Papermaking! Vol12 Nr1 2026

S.  Hu,  H.  Qi,  Z.  Wang  et  al.

Environmental  Science  and  Ecotechnology  30  (2026)  100682

fiber  stacking  area,  recovered  fiber  stacking  area,  other  stacking  areas,  wastewater  treatment  area,  thermal  power  plant  area  and  other  built-up  areas  of  the  i-th  type  of  PPP,  respectively  (m  2  );  the coefficients  a  i  ,  b  i  ,  c  i  ,  d  i  ,  e  i  ,  and  f  i represent  the  respective  contri- butions  of  the  five  functional  zones  to  the  overall  carbon  emis- sions;  ε  i  is  the  error  term.

 −  0 : 0049  ×  φ  2 

(11)

β 

+  1 : 088  ×  φ   L  pv  ×  sin  β  opt  ) tan ( 66 : 55  ∘  −  φ )  (

opt  =

(12)

d  s  =  L  pv  ×  cos  β 

opt +

where  β  s  is  the  spacing  between  PV  panels  (m),  φ  is  the  latitude  of  the  plant  location  (  ◦  ),  and  L  pv  is  the  length  of  the  PV  panel  (m). We  then  calculated  the  roof-mounted  PV  power  generation  E  pv  (kWh)  using  equations  (13)  and  (14): A  pv  =  A  a ×  1 d  s (13) opt  is  the  optimal  radiation  angle  (  ◦  ),  d 

2.5.  Model  evaluation  metrics

To  evaluate  the  classification  model's  performance,  we  used  accuracy,  precision,  recall,  and  the  F1  score  as  evaluation  metrics,  which  are  denoted  by  U ,  P ,  R ,  and  F  1  ,  respectively.  Their  formula- tions  are  given  in  equations  (4) – (7): U  = N  TP  +  N  NP N  TP  +  N  TN  +  N  FP +  N  FN  ×  100%  (4) P  = N  TP N  TP  + N  FP  ×  100%  (5) R  = N  TP N  TP  + N  FP  ×  100%  (6)

1  −  F  s 3 : 6

(14)

E  pv  =  θ  ×  A  pv  ×  H 

 δ  × 

T ×

where  A  pv  represents  the  area  of  the  rooftop  PV  panels  (m  2  ),  A  a  is  the  total  area  of  the  rooftop  (m  2  ),  H  T  is  the  total  solar  radiation  intensity  at  the  plant  location  (MJ  m  −  2  year  −  1  ),  θ  is  the  efficiency  of  PV  modules,  and  F  s  is  the  shading  coefficient.  δ  is  the  performance  ratio  (set  to  0.8  following  Wang  et  al.  [41]).  We  set  θ  =  15%  and  F  s  =  0.05,  following  Wang  et  al.  [42].  MJ  was  converted  to  kWh  by  dividing  by  3.6. Finally,  to  calculate  CER  ( C  pv  ),  we  used  the  following  equation: C  pv  =  (  Q  coal  −  Q  pv  )  ×  E  pv  (15) where  Q  coal  and  Q  pv  are  the  life  cycle  for  coal-fired  and  PV  power  generation  (g  CO  2  kWh  −  1  ),  respectively.  According  to  IPCC  [43],  we used  Q  coal  =  950  g  CO  2  kWh  −  1  and  Q  pv  =  50  g  CO  2  kWh  −  1  . 2.7.  GSA  method GSA  was  conducted  using  a  variance-based  approach  that  al- lows  for  correlated  input  parameters.  The  model  output  is  expressed  as: Y  =  g ( X  1  ; X  2  ;  … ; X  n  )  (16) where  Y  is  the  model  output  and  X  1  ; X  2  ;  … ; X  n  represent  the  un- certain  input  parameters. Assuming  the  model  is  executed  M  times,  the  sample  variance  is  calculated  as  follows:

2  ×  P  ×  R  P  +  R

(7)

F 

1  =

where  N 

 positives)  are  true  examples,  N 

 negatives) 

TP  (true

TN  (true

 positives)  are  false  positive 

are  true  counterexamples,  N 

FP  (false

examples,  and  N   negatives)  are  false  negative  examples.  For  the  semantic  segmentation  model  based  on  remote-sensing  images,  we  employed  the  intersection  over  union  ( IoU )  to  measure  the  similarity  between  the  model’s  segmentation  results  and  the  true  labels  (equation  (8)). FN  (false

S  S 

 G  k  G  k

k  ∩ k  ∪

IoU 



(8) 

k =

In  this  equation,  IoU  k  denotes  the  intersection  over  union  for class  k ,  S  k  is  the  pixel  area  predicted  by  the  model,  and  G  k  is  the annotated  pixel  area. For  the  carbon  emission  regression  model,  we  employed  the coefficient  of  determination  ( R  2  ,  equation  (9))  and  the  mean  ab- solute  percentage  error  ( MAPE ,  equation  (10))  as  evaluation metrics.

∑  n i = 1

 2

̂  y

( 

 2 

( y 

 M

i  −

i  )

 Y  )

1  M ∑

Y 

Var ( Y )  =

(17)

R  2 

j  −

=  1  −

(9)

∑  n i = 1

j = 1

2 

 y ) 

( y 

i  −

where  Y  j is  the  result  of  the  j -th  model  run,  and  Y  represents  the  mean  value  obtained  from  M  model  runs. Finally,  we  used  equations  (18)  and  (19)  to  calculate  the  first- order  and  total-order  Kucherenko  sensitivity  indices  [44]:

⃒⃒⃒ y 

1 n ∑ n i = 1⃒

̂  y  i

i  −

MAPE  =

 ×  100%

(10)

y  i⃒⃒⃒⃒

In  these  equations,  y  i  , ̂  y  i  ,  and  y  refer  to  the  actual,  predicted,  and  average  carbon  emissions  of  the  i -th  plant,  respectively,  and  n  is  the  total  number  of  samples.

 Var ( Y | X  i  ) Var ( Y )

(18)

S 

i =

Var ( Y | X  Var ( Y )

∼ i  )

(19)

S 

 1  − 

T ; i =

2.6.  Assessment  of  CER  potential

where  S  i  is  the  first-order  Kucherenko  sensitivity  index  of  input   the  total  output  variance  while  allowing  for  correlated  inputs.  The  total- effect  Kucherenko  sensitivity  index  S  T ; i  denotes  the  total-effect  Kucherenko  sensitivity  index  reflects  the  overall  influence  of  X  i paramete,  which  quantifies  the  contribution  of  X  i  to

To  assess  the  CER  potential  of  rooftop  PV  power  generation  for  existing  PPPs,  we  assumed  that  all  other  built-up  areas  were  available  for  PV  panel  installation.  For  the  modeling  calculations,  we  followed  the  approach  described  by  Wang  et  al.  [41]  (equations  (11)  and  (12)):

6

Made with FlippingBook interactive PDF creator