlier'] <- 1
weather.08.08.01[wind.speed.max > 700, 'isd_outlier'] <- 1
weather.08.08.01[wind.speed.mean < 0, 'wind.speed.mean'] <- NA
weather.08.08.01[wind.speed.max < 0, 'wind.speed.max'] <- NA
weather.08.08.01[wind.speed.mean > 350, 'wind.speed.mean'] <- NA
weather.08.08.01[wind.speed.max > 700, 'wind.speed.max'] <- NA
isd.filled[ws < 0, 'ws'] <- NA
#isd.filled[ws > 350, 'ws'] <- NA
plot(density(weather.cors[complete.cases(weather.cors), wind.bearing.mean]), col='blue')
lines(density(weather.cors[complete.cases(weather.cors), wd]), col='green')
boxplot(weather.cors[complete.cases(weather.cors), .(wind.bearing.mean, wd)])
plot(density(weather.08.08.01[, pressure], na.rm = TRUE))
boxplot(weather.08.08.01[, pressure], na.rm = TRUE)
# Restrict to 950 to 1055
#weather.08.08.01[pressure < 950, 'isd_outlier'] <- 1
#weather.08.08.01[pressure > 1055, 'isd_outlier'] <- 1
#weather.08.08.01[pressure < 950, 'pressure'] <- NA
#weather.08.08.01[pressure > 1055, 'pressure'] <- NA
# NOTE Setting these NAs to 950 or 1055 is not necessarily the right thing to do
######################################################################################################################################################
# 4. Remove long runs of consecutively equal values
# TODO Generalise to find anomolously low (but non-zero) variance observations over varying window sizes
weather.08.08.01$long_run <- 0
get_consec_run_lengths <- function(df, varcol, tscol, sd.factor=3) {
b <- rle(df[, get(varcol)])
lengths.sd <- sd(b$lengths) * sd.factor
cat(paste("sd threshold: ", round(lengths.sd, 2), "\n\n"))
flush.console()
weather.rle <- data.table(value = b$values, runlength = b$lengths)
print(summary(weather.rle))
cat("\n")
flush.console()
weather.rle$end <- cumsum(weather.rle$runlength)
weather.rle$start <- weather.rle$end - weather.rle$runlength + 1
weather.rle$starttime <- df[weather.rle$start, get(tscol)]
weather.rle$endtime <- df[weather.rle$end, get(tscol)]
print(weather.rle[runlength > lengths.sd,][order(runlength), .(value, runlength, start, end, starttime, endtime)])
cat("\n")
flush.console()
x <- weather.rle[runlength > lengths.sd,][order(runlength), .(end, start)]
return(sapply(1:nrow(x), function(i, x) { seq(x[i]$start, x[i]$end) }, x=x))
}
# Only removing the most extreme outliers
isd.temp.consecs <- get_consec_run_lengths(isd.filled, 'temp', 'time', 5) # No action
isd.rh.consecs <- get_consec_run_lengths(isd.filled, 'rh', 'time', 5) # No action
isd.ws.consecs <- get_consec_run_lengths(isd.filled, 'ws', 'time', 5) # No action
isd.wd.consecs <- get_consec_run_lengths(isd.filled, 'wd', 'time', 5) # No action
isd.dp.consecs <- get_consec_run_lengths(isd.filled, 'dew_point', 'time', 5) # No action
weather.dp.consecs <- get_consec_run_lengths(weather.08.08.01, 'dew.point', 'ds', 5) # No action
weather.pr.consecs <- get_consec_run_lengths(weather.08.08.01, 'pressure', 'ds', 5) # No action
weather.ws.consecs <- get_consec_run_lengths(weather.08.08.01, 'wind.speed.mean', 'ds', 5) # No action
weather.ws.consecs <- get_consec_run_lengths(weather.08.08.01, 'wind.speed.max', 'ds', 5) # No action
weather.wd.consecs <- get_consec_run_lengths(weather.08.08.01, 'wind.bearing.mean', 'ds', 12)
weather.08.08.01[unlist(weather.wd.consecs), 'long_run'] <- 1
weather.08.08.01[unlist(weather.wd.consecs), 'wind.bearing.mean'] <- NA
weather.rh.consecs <- get_consec_run_lengths(weather.08.08.01, 'humidity', 'ds', 50)
# max run lengths occur in winter months, all have max 100 value, none since 2015
weather.08.08.01[unlist(weather.rh.consecs), 'long_run'] <- 1
weather.08.08.01[unlist(weather.rh.consecs), 'humidity'] <- NA
weather.temp.consecs <- get_consec_run_lengths(weather.08.08.01, 'temperature', 'ds', 6)
# So far (30/03/21) the worst offender here is temperature which was
# stuck at the same value for 36 days!
# And not marked in the known inaccuracies :-(
# 2015-11-30 11:30:00 2016-01-08 15:00:00
weather.08.08.01[ds > '2015-11-30 11:30:00' & ds < '2016-01-08 15:00:00', 'long_run'] <- 1
weather.08.08.01[ds > '2015-11-30 11:30:00' & ds < '2016-01-08 15:00:00', 'temperature'] <- NA
######################################################################################################################################################
# 5. Check for measurement "spikes"
# Where spikes are sudden large increasing/decreasing observations
# followed by approximate return to previous value
# So far, spikes are limited to lengths of 1 observation
weather.08.08.01$spike <- 0
get_large_spikes <- function(df, varcol, tscol, sd.factor=3) {
df$row.no <- 1:nrow(df)
diffs <- df[, .(time=get(tscol),
row.no,
value=get(varcol),
diff.before=get(varcol) - shift(get(varcol)),
diff.after=get(varcol) - shift(get(varcol), type='lead'))]
print(summary(diffs[, .(value, diff.before, diff.after)]))
cat("\n")
flush.console()
diff.before.sd <- sd.factor * sd(diffs$diff.before, na.rm=TRUE)
diff.after.sd <- sd.factor * sd(diffs$diff.after, na.rm=TRUE)
cat(paste("sd before threshold: ", round(diff.before.sd, 2), "\n"))
cat(paste("sd after threshold: ", round(diff.after.sd, 2), "\n\n"))
flush.console()
print(diffs[abs(diff.before) > diff.before.sd &
abs(diff.after) > diff.after.sd &
((diff.before > 0 & diff.after > 0) |
(diff.before < 0 & diff.after < 0)),])
flush.console()
return(diffs[abs(diff.before) > diff.before.sd &
abs(diff.after) > diff.after.sd &
((diff.before > 0 & diff.after > 0) |
(diff.before < 0 & diff.after < 0)), row.no])
}
weather.temp.spikes <- get_large_spikes(weather.08.08.01, 'temperature', 'ds', 5)
weather.rh.spikes <- get_large_spikes(weather.08.08.01, 'humidity', 'ds', 5)
# 30 sporadic humidity spikes from 2015-11-29 20:00:00 to 2016-01-08 15:00:00
# The inaccuracies.html file lists an earlier humidity sensor problem (low measurements)
weather.pres.spikes <- get_large_spikes(weather.08.08.01, 'pressure', 'ds', 5)
weather.dp.spikes <- get_large_spikes(weather.08.08.01, 'dew.point', 'ds', 5)
weather.ws.spikes <- get_large_spikes(weather.08.08.01, 'wind.speed.mean', 'ds', 5)
weather.ws.spikes <- get_large_spikes(weather.08.08.01, 'wind.speed.max', 'ds', 5)
weather.08.08.01[weather.temp.spikes, 'spike'] <- 1
weather.08.08.01[weather.rh.spikes, 'spike'] <- 1
weather.08.08.01[weather.pres.spikes, 'spike'] <- 1
weather.08.08.01[weather.dp.spikes, 'spike'] <- 1
weather.08.08.01[weather.ws.spikes, 'spike'] <- 1
weather.08.08.01[weather.temp.spikes, 'temperature'] <- NA
weather.08.08.01[weather.rh.spikes, 'humidity'] <- NA
weather.08.08.01[weather.pres.spikes, 'pressure'] <- NA
weather.08.08.01[weather.dp.spikes, 'dew.point'] <- NA
weather.08.08.01[weather.ws.spikes, 'wind.speed.mean'] <- NA
weather.08.08.01[weather.ws.spikes, 'wind.speed.max'] <- NA
isd.temp.spikes <- get_large_spikes(isd.filled, 'temp', 'time', 5)
isd.rh.spikes <- get_large_spikes(isd.filled, 'rh', 'time', 3) # No action
isd.dp.spikes <- get_large_spikes(isd.filled, 'dew_point', 'time', 5)
isd.ws.spikes <- get_large_spikes(isd.filled, 'ws', 'time', 5)
isd.filled[isd.temp.spikes, 'temp'] <- NA
isd.filled[isd.dp.spikes, 'dp'] <- NA
isd.filled[isd.ws.spikes, 'ws'] <- NA
######################################################################################################################################################
# 6. Remove Computer Lab and Airport outliers using Cook's distance
# See http://r-statistics.co/Outlier-Treatment-With-R.html
# TODO Use more principled approach to outlier removal
# Possibly seasonal - bin cooksd values into days 48 * 12 = 576 big bins :-)
# calculate 99th percentile
# build loess model
# exclude values above loess smoothed 99th percentile
weather.08.08.01$cooksd_out <- 0
get_cooks_dist_outliers <- function(df, tscol, formula, cd.factor=3) {
df[, year:=as.numeric(strftime(get(tscol), format = "%Y"))]
df[, doy:=as.numeric(strftime(get(tscol), format = "%j"))]
df[, HMS:=strftime(get(tscol), format = "%H:%M:%S")]
df[, secs:=as.numeric(as.difftime(HMS))]
mod <- lm(formula, df)
print(summary(mod))
cooksd <- cooks.distance(mod)
cat("Cook's distance:\n")
print(summary(cooksd))
plot(cooksd, main="Influential Obs by Cooks distance") # slow
text(x=1:length(cooksd)+1,
y=cooksd,
labels=ifelse(cooksd > 2*cd.factor**mean(cooksd, na.rm=T), names(cooksd), ""),
col="red") # add labels
abline(h = cd.factor*mean(cooksd, na.rm=T), col="red") # add cutoff line
cat("\nInfluential:")
print(table(cooksd > cd.factor*mean(cooksd, na.rm=T)))
influential <- as.numeric(names(cooksd)[(cooksd > cd.factor*mean(cooksd, na.rm=T))]) # influential row numbers
df$secs <- NULL
df$year <- NULL
df$HMS <- NULL
df$doy <- NULL
cat("\nOnly influential:\n")
print(summary(df[ influential, ]))
cat("\nNo influential:\n")
print(summary(df[-influential, ]))
return(influential)
}
weather.form <- 'secs+doy+year ~ temperature+humidity+dew.point+pressure+wind.speed.mean+wind.bearing.mean+wind.speed.max'
weather.cooksd.influential <- get_cooks_dist_outliers(weather.08.08.01, 'ds', weather.form, 15)
weather.08.08.01[weather.cooksd.influential, 'cooksd_out'] <- 1
weather.08.08.01[weather.cooksd.influential, c('temperature', 'humidity', 'dew.point', 'pressure', 'wind.speed.mean', 'wind.bearing.mean', 'wind.speed.max')] <- NA
# TODO Setting all variables in line above to NA is bit extreme
# Run get_cl_outliers_using_isd for single features in turn
isd.filled[, time:=as.POSIXct(time, tz="GMT")]
isd.form <- 'secs+doy+year ~ temp+rh+dew_point+ws+wd'
isd.cooksd.influential <- get_cooks_dist_outliers(isd.filled, 'time', isd.form, 15)
isd.filled[isd.cooksd.influential, c('temp', 'wd', 'ws', 'dew_point', 'rh')] <- NA
# TODO Setting all variables in line above to NA is bit extreme
# Run get_cl_outliers_using_isd for single features in turn
######################################################################################################################################################
# 7. Use Cambridge Airport (weather.isd) to find Computer Lab (weather.08.08.01) outliers
# Potentially exclude 1 or 2 thousand more measurements
isd.renamed <- isd.filled[, .(ds=time,
temperature=temp*10,
humidity=rh,
dew.point=dew_point,
pressure=NA,
wind.speed.mean=ws*10,
wind.bearing.mean=wd,
wind.speed.max=ws*10,
ceil_hgt=ceil_hgt,
visibility=visibility)]
weather.isd <- merge(weather.08.08.01, isd.renamed, by='ds', all.x=TRUE, all.y=TRUE)
weather.isd$isd_3_sigma <- 0
get_cl_outliers_using_isd <- function(df, varcol, tscol, sd.factor=3) {
varcol.x <- paste0(varcol, '.x')
varcol.y <- paste0(varcol, '.y')
varcol.sd <- sd.factor * sd(df[, .(get(varcol.x) - get(varcol.y))][!is.na(V1), V1], na.rm=TRUE)
cat(paste("sd threshold: ", round(varcol.sd, 2), "\n\n"))
flush.console()
print(summary(df[, .(diff=get(varcol.x) - get(varcol.y))][!is.na(diff)]))
cat("\n")
flush.console()
df$row.no <- as.numeric(rownames(df))
print(df[, .(get(tscol),
get(varcol.x),
get(varcol.y),
get(varcol.x) - get(varcol.y),
row.no)][!is.na(V4) & abs(V4) > varcol.sd, .(row.no,
time=V1,
varcol.x=V2,
varcol.y=V3,
diff=V4)])
flush.console()
return(df[, .(get(varcol.x) - get(varcol.y),
row.no)][!is.na(V1) & abs(V1) > varcol.sd, row.no])
}
cl.isd.temp.outliers <- get_cl_outliers_using_isd(weather.isd, 'temperature', 'ds', 5)
cl.isd.rh.outliers <- get_cl_outliers_using_isd(weather.isd, 'humidity', 'ds', 5)
cl.isd.dp.outliers <- get_cl_outliers_using_isd(weather.isd, 'dew.point', 'ds', 4)
cl.isd.ws.outliers <- get_cl_outliers_using_isd(weather.isd, 'wind.speed.mean', 'ds', 5)
cl.isd.ws.max.outliers <- get_cl_outliers_using_isd(weather.isd, 'wind.speed.max', 'ds', 5)
weather.isd[cl.isd.temp.outliers, 'isd_3_sigma'] <- 1
weather.isd[cl.isd.rh.outliers, 'isd_3_sigma'] <- 1
weather.isd[cl.isd.dp.outliers, 'isd_3_sigma'] <- 1
weather.isd[cl.isd.ws.outliers, 'isd_3_sigma'] <- 1
weather.isd[cl.isd.temp.outliers, 'temperature.x'] <- NA
weather.isd[cl.isd.rh.outliers, 'humidity.x'] <- NA
weather.isd[cl.isd.dp.outliers, 'dew.point.x'] <- NA
weather.isd[cl.isd.ws.outliers, 'wind.speed.mean.x'] <- NA
weather.isd[cl.isd.ws.max.outliers, 'wind.speed.max.x'] <- NA
weather.isd$missing <- as.integer(!complete.cases(weather.isd[, .(ds, temperature.x, humidity.x, dew.point.x, pressure.x, wind.speed.mean.x, wind.bearing.mean.x, wind.speed.max.x)]))
na_names <- sapply(colnames(weather.isd), function(x) paste0(x, '.na'))
weather.isd[, as.vector(na_names) := lapply(.SD, function(x) as.integer(is.na(x))), .SDcols = 1:length(weather.isd)]
######################################################################################################################################################
# 8. Fill in NAs by simple linear interpolation
# Limit to 6 hours or 12 consecutive observations
weather.isd.orig <- weather.isd
weather.isd$temperature.x <- as.numeric(weather.isd$temperature.x)
weather.isd$humidity.x <- as.numeric(weather.isd$humidity.x)
weather.isd$dew.point.x <- as.numeric(weather.isd$dew.point.x)
weather.isd$pressure.x <- as.numeric(weather.isd$pressure.x)
weather.isd$wind.speed.mean.x <- as.numeric(weather.isd$wind.speed.mean.x)
weather.isd$wind.speed.max.x <- as.numeric(weather.isd$wind.speed.max.x)
weather.isd$wind.bearing.mean.x <- as.numeric(weather.isd$wind.bearing.mean.x)
interpolate_nas <- function(df, cols) {
df$row <- 1:nrow(df)
df$lin_interp <- 0
for (col in cols) {
print(col)
df[, rle := rleid(get(col))][,num_missing := max(.N + 1 , 2), by=rle]
yy <- df[num_missing <= 13, ..col][[1]]
xx <- df[num_missing <= 13, row]
nas <- df[num_missing <= 13 & is.na(get(col)), row]
df[nas, col] <- approx(xx, yy, xout=nas)$y
df[nas, 'lin_interp'] <- 1
df[nas, ]
}
summary(df)
df
df[nas,]
df$row <- NULL
df$rle <- NULL
df$num_missing <- NULL
return(df)
}
cols.1 <- c("temperature.x", "humidity.x", "dew.point.x", "pressure.x", "wind.speed.mean.x", "wind.bearing.mean.x", "wind.speed.max.x")
weather.isd.interp.1 <- interpolate_nas(weather.isd, cols.1)
cols.2 <- c("temperature.y", "humidity.y", "dew.point.y", "wind.speed.mean.y", "wind.bearing.mean.y", "wind.speed.max.y")
weather.isd.interp.2 <- interpolate_nas(weather.isd.interp.1, cols.2)
######################################################################################################################################################
# 9. Fill Computer Lab (weather.isd) NAs with Cambridge Airport (isd.filled) values
weather.isd.interp.2$isd_filled <- 0
weather.isd.interp.2[is.na(temperature.x) & !is.na(temperature.y), 'isd_filled'] <- 1
weather.isd.interp.2[is.na(humidity.x) & !is.na(humidity.y), 'isd_filled'] <- 1
weather.isd.interp.2[is.na(dew.point.x) & !is.na(dew.point.y), 'isd_filled'] <- 1
weather.isd.interp.2[is.na(wind.speed.mean.x) & !is.na(wind.speed.mean.y), 'isd_filled'] <- 1
weather.isd.interp.2[is.na(wind.speed.max.x) & !is.na(wind.speed.max.y), 'isd_filled'] <- 1
weather.isd.interp.2[is.na(wind.bearing.mean.x) & !is.na(wind.bearing.mean.y), 'isd_filled'] <- 1
weather.isd.filled <- weather.isd.interp.2[,
.(ds,
temperature=ifelse(is.na(temperature.x) & !is.na(temperature.y), temperature.y, temperature.x),
humidity=ifelse(is.na(humidity.x) & !is.na(humidity.y), humidity.y, humidity.x),
pressure=pressure.x,
dew.point=ifelse(is.na(dew.point.x) & !is.na(dew.point.y), dew.point.y, dew.point.x),
wind.speed.mean=ifelse(is.na(wind.speed.mean.x) & !is.na(wind.speed.mean.y), wind.speed.mean.y, wind.speed.mean.x),
wind.speed.max=ifelse(is.na(wind.speed.max.x) & !is.na(wind.speed.max.y), wind.speed.max.y, wind.speed.max.x),
wind.bearing.mean=ifelse(is.na(wind.bearing.mean.x) & !is.na(wind.bearing.mean.y), wind.bearing.mean.y, wind.bearing.mean.x),
rainfall,
sunshine,
ceil_hgt,
visibility,
missing,
known_inaccuracy,
isd_outlier,
long_run,
spike,
cooksd_out,
isd_3_sigma,
isd_filled)]
summary(weather.isd.filled) # lot of pressure NAs :-(
summary(weather.isd.filled[is.na(wind.speed.mean)])
summary(weather.isd.filled[is.na(wind.speed.max)])
summary(weather.isd.filled[is.na(wind.bearing.mean)])
summary(weather.isd.filled[is.na(temperature)])
cor(weather.isd.interp.2[complete.cases(weather.isd.interp.2[, .(temperature.x, temperature.y)]), .(temperature.x, temperature.y)])
# temperature.x temperature.y
# temperature.x 1.0000000 0.9783566
# temperature.y 0.9783566 1.0000000
cor(weather.isd.interp.2[complete.cases(weather.isd.interp.2[, .(dew.point.x, dew.point.y)]), .(dew.point.x, dew.point.y)])
# dew.point.x dew.point.y
# dew.point.x 1.0000000 0.9500148
# dew.point.y 0.9500148 1.0000000
# Increase over 0.9117574
cor(weather.isd.interp.2[complete.cases(weather.isd.interp.2[, .(humidity.x, humidity.y)]), .(humidity.x, humidity.y)])
# humidity.x humidity.y
# humidity.x 1.0000000 0.9064502
# humidity.y 0.9064502 1.0000000
cor(weather.isd.interp.2[complete.cases(weather.isd.interp.2[, .(wind.speed.mean.x, wind.speed.mean.y)]), .(wind.speed.mean.x, wind.speed.mean.y)])
# wind.speed.mean.x wind.speed.mean.y
# wind.speed.mean.x 1.0000000 0.8689842
# wind.speed.mean.y 0.8689842 1.0000000
# Increase over 0.7881784
cor(weather.isd.interp.2[complete.cases(weather.isd.interp.2[, .(wind.speed.max.x, wind.speed.max.y)]), .(wind.speed.max.x, wind.speed.max.y)])
cor(weather.isd.interp.2[complete.cases(weather.isd.interp.2[, .(wind.bearing.mean.x, wind.bearing.mean.y)]), .(wind.bearing.mean.x, wind.bearing.mean.y)])
# wind.bearing.mean.x wind.bearing.mean.y
# wind.bearing.mean.x 1.0000000 0.6827114
# wind.bearing.mean.y 0.6827114 1.0000000
######################################################################################################################################################
# 10. Fill in remaining missing values with historical averages
get_historical_average <- function(df, varcol, tscol, yearval=2019) {
df[, year:=as.numeric(strftime(get(tscol), format = "%Y"))]
df[, doy:=as.numeric(strftime(get(tscol), format = "%j"))]
df[, HMS:=strftime(get(tscol), format = "%H:%M:%S")]
df[, secs:=as.numeric(as.difftime(HMS))]
df.not.year <- unique(df[year != yearval | doy != 366, mean.var:=mean(get(varcol), na.rm=TRUE), by=.(doy, secs)][, .(doy, secs, mean.var)])
df.not.year <- df.not.year[doy != 366 & !is.na(mean.var),]
df.year <- df[year == yearval, .(doy, secs, get(varcol))]
df.year <- df.year[doy != 366 & !is.na(V3),]
setkey(df.not.year, doy, secs)
setkey(df.year, doy, secs)
rmse <- function(obs, pred) sqrt(mean((obs - pred)^2, na.rm=TRUE))
print(cor(df.not.year[df.year, .(mean.var, V3)]))
flush.console()
# rmse values ignore missing values so will be overestimates
cat(paste0("\nrmse (historical): ",
round(rmse(df.not.year[df.year, mean.var], df.not.year[df.year, V3]), 2),
"\n"))
mean.var <- mean(df[, get(varcol)], na.rm=TRUE)
cat(paste0("rmse (mean):\t ",
round(rmse(df[!is.na(get(varcol)), get(varcol)], jitter(rep(mean.var, nrow(df[!is.na(get(varcol)),])))), 2),
"\n"))
flush.console()
historical <- df[, mean.var:=mean(get(varcol), na.rm=TRUE), by=.(doy, secs)][, .(doy, secs, mean.var)]
historical <- unique(historical[!is.na(mean.var),])
setkey(historical, doy, secs)
return(historical)
}
##historical.temp <- get_historical_average(weather.isd.filled, 'temperature', 'ds')
## mean.var V3
## mean.var 1.0000000 0.8801195
## V3 0.8801195 1.0000000
##
## rmse (historical): 30.79
## rmse (mean): 65.25
#historical.rh <- get_historical_average(weather.isd.filled, 'humidity', 'ds')
## mean.var V3
## mean.var 1.0000000 0.7671816
## V3 0.7671816 1.0000000
##
## rmse (historical): 11.91
## rmse (mean): 17.28
#historical.dp <- get_historical_average(weather.isd.filled, 'dew.point', 'ds')
## mean.var V3
## mean.var 1.0000000 0.7608672
## V3 0.7608672 1.0000000
##
## rmse (historical): 33.03
## rmse (mean): 51.7
#historical.pres <- get_historical_average(weather.isd.filled, 'pressure', 'ds')
## mean.var V3
## mean.var 1.0000000 0.5405856
## V3 0.5405856 1.0000000
##
## rmse (historical): 12.1
## rmse (mean): 16.77
#historical.ws <- get_historical_average(weather.isd.filled, 'wind.speed.mean', 'ds')
## mean.var V3
## mean.var 1.0000000 0.4838662
## V3 0.4838662 1.0000000
##
## rmse (historical): 35.7
## rmse (mean): 40.31
#historical.ws <- get_historical_average(weather.isd.filled, 'wind.speed.max', 'ds')
#
#historical.wd <- get_historical_average(weather.isd.filled, 'wind.bearing.mean', 'ds')
## mean.var V3
## mean.var 1.0000000 0.2966302
## V3 0.2966302 1.0000000
##
## rmse (historical): 77.57
## rmse (mean): 83.31
#
## Despite the low correlation values for some variables they are all better than raw mean values.
## Historical values for pressure, wind.speed.mean and wind.bearing.mean give better results than
## multiple imputation values below. Not an ideal substition :-(
#weather.hist.filled <- weather.isd.filled
#summary(weather.hist.filled)
#weather.hist.filled[, doy:=as.numeric(strftime(ds, format = "%j"))]
#weather.hist.filled[, HMS:=strftime(ds, format = "%H:%M:%S")]
#weather.hist.filled[, secs:=as.numeric(as.difftime(HMS))]
#setkey(weather.hist.filled, doy, secs)
#
#weather.hist.filled$hist_average <- 0
#weather.hist.filled[is.na(wind.speed.mean), 'hist_average'] <- 1
#weather.hist.filled[is.na(wind.speed.max), 'hist_average'] <- 1
#weather.hist.filled[is.na(wind.bearing.mean), 'hist_average'] <- 1
#weather.hist.filled[is.na(pressure), 'hist_average'] <- 1
#
#weather.hist.filled[is.na(wind.speed.mean), 'wind.speed.mean'] <- historical.ws[weather.hist.filled[is.na(wind.speed.mean)], mean.var]
#weather.hist.filled[is.na(wind.speed.max), 'wind.speed.max'] <- historical.ws[weather.hist.filled[is.na(wind.speed.max)], mean.var]
#weather.hist.filled[is.na(wind.bearing.mean), 'wind.bearing.mean'] <- historical.wd[weather.hist.filled[is.na(wind.bearing.mean)], mean.var]
#weather.hist.filled[is.na(pressure), 'pressure'] <- historical.pres[weather.hist.filled[is.na(pressure)], mean.var]
#weather.hist.filled$doy <- NULL
#weather.hist.filled$HMS <- NULL
#weather.hist.filled$year <- NULL
#weather.hist.filled$secs <- NULL
#weather.hist.filled$mean.var <- NULL
#setkey(weather.hist.filled, ds)
#summary(weather.hist.filled)
######################################################################################################################################################
# 11. Multiple imputation for remaining temperature, humidity and dew.point NAs
# Using Hmisc library which FWIW is included in base R install
# I suspect imputation suffers when NAs co-occur across variables
#library(Hmisc)
## Very slow - approx. 10 mins :-(
#impute_arg_tlin <- aregImpute(~ temperature + humidity + pressure + dew.point + wind.speed.mean + wind.bearing.mean + wind.speed.max,
# data=weather.hist.filled, n.impute = 20, tlinear=FALSE)
#impute_arg_tlin
## aregImpute(formula = ~temperature + humidity + pressure + dew.point +
## wind.speed.mean + wind.bearing.mean, data = weather.hist.filled,
## n.impute = 20, tlinear = FALSE)
##
## n: 220799 p: 6 Imputations: 20 nk: 3
##
## Number of NAs:
## temperature humidity pressure dew.point
## 899 736 0 268
## wind.speed.mean wind.bearing.mean
## 0 0
##
## type d.f.
## temperature s 2
## humidity s 2
## pressure s 2
## dew.point s 2
## wind.speed.mean s 2
## wind.bearing.mean s 2
##
## R-squares for Predicting Non-Missing Values for Each Variable
## Using Last Imputations of Predictors
## temperature humidity dew.point
## 0.970 0.922 0.958
#
#weather.mi.filled <- weather.hist.filled
#summary(weather.mi.filled)
#
#weather.mi.filled$mi_filled <- 0
#weather.mi.filled[is.na(temperature), 'mi_filled'] <- 1
#weather.mi.filled[is.na(humidity), 'mi_filled'] <- 1
#weather.mi.filled[is.na(dew.point), 'mi_filled'] <- 1
#
#weather.mi.filled[is.na(temperature), 'temperature'] <- rowMeans(impute_arg_tlin$imputed$temperature)
#weather.mi.filled[is.na(humidity), 'humidity'] <- rowMeans(impute_arg_tlin$imputed$humidity)
#weather.mi.filled[is.na(dew.point), 'dew.point'] <- rowMeans(impute_arg_tlin$imputed$dew.point)
#summary(weather.mi.filled)
#
#
## Adding imputed and historical values introduces a few large values which are
## straight-forward to correct
#wf.temp.spikes <- get_large_spikes(weather.mi.filled, 'temperature', 'ds', 6)
#wf.rh.spikes <- get_large_spikes(weather.mi.filled, 'humidity', 'ds', 6)
#wf.pres.spikes <- get_large_spikes(weather.mi.filled, 'pressure', 'ds', 6)
#wf.dp.spikes <- get_large_spikes(weather.mi.filled, 'dew.point', 'ds', 6)
#wf.ws.spikes <- get_large_spikes(weather.mi.filled, 'wind.speed.mean', 'ds', 5) # No action
#wf.ws.spikes <- get_large_spikes(weather.mi.filled, 'wind.speed.max', 'ds', 5) # No action
#
#weather.mi.filled$mi_spike_interp <- 0
#weather.mi.filled[wf.temp.spikes, 'mi_spike_interp'] <- 1
#weather.mi.filled[wf.rh.spikes, 'mi_spike_interp'] <- 1
#weather.mi.filled[wf.pres.spikes, 'mi_spike_interp'] <- 1
#weather.mi.filled[wf.dp.spikes, 'mi_spike_interp'] <- 1
#
#weather.mi.filled[wf.temp.spikes, 'temperature'] <- NA
#weather.mi.filled[wf.rh.spikes, 'humidity'] <- NA
#weather.mi.filled[wf.pres.spikes, 'pressure'] <- NA
#weather.mi.filled[wf.dp.spikes, 'dew.point'] <- NA
#summary(weather.mi.filled)
#
#cols.3 <- c("temperature", "humidity", "dew.point", "pressure")
#weather.mi.filled.interp <- interpolate_nas(weather.mi.filled, cols.3)
#weather.mi.filled.interp <- unique(weather.mi.filled.interp)
#summary(weather.mi.filled.interp)
######################################################################################################################################################
# 12. Replace NAs using tsclean from forecast package
# https://robjhyndman.com/hyndsight/tsoutliers/
# Before NAs
# tsclean - temperature, dew.point, humidity, pressure, wind.speed.mean, wind.speed.max, rainfall, sunshine
# After NAs
# Before and after plots
library(forecast)
weather.tsclean.filled <- weather.isd.filled
weather.tsclean.filled$tsclean_filled <- 0
summary(weather.tsclean.filled)
# for (var in c('temperature', 'dew.point')) {
for (var in c('temperature', 'dew.point', 'humidity', 'pressure')) {
print(var)
var_flag <- paste0('tsclean_filled_', var)
weather.tsclean.filled[[var_flag]] <- 0
weather.tsclean.filled[is.na(get(var)), var_flag] <- 1
weather.tsclean.filled[is.na(get(var)), 'tsclean_filled'] <- 1
var_ts <- ts(weather.tsclean.filled[[var]], frequency=48)
# var_clean <- tsclean(var_ts)
var_clean <- na.interp(var_ts)
weather.tsclean.filled[[var]] <- var_clean
}
weather.tsclean.filled[humidity < 0, 'humidity'] <- 0
weather.tsclean.filled[humidity > 100, 'humidity'] <- 100
weather.tsclean.filled[pressure < 960, 'pressure'] <- 960
weather.tsclean.filled[pressure > 1060, 'pressure'] <- 1060
weather.tsclean.filled[dew.point > temperature, 'dew.point'] <- weather.tsclean.filled[dew.point > temperature, 'temperature']
# Features with some long term seasonality but much less short term seasonality
# So no frequency specification in ts() call
for (var in c('wind.speed.mean', 'wind.speed.max', 'wind.bearing.mean', 'rainfall')) {
print(var)
var_flag <- paste0('tsclean_filled_', var)
weather.tsclean.filled[[var_flag]] <- 0
weather.tsclean.filled[is.na(get(var)), var_flag] <- 1
weather.tsclean.filled[is.na(get(var)), 'tsclean_filled'] <- 1
var_ts <- ts(weather.tsclean.filled[[var]])
# var_clean <- tsclean(var_ts)
var_clean <- na.interp(var_ts)
weather.tsclean.filled[[var]] <- var_clean
weather.tsclean.filled[get(var) < 0, var] <- 0
}
weather.tsclean.filled[wind.bearing.mean > 360, 'wind.bearing.mean'] <- 360
weather.tsclean.filled[wind.speed.mean > wind.speed.max, 'wind.speed.max'] <- weather.tsclean.filled[wind.speed.mean > wind.speed.max, 'wind.speed.mean']
summary(weather.tsclean.filled)
plot_tsclean <- function(before, after, var) {
autoplot(ts(after[[var]], frequency=48), series="clean", color='red', lwd=1) +
autolayer(ts(before[[var]], frequency=48), series="original", color='gray', lwd=1)
}
# var <- 'temperature'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
# var <- 'dew.point'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
# var <- 'humidity'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
# var <- 'pressure'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
# var <- 'wind.speed.mean'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
# var <- 'wind.speed.max'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
# var <- 'rainfall'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
# var <- 'wind.bearing.mean'; plot_tsclean(weather.isd.filled, weather.tsclean.filled, var)
######################################################################################################################################################
# 13. Save data
fnRDS <- paste0("data/CamMetCleanish", format(Sys.time(), "%Y.%m.%d"), ".RData")
fnCSV <- paste0("data/CamMetCleanish", format(Sys.time(), "%Y.%m.%d"), ".csv.xz")
fnRData <- paste0("data/CambridgeTemperatureModel", format(Sys.time(), "%Y.%m.%d"), ".RData")
fnXCSV <- paste0("data/CamMetCleanishMissAnnotated", format(Sys.time(), "%Y.%m.%d"), ".csv.xz")
saveRDS(weather.tsclean.filled[, .(temperature,
dew.point,
humidity,
pressure,
wind.speed.mean,
wind.speed.max,
wind.bearing.mean,
rainfall,
ceil_hgt,
visibility,
ds)],
fnRDS)
write.csv(weather.tsclean.filled[, .(ds,
y=temperature,
humidity,
dew.point,
pressure,