Skip to content

Commit fd0c683

Browse files
The Google Earth Engine Community Authorscopybara-github
The Google Earth Engine Community Authors
authored and
copybara-github
committed
three data journey cases
PiperOrigin-RevId: 877167308
1 parent 8abf5ae commit fd0c683

File tree

4 files changed

+527
-0
lines changed

4 files changed

+527
-0
lines changed

‎experimental/data_journeys/README.md‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
Cases for the data journeys presentation/paper.
2+
3+
Originally prepared by simonf@ for AAG 2026
4+
5+
* [How the "Missing Survey" Drives India's Demolition Crisis](india_slum_demolition_survey_absence.md)
6+
* [Spain SIGPAC](spain_sigpac_error_correction.md)
7+
* [Malawi: Africa RiskView "No Trigger" Crisis (2015/16)](malawi_arc_no_trigger.md)

‎experimental/data_journeys/india_slum_demolition_survey_absence.md‎

Lines changed: 154 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,154 @@
1+
# The Cartography of Erasure: How the "Missing Survey" Drives India's Demolition Crisis
2+
3+
In the rapidly urbanizing landscapes of Delhi and Mumbai, the access to
4+
rehabilitation after slum evictions has become contingent on a specific
5+
geospatial record: the government survey. The state procures satellite maps
6+
every three months to detect new construction for demolition — demonstrating
7+
high geospatial capacity for enforcement.
8+
9+
Eligibility for rehabilitation depends on a different instrument entirely: a
10+
joint survey and documentary proof under the 2015
11+
[rehabilitation policy](https://delhishelterboard.in/main/wp-content/uploads/2019/01/Relocation-Policy-2015.pdf).
12+
We see two data journeys on the same territory: one completes, the other one is
13+
blocked.
14+
15+
Courts have [repeatedly](https://indiankanoon.org/doc/159570569/)
16+
[emphasised](https://indiankanoon.org/doc/39539866/) that a detailed survey
17+
prior to eviction and a rehabilitation plan are essential to due process; in
18+
practice, disputes often centre on whether a meaningful survey happened and who
19+
was counted.
20+
21+
In 2023 alone, approximately **280,000 people** were forcibly
22+
[evicted](https://tribe.article-14.com/post/2-8-lakh-homes-demolished-in-delhi-in-2023-but-rehab-fails-as-bjp-aap-ignore-their-own-promises-court-orders-66442cca27158)
23+
in Delhi, the
24+
[highest](https://www.newsclick.in/more-half-million-people-5-lakh-evicted-india-2023-report)
25+
number in India according to the Housing and Land Rights Network (HLRN) . Yet,
26+
rehabilitation often hinges on two gatekeepers: (1) whether a cluster is treated
27+
as ‘protected/eligible’ under the policy framework (officially limited to 675
28+
recognized *Jhuggi Jhopri* or JJ clusters) and (2) whether households meet the
29+
policy’s cut-off dates, possess correct documentation, and appear in a
30+
[joint survey](https://www.newslaundry.com/2025/06/30/delhis-demolition-drive-27000-displaced-from-9-acres-of-encroached-land).
31+
Enforcement tends to be well-instrumented (e.g., satellite monitoring aimed at
32+
spotting ‘new’ jhuggis), while rehabilitation turns on slower, contested
33+
steps—joint surveys, voter-list appearance, and pre-2015 documents—where
34+
exclusions are common.
35+
36+
A June 2025
37+
[ground report](https://www.newslaundry.com/2025/06/30/delhis-demolition-drive-27000-displaced-from-9-acres-of-encroached-land)
38+
assessed 27,000+ people displaced across multiple demolition drives since the
39+
BJP came to power; it also reports that 344 homes (about five acres) were razed
40+
at Bhoomiheen Camp on June 11. In Mumbai, despite a slum population exceeding 5
41+
million (42% of the city), Wilson Center reporting
42+
[estimates](https://www.wilsoncenter.org/article/building-slum-free-mumbai)
43+
about 0.15 million tenements rehabilitated under the model over two decades
44+
(with ~0.12 million approved but not begun), against an early target of 1
45+
million.
46+
47+
## What the State Could See — and What It Chose Not to Count
48+
49+
### Selective Mapping
50+
51+
The state demonstrates high-resolution precision when mapping land for
52+
infrastructure projects—metro lines, highways, and commercial complexes. This
53+
"structure mapping" facilitates clearance. Conversely, "social mapping"—the
54+
enumeration of families, tenure history, and eligibility—is systematically
55+
neglected.
56+
57+
The landmark *Sudama Singh* [judgment](https://indiankanoon.org/doc/39539866/)
58+
(2010) of the Delhi High Court mandated that *prior* to any eviction, the state
59+
must conduct a survey of all persons to determine eligibility. In practice,
60+
agencies circumvent this by declaring settlements as "new encroachments" not
61+
found on the official list, thereby bypassing the obligation to survey or
62+
rehabilitate.
63+
64+
### The Survey Gap: 6,343 vs. 675
65+
66+
The magnitude of the data deficit is stark. The National Sample Survey Office
67+
(NSSO) 69th Round (2012)
68+
[identified](https://des.delhi.gov.in/sites/default/files/urban_slums_in_delhi.pdf)
69+
**6,343 slums** in Delhi, housing over a million households. However, the Delhi
70+
Urban Shelter Improvement Board (DUSIB) officially recognizes only **675 JJ
71+
Bastis** (clusters) for rehabilitation purposes. These figures aren’t perfectly
72+
like-for-like (NSS uses a broad ‘slum’ definition; ‘JJ clusters’ are an
73+
administrative category), but the comparison still illustrates a major gap
74+
between the estimated universe of informal settlements and the subset treated as
75+
protected/eligible in practice.
76+
77+
This discrepancy means that nearly 90% of Delhi's slum settlements exist in a
78+
legal blind spot. Residents of these "non-notified" slums are vulnerable to
79+
summary eviction because their settlements do not appear on the JJ Bastis list
80+
used by policymakers. Slum coverage is
81+
[poorly organized](https://timesofindia.indiatimes.com/city/delhi/ud-department-seeks-information-on-82-slum-clusters-gone-missing-in-delhi/articleshow/99222772.cms),
82+
with documents often unavailable.
83+
84+
## How Absence Became Ineligibility
85+
86+
### The "Cut-Off" Date Trap
87+
88+
Eligibility for rehabilitation is anchored to arbitrary chronological
89+
milestones. In Delhi, the *Delhi Slum & JJ Rehabilitation and Relocation Policy,
90+
2015* stipulates a dual requirement: the slum cluster must have existed prior to
91+
**January 1, 2006**, and the individual family must possess documents proving
92+
residence prior to **January 1, 2015**.
93+
94+
In Mumbai, the exclusionary mechanics are even more stratified. Residents are
95+
entitled to free housing only if they can prove residence prior to **January 1,
96+
2000**. Those arriving between 2000 and 2011 fall into a "paid rehabilitation"
97+
tier, requiring them to pay
98+
[₹2.5 lakh](https://citizenmatters.in/no-more-free-rehousing-but-government-offers-discounted-homes-to-slum-dwellers/)
99+
(with some early proposals pushing even higher) for a replacement tenement—a sum
100+
often prohibitive for the urban poor. Structures outside those windows are
101+
ineligible.
102+
103+
The state's failure to update these maps creates a "relocation gap." Families
104+
displaced from surveyed areas often move to unsurveyed peripheral lands, where
105+
they are again classified as "new encroachers,"
106+
[trapping](https://www.wilsoncenter.org/article/building-slum-free-mumbai) them
107+
in a cycle of perpetual displacement.
108+
109+
### Judicial Ambivalence
110+
111+
The judiciary, once the guardian of housing rights, has increasingly
112+
[facilitated](https://article-14.com/post/how-delhi-s-govt-is-using-its-slum-rehabilitation-policy-to-deny-rehabilitation-to-those-who-need-it-6858cb6129469)
113+
this erasure. While *Sudama Singh* (2010) and *Ajay Maken* (2019) established
114+
robust protections, recent rulings have diluted them. A 2022 Delhi High Court
115+
[ruling](https://indiankanoon.org/doc/131330551/) interpreted the 2015 Policy
116+
restrictively, effectively
117+
[denying](https://scroll.in/article/1053522/how-a-delhi-high-court-judgement-has-made-slum-residents-vulnerable-to-arbitrary-evictions)
118+
relief to residents of non-notified slums because they were not on the DUSIB
119+
list.
120+
121+
This jurisprudence effectively treats the "list" as conclusive proof of
122+
existence, ignoring the reality that the list itself is outdated and incomplete.
123+
The courts now frequently treat the absence of a survey as the resident's
124+
failure rather than the state's breach of duty.
125+
126+
### The Right to be Counted
127+
128+
The "geospatial void" is a political choice. Slum dweller organizations argue
129+
that the refusal to survey is a refusal to recognize citizenship. Residents are
130+
currently
131+
[demanding](https://cpiml.org/mlupdate/slum-dwellers-convention-demands-permanent-halt-on-bulldozer-rampage-in-delhi)
132+
a fresh survey of all JJ clusters, with a revised cut-off date of **July 1,
133+
2025**, to reflect the reality of post-2015 urbanization.
134+
135+
Without a survey, a family that has lived in a *jhuggi* for twenty years—paying
136+
for electricity and voting in elections—is legally indistinguishable from a
137+
squatter who arrived yesterday.
138+
139+
## See also
140+
141+
[Policy Analyses and Recommendations for the Delhi Urban Shelter Improvement
142+
Board](https://spia.princeton.edu/sites/default/files/content/India_Policy_Workshop_Report-Final.pdf)
143+
from the Woodrow Wilson School
144+
145+
## Framework annotation
146+
147+
* blockade->void: enforcement data journey completes while recognition data
148+
journey is strategically blocked; governing effect is non-existence —
149+
families with decades of residence rendered invisible
150+
* the cut-off dates and list requirements launder the void through gate-like
151+
procedural language ("no survey, no list, no entitlement")
152+
* courts in Sudama Singh and Ajay Maken were reaching for something like a
153+
warrant card — demanding the state declare what it knows before it
154+
demolishes.

‎experimental/data_journeys/malawi_arc_no_trigger.md‎

Lines changed: 153 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,153 @@
1+
# Malawi: Africa RiskView "No Trigger" Crisis (2015/16)
2+
3+
In the 2015/16 agricultural season, Malawi declared a national state of disaster
4+
as drought left 6.5 to 6.7 million people
5+
[food-insecure](https://actionaid.org/publications/2017/wrong-model-resilience).
6+
The government had purchased drought insurance costing approximately $4.7
7+
million (often rounded to $5 million) from the African Risk Capacity (ARC), a
8+
G7-backed parametric insurance mechanism. ARC is designed to provide rapid
9+
payouts —typically within
10+
[2-4 weeks](https://www.arc.int/sites/default/files/2021-09/XCF-Policy-Brief-Summary.pdf))—
11+
when satellite-detected rainfall deficits lead to estimated response costs
12+
[exceeding](https://www.arc.int/how-arc-works) specified thresholds.
13+
14+
The insurance did not pay out immediately. The satellite model
15+
[calculated](https://www.researchgate.net/publication/322357484_The_wrong_model_for_resilience_How_G7-backed_drought_insurance_failed_Malawi_and_what_we_must_learn_from_it)
16+
that only **20,594** people had been affected—a figure over three hundred times
17+
lower than reality. The model had been calibrated for a crop that farmers no
18+
longer grew.
19+
20+
## What the Index Measured — and What It Missed
21+
22+
### The Water Requirements Satisfaction Index
23+
24+
Africa RiskView, ARC's technical engine, uses the Water Requirements
25+
Satisfaction Index (WRSI)—a [model](https://www.arc.int/drought) developed by
26+
the FAO in the 1970s that monitors water deficits throughout the growing season.
27+
The model takes satellite-based rainfall estimates and
28+
[calculates](https://africariskview.org/Content/Technical-Note_en.pdf) the ratio
29+
of actual to potential evapotranspiration: how much water the crop received
30+
versus how much it needed.
31+
32+
WRSI translates rainfall into crop stress, and crop stress into estimated
33+
affected populations. When the index crosses a predetermined threshold,
34+
insurance payouts are triggered automatically—no claims process, no ground
35+
verification, no delay. This is the promise of parametric insurance: speed at
36+
scale.
37+
38+
But WRSI requires accurate parameterization. The model must know what crop is
39+
being grown, when it was planted, how long its growing season lasts, and what
40+
its water requirements are at each growth stage. Get these parameters wrong, and
41+
the model monitors a phantom crop while the real one dies.
42+
43+
### The Phenology Mismatch
44+
45+
Malawi's ARC policy was
46+
[calibrated](https://www.arc.int/sites/default/files/2021-10/Malawi-Response-Matrix-Clean.pdf)
47+
assuming farmers primarily planted late-maturing maize varieties with 120-140
48+
day growing cycles. This assumption was embedded in the model's "crop
49+
calendar"—the temporal window during which rainfall deficits would be monitored.
50+
51+
But Malawian farmers, responding to shifting climate patterns and extension
52+
advice about drought risk, had increasingly switched to early-maturing varieties
53+
with 90-day cycles. These shorter-season crops mature faster, requiring rain at
54+
different times than their longer-season cousins.
55+
56+
The 2015/16 rainfall pattern was particularly cruel to short-cycle maize. Dry
57+
spells hit precisely during the critical growth phases of the 90-day varieties.
58+
But the model, looking for deficits in the 120-140 day window, recorded a
59+
"passable" season. The long-cycle maize the model tracked would have survived;
60+
the short-cycle maize that farmers actually planted did not.
61+
62+
### The Trigger That Didn't Trigger
63+
64+
When rainfall becomes low enough, the WRSI value drops. If this agronomic index
65+
falls below a critical threshold during the monitored window, the model
66+
estimates a high affected population and a correspondingly high response cost,
67+
triggering the insurance. In 2015/16, Malawi's WRSI—calculated for the wrong
68+
crop—stayed above this critical threshold. The model said: minor drought, 20,594
69+
affected. The government said: national emergency, 6.7 million affected.
70+
71+
## How the Proxy Became the Decision
72+
73+
### The $4.7 Million Premium, $300 Million Gap
74+
75+
Malawi paid ~$4.7 million in insurance premiums for coverage that was supposed
76+
to address drought. The eventual $8.1 million payout—itself contested as
77+
inadequate—arrived against a backdrop of a humanitarian funding gap estimated at
78+
over $300 million (total response costs were ~$395 million). ActionAid's
79+
analysis concluded the payment was "too little, too late" and "effectively
80+
represented an economic loss to Malawi".
81+
82+
The arithmetic was brutal: Millions paid for insurance that didn't trigger when
83+
needed, forcing the government to seek emergency humanitarian funding anyway,
84+
then receiving a small payout nine months later that couldn't address the
85+
immediate crisis it was meant to prevent.
86+
87+
For ActionAid and local civil society, the failure highlighted a severe
88+
opportunity cost: the $4.7 million premium could have funded decentralized
89+
weather stations, irrigation, or the Village Savings and Loans (VSLs) networks
90+
that rural Malawian women actually relied on during crises. Government officials
91+
indicated they would not renew the policy.
92+
93+
### The Epistemological Problem
94+
95+
Parametric insurance depends on the premise that satellite observations can
96+
substitute for ground truth — that the model's assessment is close enough to
97+
reality that automated triggers are preferable to slow, expensive claims
98+
verification. The Malawi case illustrates what happens when that premise fails.
99+
100+
The problem wasn't that satellites can't measure rainfall accurately. The
101+
problem was that the chain from rainfall to crop stress to food insecurity
102+
requires knowledge that satellites cannot provide: what farmers actually
103+
planted, when they planted it, and how their specific varieties respond to
104+
specific weather patterns. This "ground truth" about cropping patterns was
105+
outdated, and the model had no mechanism to detect its own obsolescence.
106+
107+
### The Retrospective Recalculation
108+
109+
In its own defense, ARC maintained that the system's safeguards functioned as
110+
intended. When discrepancies emerged between the model's output and the ground
111+
truth reported by the Malawi Vulnerability Assessment Committee (MVAC), ARC's
112+
framework allowed for a technical review. After months of dispute and field
113+
investigations, ARC recalculated the model using the corrected 90-day maize
114+
parameters. The revised index showed severe drought impact—triggering the
115+
payout.
116+
117+
In November 2016, ARC announced an $8.1 million payout. By January 2017, when
118+
the funds finally arrived, the humanitarian response had been funded through
119+
other means, the planting season had passed, and the "rapid" in "rapid response"
120+
had failed to materialize. What was designed as 2-4 week emergency financing
121+
arrived nine months after the disaster declaration.
122+
123+
### The Legitimacy Collapse
124+
125+
The Malawian government formally disputed ARC's "satellite truth," forcing an
126+
expensive retrospective ground audit that ARC's model was designed to avoid.
127+
Kenya and Malawi
128+
[pulled out](https://www.globalinclusiveinsuranceforum.org/sites/default/files/The-Future-of-Disaster-Risk-Pooling-for-Developing-Countries.pdf)
129+
of ARC for the 2016-17 season. Participation across the continent plummeted:
130+
just three countries (Burkina Faso, Senegal, The Gambia) took part in the
131+
2018/19 pool. The model's failure in one country contaminated confidence across
132+
the continent.
133+
134+
## The Aftermath: Evolution and Return
135+
136+
The 2015/16 crisis was
137+
[studied](https://www.researchgate.net/publication/322357484_The_wrong_model_for_resilience_How_G7-backed_drought_insurance_failed_Malawi_and_what_we_must_learn_from_it)
138+
and
139+
[catalyzed](https://www.arc.int/sites/default/files/2021-10/ARC_LessonsLearned.pdf)
140+
significant reforms: ARC instituted mandatory customization reviews, sensitivity
141+
analyses, and strengthened in-country Technical Working Groups. Malawi
142+
eventually rejoined the risk pool, and during the 2021/2022 season the updated
143+
system successfully triggered a $14.2 million payout — proving the model could
144+
learn.
145+
146+
## Framework annotation
147+
148+
* path->gate: linear travel from satellite to model to reinsurer to trigger;
149+
binary threshold determination
150+
* asymmetric uncertainty burden: model permitted to guess at national scale;
151+
government required to prove it wrong at field level, retroactively, at own
152+
expense
153+
* this sits on the diagonal

0 commit comments

Comments
 (0)