Repeated satellite imagery permits understanding of how our planet changes over time, letting us predict its future and plan mitigating actions. Most existing studies that rely on a time series of satellite observations forecast the spectral values at each pixel for a future target date. However, clouds, shadows, and sensor noise often corrupt these observations, so that spectral forecasters waste much of their capacity on low-level features or noise rather than real change. With the advent of Earth foundation models, we argue that forecasting in embedding space is a feasible and more effective alternative. To this end, we design a token space that handles irregular histories from different sensors and foundation models and supports forecasts at any query date. Tokens from this space are inputs to our proposed forecaster, E2Forecast, which is a transformer with two branches: a Gaussian branch that directly predicts a mean point forecast and per-dimension standard deviations, and a rectified-flow branch that generates forecast ensembles. On Sentinel-1 and Sentinel-2 benchmarks, forecasting embeddings instead of spectral values halves the cosine distance (from 0.20 to 0.10) and greatly reduces overconfidence, lowering the z-score standard deviation from 2.63 to 1.22. By comparing new observations with its forecasts, E2Forecast can also detect anomalous changes without labels or retraining, raising F1 from 41.58 to 48.70 on a global benchmark. This consistent superiority is validated on nine remote sensing foundation models. Beyond change detection, we believe our approach advances the field by permitting effective reuse of diverse pretrained task heads. Our code is publicly available at https://github.com/ucam-eo/e2forecast.
Figures & tables
Figure 1: A demo of E 2 Forecast’s flexible forecasting with any query date, producing both point and ensemble forecasts with uncertainty. We choose this Cambridge scene because it includes both urban and rural areas. With time-series OlmoEarth embeddings, we train E 2 Forecast on 2024 only and use January 2025 as context, so every later forecast comes from learned temporal patterns rather than copied observations. Both point and ensemble forecasts of 10 generated embeddings capture the seasonal changes effectively, while ensemble forecasts can better describe the diversity and details.
Figure 2: Overview of E 2 Forecast. (a) Each image in an irregular time series is converted into an Earth embedding ei by an Earth foundation model si , which may differ across observations. (b) The historical embeddings with their dates, the location and the query date τ∗ form a token sequence for a masked transformer fθ . In one pass, the probability-estimation branch predicts the per-dimension Gaussian (μ,σ) of the future embedding, and the embedding-generation branch denoises a noisy target zt=tz1+(1−t)ϵ by flow matching. (c) For forecasting, the probability branch gives a point forecast e^pt∗=μ/∥μ∥2 with uncertainty σ without sampling, and the generation branch turns J noise draws into ensemble forecasts with S Euler steps. (d) For anomaly change detection, a new observed embedding is compared with both the Gaussian forecast and the closest ensemble forecast in units of σ , giving an anomaly change map without change labels or a detection head.
Figure 3: Site locations and irregular temporal sampling in the S1_Embed, S2_Embed and GSA test sets. Left: site locations. Middle: distribution of the lead time from the last context observation to the target date. Right: context dates (circles) and target date (star) of 25 random samples, relative to the target date. Context intervals and lead times vary across samples, spanning 1–170 days for S1_Embed, 5–60 days for S2_Embed and over a year for GSA, so models must forecast embeddings for an arbitrary target date.
Method
S1_Embed
S2_Embed
Mean z→0
Std z→1
Point Cosine ↓
Ensemble Cosine ↓
CRPS ↓
Mean z→0
Std z→1
Point Cosine ↓
Ensemble Cosine ↓
CRPS ↓
Deterministic Models
Persistence
–
–
0.1760
0.1760
–
–
–
0.1699
0.1699
–
ETSformer [ 68 ]
–
–
0.1109
0.1109
–
–
–
0.0737
0.0737
–
Earthformer [ 20 ]
–
–
0.1103
0.1103
–
–
–
0.0808
0.0808
–
Image-Space Forecast with Designed Query-Date
Table 1: Forecasting results on S1_Embed and S2_Embed. Metrics are defined in Sec. 4.1 . Deterministic models give a single forecast, so only cosine distances apply. ∗ marks baselines substantially adapted to the embedding space. Bold : best; underlined : second best.
Figure 4: Image-space vs . embedding-space forecasting on S1_Embed. For each example, we show the target image with R=VV, G=VH, and B=VV-VH. Embeddings are visualized with the first three PCA components.
Figure 5: Image-space vs . embedding-space forecasting on S2_Embed. We show the RGB for the S2 image and visualize embeddings with the first three PCA components. The error maps are computed with error maps of 1−cos over all embedding dimensions.
Dam break
Burned area
Fire point
Input type
Model
R
P
F 1
Model
R
P
F 1
Model
R
P
F 1
Specialized anomaly localization models
MNDWI [ 72 ]
74.64
37.33
40.37
dNDVI [ 29 ]
39.01
87.32
48.24
TAI [ 70 ]
2.91
14.81
3.92
Reflectance
MBWI [ 65 ]
62.59
42.52
41.64
dNBR [ 45 ]
43.30
82.05
52.50
NDFI [ 42 ]
24.64
49.69
21.72
Reflectance
AWEI [ 28 ]
69.21
43.73
43.44
RdNBR [ 48 ]
63.97
67.10
62.13
AFD-S 2 [ 27 ]
2.33
89.95
4.51
Reflectance
General change detection models (bi-temporal)
Table 2: Anomaly change detection results on GSA. Pixel-level recall (R), precision (P) and F 1 (%) for each of the six event categories and their average. Baseline results are taken from [ 36 ] . Best and second best per column are in bold and underlined .
Figure 6: Qualitative anomaly change detection results on GSA. One event per category. (a) Post-event Sentinel-2 RGB image with the annotated anomaly outlined in red. (b) Embedding forecast by E 2 Forecast and (c) embedding observed on the post-event date, both projected to RGB. (d) Anomaly score of Eq. 19 .
Figure 7: Point vs . ensemble forecasts of E 2 Forecast on S1_Embed. Each column is a random test sample. Top: the point forecast against the ground truth over the 128 dimensions. Middle: the ensemble member closest to the ground truth. Bottom: five ensemble members projected onto the first two PCA components of the test targets of the same site. d is the cosine distance to the ground truth.
Historical number
3
5
7
9
12
Std z→1
1.198
1.162
1.138
1.131
1.134
Ensemble Cosine ↓
0.1007
0.0945
0.0911
0.0902
0.0898
Table 3: Effect of the historical number on S1_Embed.
Setting
Mean z→0
Std z→1
Point Cosine ↓
Ensemble Cosine ↓
CRPS ↓
w/o date token
−0.0001
1.2324
0.1188
0.1008
0.0224
w/o location
0.0012
1.1471
0.1100
0.0924
0.0214
SH [ 32 ]
0.0006
1.1627
0.1078
0.0916
0.0213
SHDD [ 69 ]
0.0007
1.1483
0.1088
0.0924
0.0216
RFF [ 9 ]
−0.0005
1.1376
0.1087
0.0911
0.0213
Table 4: Ablation of the date and location tokens on S1_Embed. “w/o date token” removes the date encoding added to the historical observation tokens ( Eq. 9 ); The bottom block compares location encoders.
Figure 8: Average F 1 of E 2 Forecast on GSA with different embedding spaces. Hatched bars add annual Tessera or AlphaEarth embeddings as extra conditions to OlmoEarth, and the dashed line marks ESIA, the best baseline in Tab. 2 . Recall and precision are given in Tab. 6 .
SatCLIP [ 32 ]
Task
Identity
GPS2Vec
CSP-iNat
CSP-FMoW
RN18 L=10
RN18 L=40
ViT-16 L=10
ViT-16 L=40
GeoCLIP
E 2 Forecast
Population
−2.9862
−0.5996
−2.0198
−0.1886
0.1468
0.2373
−0.1801
0.1028
0.3717
0.6064
Elevation
−6.8305
−0.1472
−4.4808
−0.5511
−19.2185
−3.6502
−3.4695
−0.0937
−1.9232
0.1687
Temperature
−94.6671
−0.2646
−0.6582
0.1980
0.2065
0.5840
0.3687
0.5450
0.7094
0.7628
House value
−0.0118
−0.0316
0.0774
−0.5702
0.2182
−0.2985
0.0083
−2.8809
0.4740
0.2183
Median income
−0.0185
0.5964
0.1035
−1.8341
0.1644
0.0224
0.5032
0.2731
0.6646
0.5282
Table 5: Results when adapting E 2 Forecast as INR on five regression tasks ( R2 ). All methods share the same probe and spatial splits. L : Legendre degree of SatCLIP. Best per row in bold.
Figure 9: The global embedding field compared with different INR models. Embeddings queried on a 0.5∘ grid, projected to RGB by PCA, with the ocean masked. Left: the 2025 Esri land-cover map [ 30 ] .
Figure 10: Time- and location-aware global embedding field. E 2 Forecast trained on SSL4EO-S12 [ 66 ] with the query date token, queried on a 0.5∘ grid for each month. The annual mean is subtracted at each location before projecting to RGB.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 11: Point vs . ensemble forecasts of E 2 Forecast on S2_Embed. Same layout as Fig. 7 , over the 1024 dimensions. d is the cosine distance to the ground truth.
Figure 12: Effect of the forecast lead time. Test samples of S1_Embed (a–c) and S2_Embed (d–f) are grouped by the lead time, and the tick labels give the mean lead time and the number of items of each group. (a, d) Mean and std of the z -score of the Gaussian head; (b, e) cosine distance of the point forecast and of the best of ensemble members; (c, f) CRPS of the Gaussian head. Error bars are 95% confidence intervals of the group mean, and dashed lines mark the value over the whole test set.
Figure 13: Held-out predictions on all five tasks. Top row: spatial split ( train , val , test ). Below: ground truth and each method on the test band only, sharing the ground-truth colour scale per column; both SatCLIP rows use the ViT-16 tower. Axes are longitude and latitude in degrees.
Embedding space
R
P
F 1
OlmoEarth [ 24 ]
65.48
48.84
48.70
+ Tessera [ 17 ]
65.61
49.65
49.29
+ AlphaEarth [ 7 ]
69.45
47.68
47.04
Galileo [ 63 ]
70.54
45.58
46.53
CROMA [ 19 ]
69.25
34.25
36.82
SatMAE [ 13 ]
61.51
42.56
42.61
Appendix
Table 6: Results of E 2 Forecast on GSA with different embedding spaces. Average pixel-level recall (R), precision (P) and F 1 (%) over the six event categories. “+” adds annual embeddings to OlmoEarth as extra conditions. Best and second best per column are in bold and underlined .