AgenticCity-Bench: Longitudinal Multimodal Observations of Urban Dynamics with Agentic Reasoning

Xiaofei Yu; Kailai Sun; Yanyu Mao; Baoshen Guo; Heye Huang; Yongcai Wang; Baotong Mi; Alok Prakash

cs.AIOct 26, 2026

AgenticCity-Bench: Longitudinal Multimodal Observations of Urban Dynamics with Agentic Reasoning

Authors: Xiaofei Yu, Kailai Sun, Yanyu Mao, Baoshen Guo, Heye Huang, Yongcai Wang, Baotong Mi, Alok Prakash,

Organizations: Renmin University of China · Singapore-MIT Alliance for Research and Technology, MIT · Massachusetts Institute of Technology · Nanyang Technological University ·

Abstract

City systems are shaped by interactions among infrastructure, human activity, and environmental conditions. Existing earth observation and urban benchmarks are limited in unified modeling and in capturing their joint dynamics. We introduce AgenticCity-Bench, a longitudinal multimodal benchmark for multi-layer unified state modeling and agentic city task reasoning. The benchmark covers 16 cities in 12 countries across six continents from 2000 to 2025, and includes 18 modalities spanning urban function, activity, and environmental state. It organizes heterogeneous data sources around a common spatial reference and calendar index while preserving spatial and temporal coverage, provenance and tile-level availability. On this foundation, AgenticCity-Bench supports urban cross-modal translation together with diverse agentic city tasks involving forecasting, planning and resource analysis. This study provides a unified benchmark for studying the transition from multimodal urban and earth observation to joint understanding and agentic city decision-making. The dataset is available at https://anonymous.4open.science/r/AgenticCity-Bench.

Figures & tables

Figure 1: Overview of the proposed AgenticCity-Bench. (a) Geographic distribution of the 16 cities across six continents. (b) Representative examples of the three modality groups: urban function, activity, and environment.

Family	Semantic modality	SN7	MME	PAN	CB	URB	CL	Ours
Function	Built volume (BV)	$\triangle$	–	$\triangle$	–	–	–	$\checkmark$
	Road network (RD)	–	–	–	$\checkmark$	–	–	$\checkmark$
	Land cover (LC)	–	$\checkmark$	$\checkmark$	–	$\triangle$	–	$\checkmark$
	Land use (LU)	–	–	–	–	$\checkmark$	–	$\checkmark$
	Building height & footprint (H&F)	$\checkmark$	–	$\checkmark$	–	$\triangle$	$\checkmark$	$\checkmark$
	Points of interest (POI)	–	–	–	$\checkmark$	–	–	$\checkmark$

Table 1: Comparison with existing EO and urban datasets. (

\checkmark

) denotes a released direct semantic data counterpart; (

\triangle

) denotes a related semantic counterpart; and – denotes no explicitly released or documented counterpart.

Figure 2: Overview of AgenticCity-Bench. The framework comprises four steps: (1) Data Collection, which gathers multimodal urban data; (2) Tile Construction, which organizes data into standardized spatial tiles; (3) Alignment, which aligns spatial and temporal data; and (4) Benchmark Tasks, which evaluate any-to-any multimodal translation and agentic urban tasks.

Figure 3: Selected qualitative results for any-to-any cross-modal translation. Rows denote source modalities and columns denote target modalities. Same-modality pairs are omitted. The complete $18\times 18$ any-2-any translation matrix is provided in Appendix C.2 .

(a) Visual target modalities
	OPT		DEM	NDVI	H&F	BV	LC	LU	RD
Input	MSE $\downarrow$	LPIPS $\downarrow$	NMAE (%) $\downarrow$	NMAE (%) $\downarrow$	NMAE (%) $\downarrow$	log1p NMAE (%) $\downarrow$	macro mIoU $\uparrow$	macro mIoU $\uparrow$	clDice $\uparrow$
OPT	–	–	8.147	13.009	24.976	33.548	0.213	0.580	0.288
DEM	0.115	0.614	–	8.127	27.111	36.542	0.187	0.451	0.257
NDVI	0.115	0.643	23.084	–	28.336	41.056	0.212	0.507	0.258
H&F	0.120	0.477	12.681	12.947	–	31.675	0.200	0.508	0.266

Table 2: MODUS any-to-any translation performance across target modalities. NMSE is normalized by the training-set 1st–99th percentile range. NMAE and NMSE values are reported as percentages.

Models	NDVI $\to$ WTH macro NMSE (%) $\downarrow$	POI $\to$ ELEC NMSE (%) $\downarrow$	H&F $\to$ POP NMSE (%) $\downarrow$	WTH $\to$ OPT MSE $\downarrow$	OPT $\to$ RD clDice $\uparrow$
ML-4M-3B ( Mizrahi et al., 2023 )	77.800	32.300	106.100	0.063	0.139
Show-O2-1.5B ( Xie et al., 2025 )	53.500	3.324	164.200	0.299	0.024
UniGen-1.5B ( Tian et al., 2026 )	19.000	96.900	87.9	0.036	–
MODUS-7B ( Ye et al., 2026 )	12.397	37.500	20.229	0.126	0.288

Table 3: Performance on five representative source-target urban translation routes. – indicates that the model does not natively support image-to-image generation.

Method	LST MAE ( ∘ C) $\downarrow$	Carbon MAE (tonnes C/tile/year) $\downarrow$	Property MAPE (%) $\downarrow$
Earth Agent (GPT Luna)	1.436	548.58	9.47
Earth Agent (DeepSeek)	1.400	511.62	8.64
Codex (DeepSeek)	1.350	485.78	8.21
Codex (GPT-5.5)	1.341	542.21	8.47
OpenCode (GPT Luna)	1.354	534.51	9.43
OpenCode (Gemini Flash Lite)	1.470	584.56	9.59

Table 4: Forecasting performance on 600 cases: 220 LST, 220 carbon, and 160 property cases

Figure 4: Modality access by Codex (DeepSeek) across forecasting tasks. Dark segments indicate image/raster access; light segments indicate table-only access. Labels show the total percentage of cases accessing each modality.

	EV charging		School capacity
Method	Accuracy (%)	Macro-F1	Balance rate (%)	Mean balance
Constant-class baseline	25.3	0.101	–	–
Three-year-mean baseline	–	–	84.0	96.60
Earth Agent (GPT Terra)	35.3	0.339	81.0	96.91
Earth Agent (DeepSeek)	42.0	0.382	88.0	97.50
Codex (DeepSeek)	38.0	0.366	89.0	97.61

Table 5: Agentic planning results for 150 EV charging port cases and 100 school capacity cases. The school balanced rate uses a 95–105% enrollment-to-capacity range.

Method	Cell acc. (%)	Macro-F1	Exact-case acc. (%)
Majority-class baseline	49.67	0.221	32.0
Earth Agent (GPT Terra)	50.00	0.385	33.3
Earth Agent (DeepSeek)	51.33	0.385	14.7
Codex (DeepSeek)	55.33	0.498	18.7
Codex (GPT-5.5)	53.83	0.475	37.3
OpenCode (GPT Terra)	53.67	0.429	35.3

Table 6: Food access classification on 150 cases. Exact-case accuracy requires all four cells on one tile to be correct.

Appendix figures & tables26 assets

Supplementary material from the paper’s appendix.

Appendix

City	Continent	CRS	$2\,\mathrm{km}$ grid (cells)	$5\,\mathrm{km}$ grid (cells)
Amsterdam	Europe	EPSG:32631	$10\times 14$ (140)	$3\times 4$ (12)
Boston	North America	EPSG:32619	$13\times 13$ (169)	$4\times 4$ (16)
Busan	Asia	EPSG:32652	$16\times 13$ (208)	$5\times 5$ (25)
Cape Town	Africa	EPSG:32734	$14\times 13$ (182)	$4\times 5$ (20)
Christchurch	Oceania	EPSG:32759	$30\times 32$ (960)	$11\times 12$ (132)
Washington, D.C.	North America	EPSG:32618	$13\times 11$ (143)	$5\times 4$ (20)

Appendix

Table 7: Multi-scale city-grid inventory across 16 cities. Grid dimensions are rows

\times

columns and parenthetical values give cell counts. Each scale uses its own projected grid and is evaluated separately.

Modality	Dataset name	Source website
Optical Imagery	Landsat Collection 2 Level-2	https://www.usgs.gov/landsat-missions/landsat-collection-2-level-2-science-products
Optical Imagery	Planet RapidEye(2009–2015) and Mosaics (2016–2025)	https://docs.planet.com/data/imagery/basemaps/
Roads	Open Street Map	https://planet.openstreetmap.org/planet/full-history/
POI	Open Street Map	https://www.openstreetmap.org/
GPS	Open Street Map	https://api.openstreetmap.org/api/0.6/trackpoints
Height & Footprint	3D-GloBFP global building-footprint height	https://essd.copernicus.org/articles/16/5357/2024/

Appendix

Table 8: Primary source registry for the AgenticCity-Bench modalities.

Figure 5: Temporal availability of the 18 benchmark modalities from 2000 to 2025. Columns denote calendar years and rows denote modalities. Color intensity indicates the proportion of valid tiles across the 16 cities.

Semantic modality	Spatial granularity	Temporal coverage
Built volume	$\sim$ 100 m	2000, 2005, 2010, 2015, 2020, 2025 (5-year epochs)
Road network	Vector geometry / rasterized grid	Source-specific historical snapshots
Land cover	30 m	2000–2025
Land use	Source raster grid	2000–2025
Building height & footprint	Building / raster grid	Static
Points of interest	Point / building-level raster	Source-specific historical snapshots

Appendix

Table 9: Spatial granularity and temporal coverage of the 18 semantic modalities in AgenticCity-Bench.

Modality	Available combinations.	Availability
Optical imagery	416 / 416	100.000%
Terrain	416 / 416	100.000%
Vegetation	416 / 416	100.000%
Footprint & height	364 / 416	87.500%
Built volume	416 / 416	100.000%
Land cover	416 / 416	100.000%

Appendix

Table 10: Data availability by modality in the current 2-km release. Each modality has

16\times 26=416

possible city-year combinations. Static and epoch-based data are counted as available when explicitly reused as model inputs.

City	Availability	City	Availability
Amsterdam	392 / 468 (83.761%)	Boston	413 / 468 (88.248%)
Busan	403 / 468 (86.111%)	Cape Town	395 / 468 (84.402%)
Christchurch	353 / 468 (75.427%)	Washington, D.C.	415 / 468 (88.675%)
London	420 / 468 (89.744%)	Lyon	390 / 468 (83.333%)
New York City	411 / 468 (87.821%)	San Francisco	411 / 468 (87.821%)
Santiago	379 / 468 (80.983%)	Shanghai	316 / 468 (67.521%)

Appendix

Table 11: Data availability by city. Each city has

26\times 18=468

possible year-modality combinations.

Modality	Source field(s)	MODUS representation and processing	Unit
Landsat optical imagery	B1=red, B2=green, B3=blue	Native RGB
Planet optical imagery	B1=red, B2=green, B3=blue	Native RGB	-
Road network	B1=major-road raster	Fixed white-background/red-road image	-
Height& footprint	B1= height_m	Fixed-range grayscale PNG: black is zero/background; brighter gray indicates taller buildings. Zero is a valid height.	m
DEM	B1= elevation_m	Train-global linear grayscale dem.png . Zero is valid elevation.	m
NDVI	B1= NDVI	Fixed-range grayscale PNG: darker values indicate lower NDVI and brighter values indicate higher NDVI	1

Appendix

Table 12: Visual spatial modalities used by any2any models. Each source GeoTIFF is converted to a tile-aligned PNG with a modality-specific fixed rendering rule.

Modality	JSON field(s)	Tile-level processing	Unit
Population_GHSL	population_density	Area-weighted mean of the population-density display field.	persons/km 2
Air quality NO2	annual_mean_no2	Area-weighted mean of annual surface NO 2 .	ppb
Air quality PM25	annual_mean_pm25	Area-weighted mean of annual PM 2.5 .	$\mu$ g/m 3
Temperature_LandsatST	mean_celsius ; median_celsius	Area-weighted means of source B1 (mean) and B2 (median), converted to Celsius. Source p90, p95, and standard-deviation bands are excluded in V1.	∘ C
Weather	tmin_c ; tmax_c ; tmean_c ; prcp_mm ; srad_w_m2 ; rh_fraction	The six shared fields are area-weighted over valid source cells.	tmin_c , tmax_c , tmean_c : ∘ C; prcp_mm : mm; srad_w_m2 : W/m 2 ; rh_fraction : %
Property value	property_mean_2017ppp	Area-weighted mean property value, expressed in 2017 PPP international dollars	2017 PPP international dollars per property observation

Appendix

Table 13: Statistical and geo-temporal text modalities used by MODUS. Numerical fields are serialized as typed JSON within the prompt or target text. All raster summaries are computed over valid source support only.

Source / Target	OPT	DEM	NDVI	H&F	BV	LC	LU	POI	RD
OPT	–	193,379	203,199	178,798	207,654	167,984	119,315	166,579	159,147
DEM	193,379	–	120,028	104,078	123,578	102,304	72,408	85,554	88,844
NDVI	203,199	120,028	–	111,913	133,144	110,430	77,667	92,491	86,312
H&F	178,798	104,078	111,913	–	114,894	94,346	70,077	80,046	74,822
BV	207,654	123,578	133,144	114,894	–	113,390	79,632	94,950	89,072
LC	167,984	102,304	110,430	94,346	113,390	–	72,492	74,370	69,897

Appendix

Table 14: Exhaustive one-to-one candidate pairs for 2-km and 5-km scales (28,249,584 directed pairs). Rows indicate sources and columns indicate targets. Panel (a) lists visual target modalities, and panel (b) lists numerical target modalities.

Figure 6: Qualitative any-to-any cross-modal translation across 18 urban modalities. Rows denote source modalities and columns denote target modalities. Same-modality source-target pairs are omitted.

City	A	B	C	D
Boston	$\leq 0$	$1$ – $5$	$6$ – $10$	$\geq 11$
New York City	$\leq 0$	$1$ – $4$	$5$ – $18$	$\geq 19$
San Francisco	$\leq 0$	$1$ – $3$	$4$ – $8$	$\geq 9$

Appendix

Table 15: City-specific strata for EV case selection, defined by the observed net increase in public charging ports over the evaluation horizon.

Task

Predict period-

0

annual tile-mean land-surface temperature from ten historical periods and available urban modalities.

Case and reference outcome

Case ID: Boston_2KM_r7_c8__lst__2022

City: Boston. Forecast period:

0

; calendar year withheld.

Reference: 19.910 ∘ C. Last year: 21.990. Ten-year mean: 21.310.

Submitted answer and evaluated error

Appendix

Table 16: LST forecasting: retaining the historical mean after inspecting auxiliary modalities.

Task

Predict period-

0

annual carbon emissions summed over the four complete 1 km support cells in a 2 km tile, using ten historical periods and available urban modalities.

Case and reference outcome

Case ID: London_2KM_r20_c14__carbon__2022

City: London. Forecast period:

0

; calendar year withheld.

Reference: 1,980.968 tonnes C/tile/year. Last year: 2,053.225. Ten-year mean: 2,430.620.

Submitted answer and evaluated error

Appendix

Table 17: Carbon forecasting: auxiliary histories inform the treatment of a temporary shock.

Task

Predict period-

0

annual carbon emissions summed over the four complete 1 km support cells in a 2 km tile, using ten historical periods and available urban modalities.

Case and reference outcome

Case ID: Amsterdam_2KM_r7_c11__carbon__2021

City: Amsterdam. Forecast period:

0

; calendar year withheld.

Reference: 1,759.693 tonnes C/tile/year. Last year: 1,772.050. Ten-year mean: 1,963.619.

Submitted answer and evaluated error

Appendix

Table 18: Carbon forecasting: inspected auxiliary modalities do not alter the point forecast.

Task

Predict the period-

0

property value of a 2 km London tile: the median of valid 500 m cell-level property-value medians, using ten historical periods and available urban modalities.

Case and reference outcome

Case ID: London_2KM_r16_c26__property__2021

City: London. Forecast period:

0

; calendar year withheld.

Reference: 390,000.000 GBP. Last year: 347,500.000. Ten-year mean: 303,337.375.

Submitted answer and evaluated error

Appendix

Table 19: Property forecasting: distinguishing a transformed target series from independent evidence.

Task and case

Forecast the net change in public charging ports over two years and select a city-specific growth range.

Case ID: evport_494be8490c9337a6 . Tile: Boston_2KM_r6_c3 .

Options: A:

\leq 0

; B: 1–5; C: 6–10; D:

\geq 11

additional ports.

Submitted answer and observed outcome

Codex (GPT-5.5): B (1–5 ports), confidence 0.38.

Reference: D; the observed count rises from 21 to 39 ports, a net increase of 18.

Appendix

Table 20: EV charging-port expansion: a recent buildout is interpreted as temporary saturation.

Task and case

Forecast next-period enrollment and choose nonnegative additional seats to achieve enrollment-to-capacity utilization of 95–105%.

Case ID: school_plan_d2d749da062a . Tile: London_2KM_r10_c20 .

Historical enrollment is 637, 724, and 753; existing capacity is 600 seats.

Submitted answer and observed outcome

Earth Agent (DeepSeek): forecast 782 pupils; add 145 seats; planned capacity 745. Forecast utilization is 104.966%.

Reference enrollment: 788 pupils. Forecast error is 6 pupils; observed utilization is 105.772%, exceeding the upper threshold.

Appendix

Table 21: School capacity planning: a narrow capacity margin is sensitive to enrollment error.

Task and case

Classify the four 1 km cells of a 2 km tile by food access and low-income status.

Case ID: fa_full_221fd43d1213 . Tile: DC_2KM_r8_c8 .

Labels: NLA: not low access; LA: low access; LILA: low income and low access. Reference labels are transferred from USDA census-tract indicators to the assessment cells.

Submitted answer and reference labels

Earth Agent (DeepSeek): NLA for NW, NE, SW, and SE.

Reference: NW, SW, and SE are LILA; NE is NLA. One of four labels is correct.

Appendix

Table 22: Food-access classification: commercial activity is used to infer supermarket accessibility.

Group	$n$	Earth–Luna	Earth–DS	OC–Luna	OC–Gemini	Codex–GPT	Codex–DS	OC–GLM	Last year	Mean10
By city
Amsterdam	15	1.264	1.156	1.403	1.349	1.177	1.302	1.322	1.705	1.198
Boston	15	1.269	1.403	1.309	1.290	1.223	1.326	1.341	2.063	1.216
Busan	15	0.723	0.675	0.825	0.759	0.674	0.663	0.803	0.836	0.561
Cape Town	15	1.264	1.209	1.291	1.193	1.154	1.235	1.374	1.049	1.602
Christchurch	15	1.387	1.207	1.220	1.428	1.173	1.254	1.370	1.355	1.270

Appendix

Table 29: Land-surface temperature: MAE by city and forecast year ( ∘ C).

Group	$n$	Earth–Luna	Earth–DS	OC–Luna	OC–Gemini	Codex–GPT	Codex–DS	OC–GLM	Last year	Mean10
By city
Amsterdam	15	276.613	226.749	288.957	281.078	295.122	287.385	305.042	297.730	1,006.523
Boston	10	114.388	132.012	121.332	107.740	93.660	36.772	84.891	123.812	180.381
Busan	15	514.681	488.062	581.498	517.622	473.480	287.836	453.674	304.751	403.713
Cape Town	15	331.489	331.532	281.313	288.556	380.377	324.304	318.488	256.331	385.654
Christchurch	15	0.084	0.073	0.074	0.064	0.068	0.047	0.088	0.061	0.073

Appendix

Table 30: Carbon emissions: MAE by city and forecast year (tonnes C/tile/year).

Group	$n$	Earth–Luna	Earth–DS	OC–Luna	OC–Gemini	Codex–GPT	Codex–DS	OC–GLM	Last year	Mean10
By city
Boston	20	3.120	2.897	3.137	3.768	2.887	3.278	4.152	4.460	28.308
Busan	20	15.627	17.512	19.443	15.539	17.520	19.124	15.304	17.269	18.847
DC	20	5.112	5.548	5.996	6.237	4.577	4.699	5.370	4.803	27.320
London	20	14.454	11.929	12.367	12.114	12.240	12.274	12.143	10.916	18.306
New York City	20	6.237	5.645	6.221	8.055	5.286	5.265	5.916	3.987	16.074

Appendix

Table 31: Property prices: MAPE by city and forecast year (%).

Agent	Reference	A	B	C	D	Recall (%)
Earth–Terra	A	19	14	5	0	50.000
	B	13	20	5	0	52.632
	C	12	17	5	3	13.514
	D	4	15	9	9	24.324
Earth–DeepSeek	A	27	11	0	0	71.053
	B	8	23	5	2	60.526

Appendix

Table 32: EV charging-port growth: reference-by-prediction confusion counts on the 150-case evaluation cohort. Recall is the diagonal count divided by the reference-row total. A–D denote the city-specific growth ranges.

Agent	Added seats: median [IQR]	Zero addition (%)	Balanced (%)	Shortage (%)	Surplus (%)
Earth–Terra	0.000 [0.000, 38.250]	57.000	81.000	16.000	3.000
Earth–DeepSeek	20.000 [0.000, 75.750]	35.000	88.000	7.000	5.000
Codex–GPT-5.5	0.000 [0.000, 46.500]	57.000	84.000	13.000	3.000
OpenCode–Terra	0.000 [0.000, 33.750]	53.000	80.000	17.000	3.000
OpenCode–Gemini	16.500 [0.000, 61.500]	38.000	84.000	10.000	6.000
Codex–DeepSeek	11.000 [0.000, 67.000]	45.000	89.000	8.000	3.000

Appendix

Table 33: School capacity plans across 100 cases. Planned capacity is the last observed capacity plus proposed seats; balance requires observed next-period enrollment to be 95%–105% of planned capacity.

Agent	Reference	NLA	LA	LILA	Recall (%)
Earth–Terra	NLA	242	22	34	81.208
	LA	188	43	9	17.917
	LILA	38	9	15	24.194
Earth–DeepSeek	NLA	224	68	6	75.168
	LA	146	78	16	32.500
	LILA	45	11	6	9.677

Appendix

Table 34: Food-access cell-label confusion counts on the 150-tile evaluation cohort. NLA: not low access; LA: low access; LILA: low income and low access. Rows are reference classes and columns are predicted classes.

Agent	Uniform: cell (%)	Uniform: exact tile (%)	Mixed: cell (%)	Mixed: exact tile (%)
Earth–Terra	56.183	53.763	39.912	0.000
Earth–DeepSeek	52.957	21.505	48.684	3.509
Codex–GPT-5.5	61.559	58.065	41.228	3.509
OpenCode–Terra	57.527	53.763	47.368	5.263
OpenCode–Gemini	68.548	61.290	38.158	1.754
Codex–DeepSeek	57.796	25.806	51.316	7.018

Appendix

Table 35: Food-access performance by within-tile reference-label composition: 93 uniform-label tiles and 57 mixed-label tiles. Cell accuracy counts correct labels across all four cells; exact-tile accuracy requires all four to be correct.

Task	Image/raster access	$n$	MAE $\downarrow$	$R^{2}\uparrow$
LST	Yes	198	1.359	0.9022
LST	No	22	1.268	0.9391
Carbon	Yes	203	486.841	0.9989
Carbon	No	17	473.070	0.9977
Property: image/raster access, $n=147$			MAPE = 8.472%
Property: no image/raster access, $n=13$			MAPE = 5.282%

Appendix

Table 39: Forecasting performance grouped by recorded image/raster access for Codex (DeepSeek). The two access groups partition each task’s evaluation cases.

Multimodal Dataset

See all 50 papers

UrbanWell: Benchmarking Multimodal Large Language Models for Spatio-Temporal Urban Wellbeing Analytics
Jun 14, 2026Yanxin Xi, Xiang Su, Jie Feng +3Spatiotemporal Reasoning Multimodal Large Language Models
Department of Computer Science, University of Helsinki, Helsinki, Finland · Department of Computer Science, Department of Agricultural Sciences, University of Helsinki, Helsinki, Finland · Zhongguancun Academy, Beijing, China +2
Multi-Agent Collaborative Reasoning with Tool-Augmented Evidence for Urban Region Profiling
Jul 15, 2026Xixuan Hao, Yutian Jiang, Jiabo Liu +4Multi-Agent Reasoning Multimodal Agents
The Hong Kong University of Science and Technology (Guangzhou) Guangzhou, China · University of Washington Seattle, USA · Chang’an University Xi’an, China +1
UrbanAgent: A Tool-Augmented Agent for Cross-System Urban Tasks
Aug 4, 2026Jiayu Cao, Xingyuan Zeng, feiyu Li +6LLM Agent Evaluation AI Agent Benchmarks

Jun 14, 2026J/K move · Enter open · S save

UrbanWell: Benchmarking Multimodal Large Language Models for Spatio-Temporal Urban Wellbeing Analytics

Yanxin Xi, Xiang Su, Jie Feng, Yu Liu+2

Department of Computer Science, University of Helsinki, Helsinki, Finland · Department of Computer Science, Department of Agricultural Sciences, University of Helsinki, Helsinki, Finland · Zhongguancun Academy, Beijing, China+2

Spatiotemporal Reasoning Multimodal Large Language Models Spatiotemporal Forecasting Multimodal Model Evaluation

Open paper AskPDF arXiv

Understanding urban wellbeing from multimodal data requires integrating heterogeneous spatial and temporal signals, posing significant challenges for current multimodal large language models (MLLMs). We introduce UrbanWell, a large-scale benchmark designed to systematically evaluate the spatio-temporal reasoning capabilities of MLLMs for urban wellbeing analytics through joint modeling of satellite and street view imagery. UrbanWell spans 38 cities across multiple years and includes diverse indicators covering (1) environmental conditions (CO

_2

, NO

_2

, PM

{2.5}

, and Normalized Difference Vegetation Index), (2) spatial accessibility (minimum distance to supermarkets and restaurants), (3) urban form (road length, road density, and land use), (4) urban vitality (population, economic activity diversity, and land use diversity), and (5) subjective perception attributes (e.g., safety, beauty, liveliness, wealth, and quietness). All indicators are aligned at grid level to enable standardized evaluation. Beyond static prediction, UrbanWell defines temporal reasoning tasks, including future value forecasting from historical observations and temporal trend classification. We benchmark 15 state-of-the-art representative MLLMs in a zero-shot setting, providing a comprehensive comparative evaluation across spatial and temporal dimensions. Experimental results indicate that while MLLMs capture salient spatial and perceptual cues, their performance varies substantially across heterogeneous urban indicators spanning environment and subjective perception. UrbanWell serves as a unified benchmark for evaluating multimodal spatial and temporal reasoning in urban wellbeing analytics, offering a standardized testbed for systematic assessment and future research on multimodal urban intelligence. Our codes and datasets are accessible via https://github.com/axin1301/UrbanWell-Benchmark.

Abstract

Multimodal Dataset

Explore similar work

UrbanWell: Benchmarking Multimodal Large Language Models for Spatio-Temporal Urban Wellbeing Analytics

Multi-Agent Collaborative Reasoning with Tool-Augmented Evidence for Urban Region Profiling

UrbanAgent: A Tool-Augmented Agent for Cross-System Urban Tasks