Feature Engineering

The performance of data-driven models such as machine learning depends strongly on the quality and relevance of the features used for training. To enhance the predictive capability of RECLAIM, we derived new features from the raw variables described in Section 2, a process commonly referred to as feature engineering. These derived features were designed to capture underlying processes relevant to reservoir sedimentation that are not directly represented in the raw variables. A summary of all derived features is provided in the below table. Please look out the table for the mentioned section to refer used variables.

No	Name	Abbreviation	Definition/Formula	Reference Table	Units
P14	Land cover of artificial surfaces	LCAS	=LCAS%*CA	static variables	km2
P15	Land cover of cropland	LCC	=LCC%*CA	static variables	km2
P16	Land cover of grassland	LCG	=LCG%*CA	static variables	km2
P17	Land cover of trees	LCT	=LCT%*CA	static variables	km2
P18	Land cover of shrubs	LCS	=LCS%*CA	static variables	km2
P19	Land cover of herbaceous vegetation	LCHV	=LCHV%*CA	static variables	km2
P20	Land cover of mangroves	LCM	=LCM%*CA	static variables	km2
P21	Land cover of sparse vegetation	LCSV	=LCSV%*CA	static variables	km2
P22	Land cover of bare soil	LCBS	=LCBS%*CA	static variables	km2
P23	Land cover of snow and glaciers	LCSG	=LCSG%*CA	static variables	km2
P24	Land cover of water bodies	LCWB	=LCWB%*CA	static variables	km2
P73	Age at observation end	AGE	=OEY-BY	static variables	X
P74	Relative original capacity	ROBC	=OBC/CA	static variables	m
P75	Geometry complexity	GC	=RA/RP²	static variables	DL
P76	Net vegetation gain frequency	NVGF	=VGF-VLF	dynamic variables	X
P77	Ratio tree cover to bare soil	R_tree_bare	=LCT/LCBS	static variables	DL
P78	Ratio shrubs to bare soil	R_shrub_bare	=LCS/LCBS	static variables	DL
P79	Ratio coarse to sand	R_coarse_sand	=COAR/SAND	static variables	DL
P80	Relative mean annual surface area	rel_SA_mean_clip	=SA_mean_clip/RA	static & dynamic variables	DL
P81	Ratio surface area to capacity	R_SA_cap	=SA_mean_clip/OBC	static & dynamic variables	m-1
P82	Rainfall per unit area	rain_per_area	=MAR/CA	static & dynamic variables	mm/km2
P83	Trapping efficiency	TE	100*e^(-0.0079*(MAI*3600*24*365/(OBC*1e6)))	static variables	%
P84	Residence time	RT	=OBC*1e6/(MAI*3600*24*365)	static variables	years
P85	Estimated capacity loss rate	ECLR	=TE*NSSC2_mean/RT	dynamic variables	%/year
P86	Estimated sedimentation rate	ESR	=ECLR*OBC/100	dynamic variables	million m3/year
P87	Sediment influx	SIN	=MAI*NSSC2_mean	dynamic variables	m3/s
P88	Sediment outflux	SOUT	=MAO*NSSC2_mean	dynamic variables	m3/s