-
Notifications
You must be signed in to change notification settings - Fork 36
/
TrendSeasonalFit_v12_30Line.m
871 lines (783 loc) · 43.7 KB
/
TrendSeasonalFit_v12_30Line.m
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
function rec_cg = TrendSeasonalFit_v12_30Line(dir_l,n_rst,ncols,nrows,T_cg,Tmax_cg,conse,num_c,nbands,B_detect)
% CCDC 12.30 version - Zhe Zhu, EROS USGS
% It is based on 7 bands fitting for Iterative Seasonal, Linear, and Break Models
% This function works for analyzing one line of time series pixel
%
%% Revisions: $ Date: 11/20/2015 $ Copyright: Zhe Zhu
% Version 12.30 Fixed a bug for pixels without minimum observations (11/20/2015)
% Version 12.29 Modified fit for perennial snow and Fmask failed pixels (09/20/2015)
% Version 12.29 Do not fit disturbed time period (09/18/2015)
% Version 12.28 Fixed a bug for missing values in land cover maps (09/16/2015)
% Version 12.27 Fixed bugs for persistent snow and falied Fmask pixels (06/17/2015)
% Version 12.26 Connected time for all models (05/28/2015)
% Version 12.25 Bug fixed in snow percent (05/19/2015)
% Version 12.24 Change T_const in Tmask (03/31/2015)
% Version 12.23 Update iteratively before 24 observations (03/22/2015)
% Version 12.22 Adjust mini RMSE based on temporal variability (03/22/2015)
% Version 12.21 Add more categories and update i_start in the end (03/14/2015)
% Version 12.20 Convert BT to from K to C before analysis (03/12/2015)
% Version 12.19 Fit for permanent snow if is more than 75% (03/12/2015)
% Version 12.18 No change detection if clear observation less than 25% (03/12/2015)
% Version 12.17 Use median value for very simple model & change magnitude (02/24/2015)
% Version 12.16 Finding changes in all water pixels (02/24/2015)
% Version 12.15 Use the original multitemporal cloud mask (02/15/2015)
% Version 12.14 Do not need example_img in images folder (02/09/2015)
% Version 12.13: More infromation in "category" (11/10/2014)
% This version (12.13) is used for the third round of the LCMS project.
% Command: TrendSeasonalFit_v12Plot(N_row,N_col,min=0.5,T_cg=0.99,n_times=3,conse=6,B_detect=2:6)
% Version 12.12: Fit for pixels where Fmask fails (11/09/2014)
% Version 12.11: Bug fixed in num_fc (11/09/2014)
% Version 12.10: Better multietmporal cloud detection at the beginning (11/06/2014)
% Version 12.9: Detect change only for land pixel (water/snow speical case) (10/31/2014)
% Version 12.8: Speed up by reducing time for RMSE and model computing (10/17/2014)
% Version 12.7: mini rmse should be larger than 10% of the mean (10/13/2014)
% Version 12.6: Fit model again when there are a 33.3% more data (10/08/2014)
% Version 12.5: Use subset of bands (2-6) for detecting surface change (10/01/2014)
% Version 12.4: Only apply multitemporal cloud masking during model initialization (09/29/2014)
% Version 12.3: Use subset of bands (3-5) to balance change in diferent dimensions (09/01/2014)
% This version (12.3) is used for the second round of the LCMS project.
% Command: TrendSeasonalFit_v12Plot(N_row,N_col,min=1,T_cg=0.99,n_times=3,conse=5,B_detect=3:6)
% Version 12.2: Bug fixed in model intialization (08/14/2014)
% Version 12.1: Use subset of bands (3-6) to avoid atmosphere influences (08/04/2014)
%% Version 12.0 Detecting change based on probability (07/19/2014)
% Version 11.6: No need to change folder name & faster in speed (by Christ Holden 06/06/2014)
% Version 11.5: Improved calculation of temporally adjusted RMSE (04/23/2014)
% Version 11.4: Revise "rec_cg.category" to better seperate different fit processes (04/01/2014)
% This version (11.4) is used for generating synthetic data for ACRE project and
% detecting change for LCMS project.
% Command: TrendSeasonalFit_v11Plot(N_row,N_col,min=1,T_cg=2,n_times=3,conse=6,B_detect=1:6)
% Version 11.3: Add "rec_cg.magnitude" as indicator of change magnitude (04/01/2014)
% Version 11.2: Change very simple fit with mean value for start and end of timeseries (04/01/2014)
% Version 11.1: Do not need metadata in the image folder to run CCDC (03/25/2014)
%% Version 11.0: Use change vector magnitude as threshold for detecting change (03/25/2014)
% Version 10.13: Use subset of bands (1-6) to avoid atmosphere influences (01/31/2014)
% Version 10.12: More accurate number of days per year "num_yrs" (01/30/2014)
% Version 10.11: RMSE updates with time series fit (01/26/2014)
% Version 10.10: Update temperature extreme in recent studies (01/16/2014)
% Version 10.9: Find break in max value in any of the band (01/08/2014)
% Version 10.8: Add very simple fit with median value for start and end of timeseries (10/21/2013)
% This version (10.8) is used for generating synthetic data for the LCMS project.
% Command: TrendSeasonalFit_v10Plot('stack',N_row,N_col,mini=0.5,T_cg=3,n_times=3,conse=6,B_detect=2:6)
% Version 10.7: Better multitemporal cloud detection (10/19/2013)
% Version 10.6: Add "Tmax_cg" for last step noise removal (10/18/2013)
% Version 10.5: Use subset of bands (2-6) to avoid atmosphere influences (10/18/2013)
% Version 10.4: Let dynamic fitting for pixels at the beginning (09/23/2013)
% Version 10.3: Able to detect change at the verying beginning (09/06/2013)
% Version 10.2: Add mini years "mini_yrs" in model intialization (09/03/2013)
% Version 10.1: Reduce time for calcuating "v_dif" (09/02/2013)
%% Version 10.0: Fit for beginning and end of the time series (08/31/2013)
% Version 9.9: Only fit more than 50% of Landat images overlap area (08/28/2013)
% Version 9.8: Force model fit for persistent snow pixels (08/27/2013)
% Version 9.7: Add "rec_cg.category" as indicator of fitting procudure (08/20/2013)
% Add rec_cg.change_prob as indicator of change probability (08/20/2013)
% Add rec_cg.num_obs ad indicator of number of observations (08/20/2013)
% Version 9.6: Remove mininum rmse "mini" and minimum years "mini_yrs" (08/16/2013)
% Version 9.5: Model gets more coefficients with more observations (08/16/2013)
% Version 9.4: Bug fixed in calculating temporally adjusted rmse (08/01/2013)
% Version 9.3: Fit curve again after one year (03/28/2013)
% This version (9.3) is used for mapping land cover for the IDS project.
% Command: TrendSeasonalFit_v9Plot('stack',N_row,N_col,T_cg=2,n_times=3,conse=4)
% Version 9.2: Use "mini = T_const/T_cg" for small rmse cases (03/26/2013)
% Version 9.1: Remove out of range pixels before time series analysis (02/09/2013)
%% Version 9.0: Using 8 coefficients and lasso fit (02/01/2013)
% Version 8.4: Use "max v_slope" instead of "average v_slope" (01/16/2013)
% Version 8.3: Start initialization when "time_span" > 1 year (01/16/2013)
% Version 8.2: Bug fixed in not fitting models at the begining (01/16/2013)
% Version 8.1: Bug fixed in counting "i" and "i_span"(01/13/2013)
%% Version 8.0: Temporally changing RMSE (01/09/2013)
%% Version 7.3: Continuous Change Detection and Classification (CCDC) (07/11/2012)
% This version (7.3) is explained by Zhu, Z. & Woodcock, C.E., Continuous Change
% Detection and Classification (CCDC) of land cover using all available
% Landsat data, Remote Sensing of Environment (2014).
% Command: TrendSeasonalFit_v7Plot('stack',N_row,N_col,T_cg=3,n_times=3,conse=3)
%% Version 1.0: Continous Monitoring of Forest Disturbance Algorithm (CMFDA) (07/13/2010)
% This version (1.0) is explained by Zhu, Z., Woodcock, C.E., Olofsson, P.,
% Continuous monitoring of forest disturbance using all available Landsat
% data, Remote Sensing of Environment (2012).
%
%% Inputs:
% stk_n='stack'; stack image name
% ncols = 8021; % number of pixels processed per line
% nrows=1; % the nrowsth lines
% for example 1 2 3 4 5
% 6 7 8 9 10
%
%% Outputs:
%
% rec_cg RECord information about all curves between ChanGes
% rec_cg(i).t_start record the start of the ith curve fitting (julian_date)
% rec_cg(i).t_end record the end of the ith curve fitting (julian_date)
% rec_cg(i).t_break record the first observed break time (julian_date)
% rec_cg(i).coefs record the coefficients of the ith curve
% rec_cg(i).pos record the position of the ith pixel (pixel id)
% rec_cg(i).magnitude record the change vector of all spectral bands
% rec_cg(i).category record what fitting procudure and model is used
% cateogry category 5x: persistent snow 4x: Fmask fails
% cateogry category 3x: modified fit 2x: end fit
% category category 1x: start fit x: normal procedure
% cateogry category x1: mean value x4: simple model
% category category x6: advanced model x8: full model
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% defining variables
%% Constants
% maximum number of coefficient required
% 2 for tri-modal; 2 for bi-modal; 2 for seasonality; 2 for linear;
min_num_c = 4;
mid_num_c = 6;
max_num_c = 8;
% number of clear observation / number of coefficients
n_times = 3;
% initialize NUM of Functional Curves
num_fc = 0;
% number of days per year
num_yrs = 365.25;
% number of bytes: int16
num_byte = 2;
% Band for multitemporal cloud/snow detection (Green)
num_B1 = 2;
% Band for multitemporal shadow/snow shadow detection (SWIR)
num_B2 = 5;
% Threshold for cloud, shadow, and snow detection.
T_const = 3.89;
% minimum year for model intialization
mini_yrs = 1;
% no change detection for permanent snow pixels
t_sn = 0.75;
% Fmask fails threshold
t_clr = 0.25;
% get num of total folders start with "L"
imf = dir(fullfile(dir_l,'L*')); % folder names
% filter for Landsat folders
imf = regexpi({imf.name}, 'L(T5|T4|E7|C8|ND)(\w*)', 'match');
imf = [imf{:}];
imf = vertcat(imf{:});
% sort according to yeardoy
yeardoy = str2num(imf(:, 10:16));
[~, sort_order] = sort(yeardoy);
imf = imf(sort_order, :);
% number of folders start with "L"
num_t = size(imf,1);
% initialize the struct data of RECording of ChanGe (rec_cg)
rec_cg = struct('t_start',[],'t_end',[],'t_break',[],'coefs',[],'rmse',[],...
'pos',[],'change_prob',[],'num_obs',[],'category',[],'magnitude',[]);
% % mask for study area (1 fit, 0 no fit)
% fit_mask = enviread('GZ_Mask');
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Get ready for Xs & Ys
%% Read in Xs & Ysq
% transforming to serial date number (0000 year)
sdate = zeros(num_t,1); % Xs
line_t = zeros(num_t,nbands*ncols); %Ys
for i=1:num_t
im_dir = dir(fullfile(dir_l,imf(i, :)));
im = '';
for f = 1:size(im_dir, 1)
% use regular expression to match:
% 'L(\w*)' Any word begining with L that has any following chars
% stk_n includes stack name somewhere after L
% '$' ends with the stack name (e.g., no .hdr, .aux.xml)
if regexp(im_dir(f).name, ['L(\w*)', 'stack', '$']) == 1
im = fullfile(dir_l,imf(i, :), im_dir(f).name);
break
end
end
% Check to make sure we found something
if strcmp(im, '')
error('Could not find stack image for directory %s\n', imf(i));
end
% Find date for folder imf(i)
yr = str2num(imf(i, 10:13));
doy = str2num(imf(i, 14:16));
sdate(i) = datenum(yr, 1, 0) + doy;
dummy_name = im;
fid_t = fopen(dummy_name,'r'); % get file ids
fseek(fid_t,num_byte*(nrows-1)*ncols*nbands,'bof');
line_t(i,:) = fread(fid_t,nbands*ncols,'int16=>double','ieee-le'); % get Ys
end
fclose('all'); % close all files
for i_ids = 1:ncols
% mask data
line_m = line_t(:,nbands*i_ids);
% Only run CCDC for places where more than 50% of images has data
idexist = line_m < 255;
overlap_pct = sum(idexist)/num_t;
if overlap_pct < 0.5
continue;
end
% convert Kelvin to Celsius
line_t(:,nbands*(i_ids-1)+7) = line_t(:,nbands*(i_ids-1)+7)*10 - 27315;
% clear pixel should have reflectance between 0 and 1
% brightness temperature should between -93.2 to 70.7 celsius degree
idrange = line_t(:,nbands*(i_ids-1)+1)>0&line_t(:,nbands*(i_ids-1)+1)<10000&...
line_t(:,nbands*(i_ids-1)+2)>0&line_t(:,nbands*(i_ids-1)+2)<10000&...
line_t(:,nbands*(i_ids-1)+3)>0&line_t(:,nbands*(i_ids-1)+3)<10000&...
line_t(:,nbands*(i_ids-1)+4)>0&line_t(:,nbands*(i_ids-1)+4)<10000&...
line_t(:,nbands*(i_ids-1)+5)>0&line_t(:,nbands*(i_ids-1)+5)<10000&...
line_t(:,nbands*(i_ids-1)+6)>0&line_t(:,nbands*(i_ids-1)+6)<10000&...
line_t(:,nbands*(i_ids-1)+7)>-9320&line_t(:,nbands*(i_ids-1)+7)<7070;
% # of clear observatons
idclr = line_m < 2;
% # of all available observations
idall = line_m < 255;
% clear observation percentage
clr_pct = sum(idclr)/sum(idall);
% snow pixels
idsn = line_m == 3;
% percent of snow observations
sn_pct = sum(idsn)/(sum(idclr)+sum(idsn)+0.01);
% not enough clear observations for change detection
if clr_pct < t_clr
% permanent snow pixels
if sn_pct > t_sn
% snow observations are "good" now
idgood = idsn|idclr;
% number of snow pixel within range
n_sn = sum(idgood);
if n_sn < n_times*min_num_c % not enough snow pixels
continue
else
% Xs & Ys for computation
clrx = sdate(idgood);
% bands 1-5,7,6
clry = line_t(idgood,(nbands*(i_ids-1)+1):(nbands*(i_ids-1)+nbands-1));
% the first observation for TSFit
i_start = 1;
% identified and move on for the next curve
num_fc = num_fc + 1; % NUM of Fitted Curves (num_fc)
% defining computed variables
fit_cft = zeros(max_num_c,nbands-1);
% rmse for each band
rmse = zeros(nbands-1,1);
% snow qa = 50
qa = 50;
for i_B=1:nbands-1
if i_B ~= nbands-1 % treat saturated and unsaturated pixels differently
idgood = clry(:,i_B) < 10000; % saturate if ref > 1 or NBR NDVR > 1
i_span = sum(idgood);
if i_span < min_num_c*n_times % fill value for frequently saturated snow pixels
fit_cft(1,i_B) = 10000; % give constant value
else % fit for enough unsaturat snow pixels
[fit_cft(:,i_B),rmse(i_B)]=autoTSFit(clrx(idgood),clry(idgood,i_B),min_num_c);
end
else % fit for temperature band
idgood = clry(:,i_B)>-9320&clry(:,i_B)<7070;
[fit_cft(:,i_B),rmse(i_B)]=autoTSFit(clrx(idgood),clry(idgood,i_B),min_num_c);
end
end
% updating information at each iteration
% record time of curve start
rec_cg(num_fc).t_start=clrx(i_start);
% record time of curve end
rec_cg(num_fc).t_end=clrx(end);
% record break time
rec_cg(num_fc).t_break = 0; % no break at the moment
% record postion of the pixel
rec_cg(num_fc).pos = (nrows-1)*ncols+i_ids;
% record fitted coefficients
rec_cg(num_fc).coefs = fit_cft;
% record rmse of the pixel
rec_cg(num_fc).rmse = rmse;
% record change probability
rec_cg(num_fc).change_prob = 0;
% record number of observations
rec_cg(num_fc).num_obs = n_sn;
% record fit category
rec_cg(num_fc).category = qa + min_num_c;
% record change magnitude
rec_cg(num_fc).magnitude = zeros(1,nbands-1);
end
else % no change detection for clear observations
% within physical range pixels
idgood = idrange;
% Xs & Ys for computation
clrx = sdate(idgood);
% bands 1-5,7,6
clry = line_t(idgood,(nbands*(i_ids-1)+1):(nbands*(i_ids-1)+nbands-1));
idclr = clry(:,num_B1) < median(clry(:,num_B1)) + 400;
n_clr = sum(idclr);
if n_clr < n_times*min_num_c % not enough clear pixels
continue
else
% Xs & Ys for computation
clrx = clrx(idclr);
clry = clry(idclr,:);
% the first observation for TSFit
i_start = 1;
% identified and move on for the next curve
num_fc = num_fc + 1; % NUM of Fitted Curves (num_fc)
% defining computed variables
fit_cft = zeros(max_num_c,nbands-1);
% rmse for each band
rmse = zeros(nbands-1,1);
% Fmask fail qa = 40
qa = 40;
for i_B = 1:nbands-1
% fit basic model for all within range snow pixels
[fit_cft(:,i_B),rmse(i_B)] = autoTSFit(clrx,clry(:,i_B),min_num_c);
end
% record time of curve start
rec_cg(num_fc).t_start = clrx(i_start);
% record time of curve end
rec_cg(num_fc).t_end = clrx(end);
% record break time
rec_cg(num_fc).t_break = 0;
% record postion of the pixel
rec_cg(num_fc).pos = (nrows-1)*ncols+i_ids;
% record fitted coefficients
rec_cg(num_fc).coefs = fit_cft;
% record rmse of the pixel
rec_cg(num_fc).rmse = rmse;
% record change probability
rec_cg(num_fc).change_prob = 0;
% record number of observations
rec_cg(num_fc).num_obs = length(clrx);
% record fit category
rec_cg(num_fc).category = qa + min_num_c;
% record change magnitude
rec_cg(num_fc).magnitude = zeros(1,nbands-1);
end
end
else % normal CCDC procedure
% clear and within physical range pixels
idgood = idclr & idrange;
% Xs & Ys for computation
clrx = sdate(idgood);
% bands 1-5,7,6
clry = line_t(idgood,(nbands*(i_ids-1)+1):(nbands*(i_ids-1)+nbands-1));
% caculate median variogram
var_clry = clry(2:end,:)-clry(1:end-1,:);
adj_rmse = median(abs(var_clry),1);
% start with the miminum requirement of clear obs
i = n_times*min_num_c;
% initializing variables
% the first observation for TSFit
i_start = 1;
% record the start of the model initialization (0=>initial;1=>done)
BL_train = 0;
% identified and move on for the next curve
num_fc = num_fc+1; % NUM of Fitted Curves (num_fc)
% record the num_fc at the beginning of each pixel
rec_fc = num_fc;
% while loop - process till the last clear observation - conse
while i<= length(clrx)-conse
% span of "i"
i_span = i-i_start+1;
% span of time (num of years)
time_span=(clrx(i)-clrx(i_start))/num_yrs;
% basic requrirements: 1) enough observations; 2) enough time
if i_span >= n_times*min_num_c && time_span >= mini_yrs
% initializing model
if BL_train == 0
% Tmask: noise removal (good => 0 & noise => 1)
blIDs = autoTmask(clrx(i_start:i+conse),clry(i_start:i+conse,[num_B1,num_B2]),...
(clrx(i+conse)-clrx(i_start))/num_yrs,adj_rmse(num_B1),adj_rmse(num_B2),T_const);
% IDs to be removed
IDs = i_start:i+conse;
rmIDs = IDs(blIDs(1:end-conse) == 1);
% update i_span after noise removal
i_span = sum(~blIDs(1:end-conse));
% check if there is enough observation
if i_span < n_times*min_num_c
% move forward to the i+1th clear observation
i = i+1;
% not enough clear observations
continue;
% check if there is enough time
else
% copy x & y
cpx = clrx;
cpy = clry;
% remove noise pixels between i_start & i
cpx(rmIDs) = [];
cpy(rmIDs,:) = [];
% record i before noise removal
% This is very important as if model is not initialized
% the multitemporal masking shall be done again instead
% of removing outliers in every masking
i_rec = i;
% update i afer noise removal (i_start stays the same)
i = i_start+i_span-1;
% update span of time (num of years)
time_span = (cpx(i)-cpx(i_start))/num_yrs;
% check if there is enough time
if time_span < mini_yrs
% keep the original i
i = i_rec;
% move forward to the i+1th clear observation
i = i+1;
% not enough time
continue;
% Step 2: model fitting
else
% remove noise
clrx = cpx;
clry = cpy;
% Step 2: model fitting
% initialize model testing variables
% defining computed variables
fit_cft = zeros(max_num_c,nbands-1);
% rmse for each band
rmse = zeros(nbands-1,1);
% value of differnce
v_dif = zeros(nbands-1,1);
% record the diference in all bands
rec_v_dif = zeros(i-i_start+1,nbands-1);
for i_B = 1:nbands-1
% initial model fit
[fit_cft(:,i_B),rmse(i_B),rec_v_dif(:,i_B)] = ...
autoTSFit(clrx(i_start:i),clry(i_start:i,i_B),min_num_c);
end
% normalized to z-score
for i_B = B_detect
% minimum rmse
mini_rmse = max(adj_rmse(i_B),rmse(i_B));
% compare the first clear obs
v_start = rec_v_dif(1,i_B)/mini_rmse;
% compare the last clear observation
v_end = rec_v_dif(end,i_B)/mini_rmse;
% anormalized slope values
v_slope = fit_cft(2,i_B)*(clrx(i)-clrx(i_start))/mini_rmse;
% differece in model intialization
v_dif(i_B) = abs(v_slope) + abs(v_start) + abs(v_end);
end
v_dif = norm(v_dif(B_detect))^2;
% find stable start for each curve
if v_dif > T_cg
% start from next clear obs
i_start = i_start + 1;
% move forward to the i+1th clear observation
i = i + 1;
% keep all data and move to the next obs
continue;
else
% model ready!
BL_train = 1;
% count difference of i for each iteration
i_count = 0;
% find the previous break point
if num_fc == rec_fc
% first curve
i_break = 1;
else
% after the first curve
i_break = find(clrx >= rec_cg(num_fc-1).t_break);
i_break = i_break(1);
end
if i_start > i_break
% model fit at the beginning of the time series
for i_ini = i_start-1:-1:i_break
if i_start - i_break < conse
ini_conse = i_start - i_break;
else
ini_conse = conse;
end
% value of difference for conse obs
v_dif = zeros(ini_conse,nbands-1);
% record the magnitude of change
v_dif_mag = v_dif;
% chagne vector magnitude
vec_mag = zeros(ini_conse,1);
for i_conse = 1:ini_conse
for i_B = 1:nbands-1
% absolute difference
v_dif_mag(i_conse,i_B) = clry(i_ini-i_conse+1,i_B)-autoTSPred(clrx(i_ini-i_conse+1),fit_cft(:,i_B));
% normalized to z-scores
if sum(i_B == B_detect)
% minimum rmse
mini_rmse = max(adj_rmse(i_B),rmse(i_B));
% z-scores
v_dif(i_conse,i_B) = v_dif_mag(i_conse,i_B)/mini_rmse;
end
end
vec_mag(i_conse) = norm(v_dif(i_conse,B_detect))^2;
end
if min(vec_mag) > T_cg % change detected
break
elseif vec_mag(1) > Tmax_cg % false change
% remove noise
clrx(i_ini) = [];
clry(i_ini,:) = [];
i=i-1; % stay & check again after noise removal
end
% update new_i_start if i_ini is not a confirmed break
i_start = i_ini;
end
end
if num_fc == rec_fc && i_start - i_break >= conse
% defining computed variables
fit_cft = zeros(max_num_c,nbands-1);
% rmse for each band
rmse = zeros(nbands-1,1);
% start fit qa = 10
qa = 10;
for i_B=1:nbands-1
[fit_cft(:,i_B),rmse(i_B)] = ...
autoTSFit(clrx(i_break:i_start-1),clry(i_break:i_start-1,i_B),min_num_c);
end
% record time of curve end
rec_cg(num_fc).t_end = clrx(i_start-1);
% record postion of the pixel
rec_cg(num_fc).pos = (nrows-1)*ncols + i_ids;
% record fitted coefficients
rec_cg(num_fc).coefs = fit_cft;
% record rmse of the pixel
rec_cg(num_fc).rmse = rmse;
% record break time
rec_cg(num_fc).t_break = clrx(i_start);
% record change probability
rec_cg(num_fc).change_prob = 1;
% record time of curve start
rec_cg(num_fc).t_start = clrx(1);
% record fit category
rec_cg(num_fc).category = qa + min_num_c;
% record number of observations
rec_cg(num_fc).num_obs = i_start - i_break;
% record change magnitude
rec_cg(num_fc).magnitude = - median(v_dif_mag,1);
% identified and move on for the next functional curve
num_fc = num_fc + 1;
end
end
end
end
end
% continuous monitoring started!!!
if BL_train == 1
% all IDs
IDs = i_start:i;
% span of "i"
i_span = i-i_start+1;
% determine the time series model
update_num_c = update_cft(i_span,n_times,min_num_c,mid_num_c,max_num_c,num_c);
% initial model fit when there are not many obs
if i_count == 0 || i_span <= max_num_c*n_times
% update i_count at each interation
i_count = clrx(i)-clrx(i_start);
% defining computed variables
fit_cft = zeros(max_num_c,nbands-1);
% rmse for each band
rmse = zeros(nbands-1,1);
% record the diference in all bands
rec_v_dif = zeros(length(IDs),nbands-1);
% normal fit qa = 0
qa = 0;
for i_B=1:nbands-1
[fit_cft(:,i_B),rmse(i_B),rec_v_dif(:,i_B)] = ...
autoTSFit(clrx(IDs),clry(IDs,i_B),update_num_c);
end
% updating information for the first iteration
% record time of curve start
rec_cg(num_fc).t_start = clrx(i_start);
% record time of curve end
rec_cg(num_fc).t_end = clrx(i);
% record break time
rec_cg(num_fc).t_break = 0; % no break at the moment
% record postion of the pixel
rec_cg(num_fc).pos = (nrows-1)*ncols+i_ids;
% record fitted coefficients
rec_cg(num_fc).coefs = fit_cft;
% record rmse of the pixel
rec_cg(num_fc).rmse = rmse;
% record change probability
rec_cg(num_fc).change_prob = 0;
% record number of observations
rec_cg(num_fc).num_obs = i-i_start+1;
% record fit category
rec_cg(num_fc).category = qa + update_num_c;
% record change magnitude
rec_cg(num_fc).magnitude = zeros(1,nbands-1);
% detect change
% value of difference for conse obs
v_dif = zeros(conse,nbands-1);
% record the magnitude of change
v_dif_mag = v_dif;
vec_mag = zeros(conse,1);
for i_conse = 1:conse
for i_B = 1:nbands-1
% absolute difference
v_dif_mag(i_conse,i_B) = clry(i+i_conse,i_B)-autoTSPred(clrx(i+i_conse),fit_cft(:,i_B));
% normalized to z-scores
if sum(i_B == B_detect)
% minimum rmse
mini_rmse = max(adj_rmse(i_B),rmse(i_B));
% z-scores
v_dif(i_conse,i_B) = v_dif_mag(i_conse,i_B)/mini_rmse;
end
end
vec_mag(i_conse) = norm(v_dif(i_conse,B_detect))^2;
end
% IDs that haven't updated
IDsOld = IDs;
else
if clrx(i)-clrx(i_start) >= 1.33*i_count
% update i_count at each interation
i_count = clrx(i)-clrx(i_start);
% defining computed variables
fit_cft = zeros(max_num_c,nbands-1);
% rmse for each band
rmse = zeros(nbands-1,1);
% record the diference in all bands
rec_v_dif = zeros(length(IDs),nbands-1);
% normal fit qa = 0
qa = 0;
for i_B=1:nbands-1
[fit_cft(:,i_B),rmse(i_B),rec_v_dif(:,i_B)] = ...
autoTSFit(clrx(IDs),clry(IDs,i_B),update_num_c);
end
% record fitted coefficients
rec_cg(num_fc).coefs = fit_cft;
% record rmse of the pixel
rec_cg(num_fc).rmse = rmse;
% record number of observations
rec_cg(num_fc).num_obs = i-i_start+1;
% record fit category
rec_cg(num_fc).category = qa + update_num_c;
% IDs that haven't updated
IDsOld = IDs;
end
% record time of curve end
rec_cg(num_fc).t_end = clrx(i);
% use fixed number for RMSE computing
n_rmse = n_times*rec_cg(num_fc).category;
tmpcg = zeros(nbands-1,1);
% better days counting for RMSE calculating
% relative days distance
d_rt = clrx(IDsOld) - clrx(i+conse);
d_yr = abs(round(d_rt/num_yrs)*num_yrs-d_rt);
[~,sorted_indx] = sort(d_yr);
sorted_indx = sorted_indx(1:n_rmse);
for i_B = B_detect
% temporally changing RMSE
tmpcg_rmse(i_B) = norm(rec_v_dif(IDsOld(sorted_indx)-IDsOld(1)+1,i_B))/...
sqrt(n_rmse-rec_cg(num_fc).category);
end
% move the ith col to i-1th col
v_dif(1:conse-1,:) = v_dif(2:conse,:);
% only compute the difference of last consecutive obs
v_dif(conse,:) = 0;
% move the ith col to i-1th col
v_dif_mag(1:conse-1,:) = v_dif_mag(2:conse,:);
% record the magnitude of change of the last conse obs
v_dif_mag(conse,:) = 0;
% move the ith col to i-1th col
vec_mag(1:conse-1) = vec_mag(2:conse);
% change vector magnitude
vec_mag(conse) = 0;
for i_B = 1:nbands-1
% absolute difference
v_dif_mag(conse,i_B) = clry(i+conse,i_B)-autoTSPred(clrx(i+conse),fit_cft(:,i_B));
% normalized to z-scores
if sum(i_B == B_detect)
% minimum rmse
mini_rmse = max(adj_rmse(i_B),tmpcg_rmse(i_B));
% z-scores
v_dif(conse,i_B) = v_dif_mag(conse,i_B)/mini_rmse;
end
end
vec_mag(conse) = norm(v_dif(end,B_detect))^2;
end
% change detection
if min(vec_mag) > T_cg % change detected
% record break time
rec_cg(num_fc).t_break = clrx(i+1);
% record change probability
rec_cg(num_fc).change_prob = 1;
% record change magnitude
rec_cg(num_fc).magnitude = median(v_dif_mag,1);
% identified and move on for the next functional curve
num_fc = num_fc + 1;
% start from i+1 for the next functional curve
i_start = i + 1;
% start training again
BL_train = 0;
elseif vec_mag(1) > Tmax_cg % false change
% remove noise
clrx(i+1) = [];
clry(i+1,:) = [];
i=i-1; % stay & check again after noise removal
end
end % end of continuous monitoring
end % end of checking basic requrirements
% move forward to the i+1th clear observation
i=i+1;
end % end of while iterative
% Two ways for processing the end of the time series
if BL_train == 1
% 1) if no break find at the end of the time series
% define probability of change based on conse
for i_conse = conse:-1:1
if vec_mag(i_conse) <= T_cg
% the last stable id
id_last = i_conse;
break;
end
end
% update change probability
rec_cg(num_fc).change_prob = (conse-id_last)/conse;
% update end time of the curve
rec_cg(num_fc).t_end=clrx(end-conse+id_last);
if conse > id_last % > 1
% update time of the probable change
rec_cg(num_fc).t_break = clrx(end-conse+id_last+1);
% update magnitude of change
rec_cg(num_fc).magnitude = median(v_dif_mag(id_last+1:conse,:),1);
end
elseif BL_train == 0
% 2) if break find close to the end of the time series
% Use [conse,min_num_c*n_times+conse) to fit curve
if num_fc == rec_fc
% first curve
i_start = 1;
else
i_start = find(clrx >= rec_cg(num_fc-1).t_break);
i_start = i_start(1);
end
% Tmask
if length(clrx(i_start:end)) > conse
blIDs = autoTmask(clrx(i_start:end),clry(i_start:end,[num_B1,num_B2]),...
(clrx(end)-clrx(i_start))/num_yrs,adj_rmse(num_B1),adj_rmse(num_B2),T_const);
% update i_span after noise removal
i_span = sum(~blIDs);
IDs = i_start:length(clrx); % all IDs
rmIDs = IDs(blIDs(1:end-conse) == 1); % IDs to be removed
% remove noise pixels between i_start & i
clrx(rmIDs) = [];
clry(rmIDs,:) = [];
end
% enough data
if length(clrx(i_start:end)) >= conse
% defining computed variables
fit_cft = zeros(max_num_c,nbands-1);
% rmse for each band
rmse = zeros(nbands-1,1);
% end of fit qa = 20
qa = 20;
for i_B = 1:nbands-1
[fit_cft(:,i_B),rmse(i_B)] = ...
autoTSFit(clrx(i_start:end),clry(i_start:end,i_B),min_num_c);
end
% record time of curve start
rec_cg(num_fc).t_start = clrx(i_start);
% record time of curve end
rec_cg(num_fc).t_end=clrx(end);
% record break time
rec_cg(num_fc).t_break = 0;
% record postion of the pixel
rec_cg(num_fc).pos = (nrows-1)*ncols+i_ids;
% record fitted coefficients
rec_cg(num_fc).coefs = fit_cft;
% record rmse of the pixel
rec_cg(num_fc).rmse = rmse;
% record change probability
rec_cg(num_fc).change_prob = 0;
% record number of observations
rec_cg(num_fc).num_obs = length(clrx(i_start:end));
% record fit category
rec_cg(num_fc).category = qa + min_num_c;
% record change magnitude
rec_cg(num_fc).magnitude = zeros(1,nbands-1);
end
end
end % end of if sum(idgood) statement
end % end of for i_ids loop
save(fullfile(dir_l,n_rst,['record_change',num2str(nrows)]),'rec_cg');
end % end of function