Table 5 Comparison of approaches’ performance on the test set from the STTORM-CD-Floods dataset via standard metrics. The “Most recent” column represents results from comparing the changed tile with only the most recent non-cloudy tile. In other columns, all non-cloudy tiles from memory were used to derive the final change predictions using the min() or average() function.

From: STTORM-CD low-demand and high-impact disaster monitoring onboard satellites using change detection

Method

AUPRC [%] \(\uparrow\)

F1 [%] \(\uparrow\)

Most recent

min(memory)

avg(memory)

Most recent

min(memory)

avg(memory)

Geo-Index

60.28

59.48

59.86

59.24

58.73

59.96

Cosine baseline

48.04

63.57

57.99

52.97

58.84

55.74

RaVAEn – small

39.21

43.92

39.62

46.90

50.57

48.15

RaVAEn – medium

40.74

45.07

39.60

47.77

50.77

48.60

RaVAEn – large

39.10

43.33

38.32

47.52

49.93

47.03

STTORM-CD – small

72.45

75.43

75.70

66.42

69.24

69.35

STTORM-CD – medium

73.45

76.00

76.12

69.00

71.25

71.15

STTORM-CD – large

69.05

72.31

72.26

63.18

66.70

66.67

  1. Highlighted are the best scores for STTORM-CD models and RaVAEn models for easy comparison.