-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
548 lines (494 loc) · 27.9 KB
/
index.html
File metadata and controls
548 lines (494 loc) · 27.9 KB
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
<!DOCTYPE html>
<html lang="en">
<head>
<title>Presto Embeddings</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css?family=Open+Sans:400,700,300" rel="stylesheet" type="text/css">
<link rel="icon" href="assets/NASA_Harvest_favicon.png">
<link rel="stylesheet" type="text/css" rel="noopener" target="_blank" href="main.css">
</head>
<body>
<!-- Header -->
<div id="navbar">
<a href="https://nasaharvest.github.io/">
<div id="logo">
<img src="assets/logo.png"/>
<div id="logotext">
<h1>NASA Harvest</h1>
<h3>Machine Learning</h3>
</div>
</div>
</a>
<div class="menu">
<a href="https://nasaharvest.github.io/">About</a>
<a href="https://nasaharvest.github.io/#sessions">Sessions</a>
<a href="https://nasaharvest.github.io/#profiles">Team</a>
</div>
</div>
<div id="content">
<h1 style="text-align: center; margin-block-end: 0">Presto Embeddings for Cropland Mapping</h1>
<p style="text-align: center;">by Ivan Zvonkov, Gabriel Tseng, Hannah Kerner</p>
<img src="assets/pipeline.png" alt="Presto Embeddings Pipeline" style="width: 100%;">
<p>
Geospatial embeddings offer a novel, efficient, and accessible way to map landscape features.
<br>
In part 1 of this blog post, we show how to generate embeddings using a geospatial foundation model (Presto <a href="#presto" id="presto-ref">[1]</a>).
<br>
In part 2 we show that geospatial embeddings can be used to map cropland with high accuracy.
<br><br>
We have written and linked several Google Earth Engine scripts throughout the blog post,
the scripts can also be accessed through this repository:
<br>
<a href="https://code.earthengine.google.com/?accept_repo=users/izvonkov/presto">https://code.earthengine.google.com/?accept_repo=users/izvonkov/presto</a>
<br><br>
</p>
<details>
<summary>🛰️ <strong>Part 1:</strong> Generating Presto Embeddings</summary>
<h4>Why Generate Embeddings?</h4>
<p>
Presto geospatial embeddings provide a compressed representation of Earth Observation data,
enabling more efficient mapping and analysis. Embeddings are generated by using the Presto encoder
to compress location information, optical imagery (Sentinel-2), radar imagery (Sentinel-1),
climatology data (ERA5), and elevation data (SRTM) over the course of a year.
Each embedding contains 128 features representing a single 10m<sup>2</sup> pixel on Earth.
Embeddings can be used in place of raw Earth Observation data for various machine-learning tasks,
such as classification, clustering, and anomaly detection.
In order to use embeddings they must first be generated for the time frame and area of interest.
We provide an open-source pipeline to do this using Vertex AI and Google Earth Engine.
</p>
<h4>Generating Embeddings</h4>
<p>
The embedding generation pipeline consists of two steps:
<ol>
<li>Deploying Presto to Google Cloud Vertex AI in Google Colab<br>
<a href="https://colab.research.google.com/github/nasaharvest/presto/blob/main/deploy/1_Presto_to_VertexAI.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a></li>
<li>Using the ee.Model.fromVertexAi function in Google Earth Engine<br><a href="https://code.earthengine.google.com/1d196e8466506239c4780585c0e28d26">1_Generate_Embeddings</a></li>
</ol>
In step 1, we package the default Presto model into a TorchServe container.
The container is deployed to a Google Cloud Vertex AI endpoint, which allows for scalable inference
using data from Google Earth Engine.
<br><br>
In step 2, we call the Vertex AI endpoint from Google Earth Engine, sending Earth observation data for our
time frame and area of interest to the deployed Presto model. Once inference is complete, the embeddings are
saved as an Earth Engine asset.
<br><br>
Once predictions are made, you must <strong>undeploy your model</strong> to stop incurring further charges.
This can be done in the Vertex AI console or in the provided Google Colab notebook.
</p>
<h4>Cost Considerations</h4>
<p>
While Google Colab and Google Earth Engine (non-commercial) are free to use,
deploying Presto to Vertex AI and running large-scale inference incurs cloud costs.
<br>The current cost formula is:
<strong style="background-color: pink">$5.37 - $10.14 / 1000 km<sup>2</sup></strong>.
<br><br>
The cost formula is a range due to scaling in Google Cloud.
Small regions will tend toward the higher range,
and larger regions, such as all of Togo, will tend towards the lower range.
</p>
<h4>Case Study: Generating Embeddings for Togo</h4>
<p>
To test our pipeline, we generated embeddings for all of Togo (56,785 km<sup>2</sup>)
for the time frame of March 2019 - March 2020.
We will use labels collected in March 2020 as part of a Rapid Response crop mapping effort in Togo <a href="#togo-rapid" id="togo-rapid-ref">[2]</a>.
The embedding generation took 16 hours and cost $313.40.
The final Togo asset size was 128.8 GB.
</p>
<div class='click-zoom width2'>
<label>
<input type='checkbox' />
<img src='assets/Togo_cost.png' alt='Cost of Togo Embeddings' style="width: 100%;">
</label>
</div>
<div class='click-zoom width2'>
<label>
<input type='checkbox' />
<img src="assets/Togo_stats.png" alt="Asset Size" style="width: 100%;">
</label>
</div>
<p>
The asset can be accessed here:<br><a href="https://code.earthengine.google.com/?asset=users/izvonkov/Togo/Presto_embeddings_v2025_06_19">https://code.earthengine.google.com/?asset=users/izvonkov/Togo/Presto_embeddings_v2025_06_19</a>
<br><br>
It's important to note that Google Earth Engine has a default asset storage limit of 250 GB.
<br>
For larger assets, we recommend exploring
<a href="https://developers.google.com/earth-engine/Earth_Engine_asset_from_cloud_geotiff">Cloud GeoTiff-Backed Earth Engine Assets</a>.
</p>
<h4>Sanity Checking Embeddings</h4>
<p>
To verify that the embeddings contain meaningful information we cluster the embeddings
using k-means and visually compare the clusters to an existing land cover map (WorldCover 2020).
<br>This requires just 9 lines of code:
<a href="https://code.earthengine.google.com/c90e3bcc3b26f2eaee125ccf7d105942">2_Kmeans_clustering_embeddings</a>
</p>
<pre>
// Region of interest
var roi = ee.FeatureCollection("FAO/GAUL/2015/level2").filter(ee.Filter.eq('ADM0_NAME', 'Togo'));
Map.centerObject(roi, 7)
// Load and cluster embeddings
var embeddings = ee.Image("users/izvonkov/Togo/Presto_embeddings_v2025_06_19")
var training = embeddings.sample({region: roi, scale: 10, numPixels: 10000});
var trainedClusterer = ee.Clusterer.wekaKMeans(7).train(training);
var result = embeddings.cluster(trainedClusterer);
Map.addLayer(result.randomVisualizer(), {}, 'clusters');
// Display WorldCover
var WorldCover = ee.ImageCollection('ESA/WorldCover/v200').first().clip(roi)
Map.addLayer(WorldCover, {bands: ['Map']}, 'WorldCover')
</pre>
<div class='click-zoom' >
<label>
<input type='checkbox' />
<img src="assets/Togo_clusters.png" alt="Togo Clusters" style="width: 90%;">
</label>
</div>
<p>
We see that the clusters visually align well with the land cover map.
In particular, the green embedding cluster correlates very closely with built-up areas in the land cover map.
This suggests that the embeddings capture meaningful information about the land cover in Togo
and can be used for further mapping and analysis.
</p>
</details>
<br><br>
<details>
<summary>🌾 <strong>Part 2</strong>: Mapping Cropland in Togo using Embeddings</summary>
<h4>Why use embeddings for mapping?</h4>
<p>
Embeddings are a compressed version of processed Earth Observation data.
So using embeddings for mapping will be more computationally efficient compared
to using traditional Earth Observation data.
Using embeddings also makes it possible to skip data processing (such as cloud processing) altogether.
</p>
<h4>How does it work?</h4>
<p>
A traditional mapping approach usually involves the following:
<ol>
<li>Loading Earth Observation data for the region of interest.</li>
<li>Loading labeled points representing the classes to be mapped.</li>
<li>Creating a training dataset by combining labeled points with Earth Observation data.</li>
<li>Training a classifier (e.g. Random Forest) on the training dataset.</li>
<li>Classifying all the Earth Observation data using the trained classifier.</li>
</ol>
Replace <strong>Earth Observation data</strong> with <strong>embeddings</strong>
and the result is an embedding-based mapping approach:
<ol>
<li>Loading embeddings for the region of interest.</li>
<li>Loading labeled points representing the classes to be mapped.</li>
<li>Creating a training dataset by combining labeled points with embeddings.</li>
<li>Training a classifier (e.g. Random Forest) on the training dataset.</li>
<li>Classifying all the embeddings using the trained classifier.</li>
</ol>
</p>
<h4>Mapping Cropland in Togo</h4>
We need just 12 lines of code to map cropland using embeddings in Togo:
<a href="https://code.earthengine.google.com/a022104b3793e63dcb9459b97886c7d9">3_Togo_cropland_presto</a>.
<br>
We use a random forest classifier with 100 trees.
We initially saw a lot of crops in the output and so used a probability threshold of 0.7.
<pre>
// 1. Load embeddings for region of interest
var roi = ee.FeatureCollection("FAO/GAUL/2015/level2").filter("ADM0_NAME=='Togo'");
Map.centerObject(roi, 7)
var embeddings = ee.Image("users/izvonkov/Togo/Presto_embeddings_v2025_06_19")
// 2. Load Togo points
var points = ee.FeatureCollection("users/izvonkov/Togo/points_2019")
var trainingPoints = points.filter(ee.Filter.eq("subset", "training"))
// 3. Create training dataset (training points + embeddings)
var trainingSet = embeddings.sampleRegions(trainingPoints, ["is_crop"], 10 )
// 4. Train a classifier (100 trees)
var model = ee.Classifier.smileRandomForest(100).setOutputMode('probability')
var trainedModel = model.train(trainingSet, 'is_crop', embeddings.bandNames());
// 5. Classify embeddings using trained model
var croplandPreds = embeddings.classify(trainedModel).clip(roi)
var croplandMap = croplandPreds.gte(0.7).rename("map_crop")
// 6. Display cropland maps
var classVis = {min: 0, max: 1.0, palette: ['yellow', 'green']}
Map.addLayer(croplandMap, classVis, 'Presto Embeddings Based Cropland');
</pre>
<p style="text-align: center">
The resulting cropland map is generated in less than 5 seconds:
</p>
<div class='click-zoom' style="width: 10em; margin-left: 20em">
<label>
<input type='checkbox' />
<img src="assets/Togo_cropland_Presto.png" alt="Togo Cropland" style="width: 100%;">
</label>
</div>
<h4>Alternative Embeddings: Google DeepMind</h4>
<p>
Google DeepMind recently released their own pixel-level
<a href="https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL">embeddings</a>.
<br>Here are some similarities and differences between the two:
</p>
<table>
<tr>
<th></th>
<th>Presto Embeddings</th>
<th>DeepMind Embeddings</th>
</tr>
<tr>
<th>Earth Engine Code</th>
<td>
<pre style="font-size: 0.5em">
<strong>ee.Image(</strong>"users/izvonkov/Togo/Presto_embeddings_v2025_06_19"<strong>)</strong></pre>
</td>
<td>
<pre style="font-size: 0.5em">
<strong>ee.ImageCollection(</strong>"GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL"<strong>)</strong>
<strong>.filterDate(</strong>"2021-01-01", "2022-01-01"<strong>)</strong>
<strong>.mosaic()</strong>
<strong>.clip(</strong>roi<strong>)</strong></pre>
</td>
</tr>
<tr>
<th>Data Sources</th>
<td>Sentinel-1, Sentinel-2, ERA5, SRTM</td>
<td>Optical, radar, LiDAR, and other sources</td>
</tr>
<tr>
<th>Timeframe</th>
<td>March 2019 - March 2020</td>
<td>January 2021 - January 2022<br>[earlier than 2021 not available]</td>
</tr>
<tr>
<th>Embedding Size</th>
<td>128 values</td>
<td>64 values</td>
</tr>
<tr>
<th>Scale</th>
<td>10 m<sup>2</sup></td>
<td>10 m<sup>2</sup></td>
</tr>
<tr>
<th>Data Type</th>
<td><pre>uint16</pre></td>
<td><pre>double</pre></td>
</tr>
<tr>
<th>Bytes per pixel</th>
<td>256</td>
<td>512</td>
</tr>
</table>
<p>
We create an alternative cropland map with Google DeepMind embeddings by simply substituting:
<pre>
<strong>ee.Image(</strong>"users/izvonkov/Togo/Presto_embeddings_v2025_06_19"<strong>)</strong></pre>
with
<pre>
<strong>ee.ImageCollection(</strong>"GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL"<strong>)</strong>
<strong>.filterDate(</strong>"2021-01-01", "2022-01-01"<strong>)</strong>
<strong>.mosaic()</strong>
<strong>.clip(</strong>roi<strong>)</strong></pre>
Here is the updated script on Google Earth Engine:
<a href="https://code.earthengine.google.com/fa52c369fd0d6e5336e9686902972b49">4_Togo_cropland_deepmind</a>.
</p>
<h4>Visual Assessment and Comparison</h4>
<p>
One way of comparing embeddings is visually assessing and comparing the generated cropland maps.
We display the embedding-generated maps below, in addition to the best available cropland maps in Togo.
<br><br>
<span style="background-color: green; color: white">Green represents crops</span> and <span style="background-color: yellow">yellow is everything else</span>. The displayed maps are:
<ol>
<li>a cropland map derived from the WorldCover land cover map <a href="#worldcover" id="worldcover-ref">[3]</a>,</li>
<li>the GLAD cropland map <a href="#glad" id="glad-ref">[4]</a>,</li>
<li>the Presto embedding-based cropland map,</li>
<li>the DeepMind embedding-based cropland map <a href="#dm-emb" id="dm-emb-ref">[5]</a>.</li>
</ol>
</p>
<div class='click-zoom width4'>
<label>
<input type='checkbox' />
<img src='assets/Togo_cropland_WorldCover.png' alt='Togo Cropland WorldCover'>
</label>
<p class="textRight800">1. WorldCover<br><br></p>
</div>
<div class='click-zoom width4'>
<label>
<input type='checkbox' />
<img src='assets/Togo_cropland_GLAD.png' alt='Togo Cropland GLAD'>
</label>
<p class="textRight800">2. GLAD<br><br></p>
</div>
<div class='click-zoom width4'>
<label>
<input type='checkbox' />
<img src='assets/Togo_cropland_Presto.png' alt='Togo Cropland Presto'>
</label>
<p class="textRight800">3. Presto<br>Embeddings</p>
</div>
<div class='click-zoom width4'>
<label>
<input type='checkbox' />
<img src='assets/Togo_cropland_DeepMind.png' alt='Togo Cropland DeepMind'>
</label>
<p class="textRight800">4. DeepMind<br>Embeddings</p>
</div>
<p>
Observations from examining the maps from a country level:
<ul>
<li>Presto and DeepMind embedding-based have a similar cropland amount and distribution</li>
<li>Presto and DeepMind embedding-based show more crops than WorldCover</li>
<li>The South of Togo is where all maps have differences</li>
</ul>
This Google Earth Engine script allows flexible map comparison at different zoom levels:
<a href="https://code.earthengine.google.com/29b52c0bfed621f07c29798a966fb63b">5_Togo_cropland_comparison</a>
For demonstration purposes, we zoom into the South West of Togo.
Here are some screenshots from a window centered on the coordinate: (0.84445, 6.46924):
</p>
<!-- Slideshow container -->
<div class="slideshow-container">
<div class="mySlides click-zoom">
<label>
<input type='checkbox' />
<img src='assets/Togo_zoom_satellite.png' alt="Togo Cropland Zoom Satellite" />
</label>
<div class="numbertext">1 / 5</div>
<div class="text">Satellite view in Google Earth Engine</div>
</div>
<div class="mySlides click-zoom">
<label>
<input type='checkbox' />
<img src='assets/Togo_zoom_WorldCover.png' alt="Togo Cropland Zoom WorldCover" />
</label>
<div class="numbertext">2 / 5</div>
<div class="text">WorldCover Cropland</div>
</div>
<div class="mySlides click-zoom">
<label>
<input type='checkbox' />
<img src='assets/Togo_zoom_GLAD.png' alt="Togo Cropland Zoom GLAD " />
</label>
<div class="numbertext">3 / 5</div>
<div class="text">GLAD Cropland</div>
</div>
<div class="mySlides click-zoom">
<label>
<input type='checkbox' />
<img src='assets/Togo_zoom_Presto.png' alt="Togo Cropland Zoom Presto" />
</label>
<div class="numbertext">4 / 5</div>
<div class="text">Presto Embeddings-based Cropland</div>
</div>
<div class="mySlides click-zoom">
<label>
<input type='checkbox' />
<img src='assets/Togo_zoom_DeepMind.png' alt="Togo Cropland Zoom DeepMind" />
</label>
<div class="numbertext">5 / 5</div>
<div class="text">DeepMind Embeddings-based Cropland</div>
</div>
<!-- Next and previous buttons -->
<a class="prev" onclick="plusSlides(-1)">❮</a>
<a class="next" onclick="plusSlides(1)">❯</a>
</div>
<br>
<!-- The dots/circles -->
<div style="text-align:center">
<span class="dot" onclick="currentSlide(1)"></span>
<span class="dot" onclick="currentSlide(2)"></span>
<span class="dot" onclick="currentSlide(3)"></span>
<span class="dot" onclick="currentSlide(4)"></span>
<span class="dot" onclick="currentSlide(5)"></span>
</div>
<p>
<ul>
<li>All maps with the exception of GLAD correctly identify built-up areas as <span style="background-color: yellow">non-crop (yellow)</span></li>
<li>Embedding-based maps appear to classify more <span style="background-color: green; color: white">cropland (green)</span> correctly</li>
<li>The DeepMind embedding-based map over predicts <span style="background-color: green; color: white">crops</span> in some shrub/tree areas</li>
<li>Presto embedding-based map visually accurately identifies many of the tree areas as <span style="background-color: yellow">non-crop</span></li>
</ul>
</p>
<h4>Metrics Assessment and Comparison</h4>
<p>
To get a more comprehensive understanding of the map quality
we conducted an accuracy assessment using the Togo test set from CropHarvest.
It's important to note that Presto was benchmarked against CropHarvest.
</p>
<table style="text-align: center;">
<tr>
<th style="text-align: left">Metric / Cropland Map</th>
<th>WorldCover</th>
<th>GLAD</th>
<th>Presto<br>Embeddings</th>
<th>DeepMind<br>Embeddings</th>
</tr>
<tr>
<td style="text-align: left">Overall Accuracy</td>
<td>0.880</td>
<td>0.859</td>
<th><span style="color: blue">0.897</span></th>
<td>0.859</td>
</tr>
<tr>
<td style="text-align: left">User's Accuracy</td>
<th><span style="color: blue">0.892</span></th>
<td>0.821</td>
<td>0.833</td>
<td>0.745</td>
</tr>
<tr>
<td style="text-align: left">Producer's Accuracy</td>
<td>0.647</td>
<td>0.627</td>
<th><span style="color: blue">0.784</span></th>
<td>0.745</td>
</tr>
<tr>
<td style="text-align: left">F1-Score</td>
<td>0.750</td>
<td>0.711</td>
<th><span style="color: blue">0.808</span></th>
<td>0.745</td>
</tr>
</table>
<p>Script for assessment: <a href="https://code.earthengine.google.com/52d99816df0fc7b2cda3648cf7f708d5">6_Togo_cropland_embeddings_metrics</a></p>
<h4>Concluding Thoughts</h4>
<p>
We show that using Presto embeddings for cropland mapping is simple, quick, and effective.
Our results indicate the potential for using embeddings for more effective mapping
in different regions and for different use cases.
</p>
</details>
<br><br>
<details>
<summary>Citations</summary>
<p id="presto">[1] <i>Lightweight, Pre-trained Transformers for Remote Sensing Timeseries</i>. Gabriel Tseng, Ruben Cartuyvels, Ivan Zvonkov, Mirali Purohit, David Rolnick, Hannah Kerner, 2023. Neural Information Processing Systems (NeurIPS), Workshop on Tackling climate change with machine learning (<a href=https://arxiv.org/abs/2304.14065>link</a>) <a href="#presto-ref">↵</a></p>
<p id="togo-rapid">[2] <i>Rapid Response Crop Maps in Data Sparse Regions</i>. Hannah Kerner, Gabriel Tseng, Inbal Becker-Reshef, Catherine Nakalembe, Brian Barker, Blake Munshell, Madhava Paliyam, Mehdi Hosseini, 2020. Knowledge Discovery and Data (KDD), Workshop on Humanitarian Mapping (<a href=https://arxiv.org/abs/2006.16866>link</a>) <a href="#togo-rapid-ref">↵</a></p>
<p id="worldcover">[3] <i>ESA WorldCover 10 m 2021 v200</i>. Daniele Zanaga, Ruben Van De Kerchove, Dirk Daems, Wanda De Keersmaecker, Carsten Brockmann, Grit Kirches, Jan Wevers, Oliver Cartus, Maurizio Santoro, Steffen Fritz, Myroslava Lesiv, Martin Herold, N-E Tsendbazar, Panpan Xu, Fabrizio Ramoino, Olivier Arino, Zenodo (<a href=https://esa-worldcover.org/en>link</a>) <a href="#worldcover-ref">↵</a></p>
<p id="glad">[4] <i>Global maps of cropland extent and change show accelerated cropland expansion in the twenty-first century</i>. Peter Potapov, Svetlana Turubanova, Matthew C Hansen, Alexandra Tyukavina, Viviana Zalles, Ahmad Khan, Xiao-Peng Song, Amy Pickens, Quan Shen, Jocelyn Cortez, Nature Food (<a href=https://www.nature.com/articles/s43016-021-00429-z>link</a>) <a href="#glad-ref">↵</a></p>
<p id="dm-emb">[5] <i>AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label data</i>. Christopher F. Brown, Michal R. Kazmierski, Valerie J. Pasquarella, William J. Rucklidge, Masha Samsikova, Chenhui Zhang, Evan Shelhamer, Estefania Lahera, Olivia Wiles, Simon Ilyushchenko, Noel Gorelick, Lihui Lydia Zhang, Sophia Alj, Emily Schechter, Sean Askay, Oliver Guinan, Rebecca Moore, Alexis Boukouvalas, Pushmeet Kohli (<a href=https://arxiv.org/abs/2507.22291>link</a>) <a href="#dm-emb-ref">↵</a></p>
</details>
<br>
<br>
<hr>
<p>This work was supported by the NASA Harvest Consortium.</p>
</div>
<script>
let slideIndex = 1;
showSlides(slideIndex);
function plusSlides(n) {
showSlides(slideIndex += n);
}
function currentSlide(n) {
showSlides(slideIndex = n);
}
function showSlides(n) {
let i;
let slides = document.getElementsByClassName("mySlides");
let dots = document.getElementsByClassName("dot");
if (n > slides.length) {slideIndex = 1}
if (n < 1) {slideIndex = slides.length}
for (i = 0; i < slides.length; i++) {
slides[i].style.display = "none";
}
for (i = 0; i < dots.length; i++) {
dots[i].className = dots[i].className.replace(" active", "");
}
slides[slideIndex-1].style.display = "block";
dots[slideIndex-1].className += " active";
}
</script>
</body>
</html>