Chapter Four · failure evidence
What Remote Sensing & Satellite Imagery got wrong, from 82 dissertations
Remote sensing and satellite imagery models frequently struggle with spatial resolution limits, cloud contamination, and discrepancies against in situ ground measurements. In addition, complex terrain and spectral confusion hinder accurate surface classification and out-of-sample regional generalization. These records come from PhD theses at 24 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Coarse spatial resolution prevents the detection of fine-scale objects and within-pixel variation
Moderate and coarse satellite footprints fail to resolve small targets such as individual buildings, burrows, micro-topography, and localized erosion. Pixel blending causes surrounding matrix contamination, mixed-pixel grouping, and severe information loss compared to higher-resolution aerial or manual methods.
Tried and failed
GIS digital elevation models for hydrological modeling applied to excavated terrain and plateau surface hydrology. Outcome: data insufficient. Reason: Coarse spatial resolution and missing historical imagery failed to resolve excavation micro-topography and hydrology
From Wasteland to Biocultural Heritage: Negotiation by Design in Khotale, Konkan, India · Cornell
Tried and failed
moderate resolution multispectral satellite vegetation index analysis applied to small urban green infrastructure features. Outcome: no signal. Reason: Feature sizes were smaller than pixel spatial resolution, causing surrounding matrix contamination to dominate pixel reflectance values.
Lost to a baseline
Historically, black-and-white aerial imagery beat satellite remote sensing in municipal planning due to higher resolution, oblique angles, immediate interpretability, and low cost.
Using Earth Observation-Informed Modeling to Inform Sustainable Development Decision-Making · MIT
Considered and rejected
Considered and rejected: Rejected analyzing small cities with urban radius < 5 km because 0.1° satellite data cannot resolve them and small areas do not exert sufficient rainfall impacts.
Urbanization’s impacts on precipitation : observational analysis of global patterns and storm dynamics · UT Austin
Considered and rejected
Considered and rejected: Rejected using Moderate Resolution Imaging Spectroradiometer (MODIS) 250m imagery for NDVI because spatial resolution was too coarse for small-scale urban greenness.
Considered and rejected
Considered and rejected: Decided against building-by-building footprint shape extraction directly from raster-to-vector converted satellite data in dense/undulated areas due to mixed pixels grouping multiple buildings.
On the reliable generation of 3D city models from open data · University of Nottingham Repository
Tried and failed
coarse-resolution satellite burned area mapping applied to low-severity wildfire detection. Outcome: no signal. Reason: spatial resolution limitations and subtle spectral changes prevented detection of low-severity events
Burned severity assessment and mapping in the Hindu Kush-Himalayan region of Asia · Texas Tech
Tried and failed
multispectral vegetation indices from UAV imagery applied to canopy wilting assessment. Outcome: no signal. Reason: lower spatial resolution in multispectral bands reduced correlation with ground-truth visual scores compared to high-resolution RGB
Multi-modal phenotyping and breeding for drought tolerance in soybean · Iowa State
Tried and failed
spectral mixture analysis with lower-resolution bands applied to satellite imagery land cover classification. Reason: model misspecification occurred due to lack of pure endmember pixels at the coarser spatial resolution
Multiscale spatiotemporal approaches to contemporary pyrogeography in savanna systems · UT Austin
Tried and failed
satellite-guided spatial cluster sampling applied to short-range spatial nutrient variation. Outcome: no signal. Reason: pixel resolution and sampling box design failed to resolve variables with high short-distance spatial variance
Considered and rejected
Considered and rejected: Using satellite-derived land surface temperature or net ecosystem productivity was rejected due to coarse spatial resolution.
Considered and rejected
Considered and rejected: Rejected relying solely on satellite-informed PlanetScope/ClimateTRACE feedlot imagery as prior due to lack of comprehensiveness for small facilities
Considered and rejected
Considered and rejected: Rejected using Planet's SkySat 50 cm satellite imagery for direct burrow object detection because the resolution was insufficient.
Analyzing Colony Structural Characteristics to Assess Establishment of Reintroduced Black-Tailed Prairie Dogs (Cynomys ludovicianus) · TXST Digital Repository
Considered and rejected
Considered and rejected: Rejected using 60 m Landsat MSS imagery prior to 1999 due to spatial resolution being too coarse for meaningful localized erosion delineation.
Channel erosion projections for threatened Alaska-Native communities along the Kuskokwim River · Iowa State
Considered and rejected
Considered and rejected: Rejected standard MODIS 250 m imagery because the pixel footprint (6.25 ha) was too coarse for smallholder median field sizes (0.63–1.05 ha).
Earth Observation for Sustainable Sugarcane Production · Cranfield
Tried and failed
consumer GPS and satellite imagery georeferencing applied to camera calibration over small spatial extents. Outcome: data insufficient. Reason: spatial resolution and consumer GPS accuracy were inadequate over short distances
Towards a Comprehensive Bicycle Motion Behavior Model and Naturalistic Cycling Dataset · Virginia Tech
Tried and failed
social media and remote sensing damage mapping applied to asset-level infrastructure damage assessment. Outcome: data insufficient. Reason: high uncertainty and insufficient precision prevent reliable asset-level structural or utility status evaluation
Leveraging Post-Disaster Data for Rapid Damage Assessment of Infrastructure Systems · Georgia Tech
Tried and failed
manual feature extraction from low-resolution aerial imagery applied to small roadway infrastructure inventorying. Outcome: data insufficient. Reason: spatial resolution was too coarse to resolve small infrastructure elements like signs and markings
Evaluating remotely sensed images for use in inventorying roadway infrastructure features · Iowa State
Considered and rejected
Considered and rejected: Rejected conducting a randomized household survey across Nakuru peri-urban settlements due to lack of formal postal addresses and aerial imagery inability to resolve multi-household informal rental units per structure
Safety in Separation? Ethnic Demography and Violence in Kenya · Cornell
Considered and rejected
Considered and rejected: Rejected grid-based spatial random point sampling stratified by LANDSAT SEDAC satellite density due to aerial inability to detect multiple informal single-story rental units per structure
Safety in Separation? Ethnic Demography and Violence in Kenya · Cornell
Cloud cover and atmospheric contamination cause widespread data loss and masking errors
Persistent cloud cover, shadows, and darkness obscure optical observations, resulting in massive missing data gaps and severe observation dropouts. Automated cloud masking algorithms struggle to separate clouds from snow, ice, or surface water, while aggressive cloud filtering introduces geographic and seasonal sampling biases.
Considered and rejected
Considered and rejected: Rejected using raw satellite imagery due to inaccuracies caused by cloud cover over UK study sites.
Enhancing Winter Wheat Crop Yield Predictions: A Data-Driven, Incremental and Integrative Approach with Machine Learning · Research Repository UCD
Tried and failed
daily satellite nighttime light time-series analysis applied to flood exposure and vulnerability mapping. Outcome: data insufficient. Reason: daily observations had severe cloud contamination and satellite fluctuations, requiring temporal aggregation to filter noise
Considered and rejected
Considered and rejected: Rejected using observational satellite flood extent (Sentinel-1/2) as ML ground truth due to cloud cover, small flood extents, and lack of validation data, switching to Floods Directive numerical model outputs.
Assessing Vulnerability to Rapid-Onset Coastal Flooding: Advancing Methodological Approaches with a Focus on the Mediterranean Sea Basin. · IRIS - POLITO - prod
Tried and failed
satellite optical remote sensing column retrievals applied to regional greenhouse gas emission estimation. Outcome: data insufficient. Reason: frequent cloud cover resulted in extremely low clear-sky observation availability across the target region
Tried and failed
Scene-wide cloud cover threshold filtering applied to satellite vegetation index time series. Reason: retained localized cloud and shadow artifacts causing erroneous drops in mid-season vegetation indices
Earth Observation for Sustainable Sugarcane Production · Cranfield
Tried and failed
rule-based optical satellite cloud masking applied to mountain glacier satellite imagery. Reason: failed to distinguish high-altitude clouds from snow and ice, causing unmasked clouds to be misclassified as water
Tried and failed
satellite multispectral imagery classification applied to crop disease detection across years. Outcome: did not generalise. Reason: poor cross-year model transferability and cloud cover interference
Multidisciplinary strategies for grapevine disease monitoring in the age of digital viticulture · Cornell
Lost to a baseline
229 reservations in the southeast corner of the satellite imagery footprint were excluded due to cloud cover in the 1973 Landsat MSS imagery preventing baseline land cover estimation.
Essays on the Influence of Western Institutions on Indigenous Societies · Harvard
Considered and rejected
Considered and rejected: Rejected satellite images containing cloud cover over more than 50% of the study area
Using Multitemporal Satellite Imagery to Monitor the Response of Vegetation to Drought in the Great Lakes Region · TXST Digital Repository
Considered and rejected
Considered and rejected: Rejected using satellite remote-sensed chlorophyll-a data (NDCI index) due to massive missing data (<30% data availability) caused by persistent Irish cloud cover.
A risk assessment and an early warning system for marine biotoxins in Irish produced shellfish · Research Repository UCD
Considered and rejected
Considered and rejected: Excluded Landsat 8 2014 imagery due to excessive cloud obscuration over the study area.
An approach to map soil texture class on the Iowan Erosion Surface · Iowa State
Considered and rejected
Considered and rejected: Rejected using daily satellite Aerosol Optical Depth (AOD) for the primary subway expansion study due to large numbers of missing observations caused by cloud coverage.
Essays on Environmental Challenges and Regulations in China · Cornell
Considered and rejected
Considered and rejected: Rejected using satellite imagery alone for estimating ice-on/ice-off dates for Lake Pesiöjärvi due to polar night darkness and cloud obstructions, requiring integration of FDD and water temperature profiles.
Considered and rejected
Considered and rejected: Decided against purely satellite DTC/downscaling models due to dependency on clear skies and lack of atmospheric physics.
Impact of urbanization and climate change on building energy use and heat emissions · Iowa State
Considered and rejected
Considered and rejected: Filtering out satellite images with cloud coverage above 5% was rejected because it caused substantial geographic and seasonal data imbalance (e.g., selecting 80% of Yuma vs 19% of Broward data)
Machine Learning Approaches to Predict PM2.5 Using Satellite Images · JScholarship
Optical and multispectral reflectance fails to detect subsurface conditions and subtle physiological stress
Surface spectral bands cannot track subsurface phenomena such as deep dry-season groundwater depletion, understory biomass, or distinct soil classes spanning multiple reflectance values. Spectral indices and reflectance profiles also lack the sensitivity to distinguish specific co-occurring biological agents, disease severity, or agricultural drought impacts.
Tried and failed
random forest classification on satellite imagery applied to archaeological soil and terrain mapping. Outcome: did not generalise. Reason: sensor artifacts and spectral overlap caused high false positives and missed recent land-use changes
Tectonic and Climatic Controls on Continental River Systems · MIT
Tried and failed
satellite-derived vegetation productivity models applied to regional forest productivity estimation. Outcome: no signal. Reason: optical remote sensing product failed to capture known productivity differences across distinct forest management types and regions
Relationships between Forest Productivity and Water Yield Across Virginia's Forests · Virginia Tech
Tried and failed
Gaussian Naive Bayes classification applied to satellite urban land cover mapping. Outcome: worse than baseline. Reason: overlapping spectral features violated naive independence and distribution assumptions across complex urban surface classes
Improved urban extreme weather simulation by capturing urban heterogeneity · UT Austin
Tried and failed
multispectral satellite regression without understory biomass covariates applied to forest basal area estimation. Outcome: no signal. Reason: omitting understory biomass predictors severely degraded overstory basal area model accuracy
Tried and failed
satellite remote sensing proxy validation against in-situ sensors applied to upper ocean chlorophyll concentration estimation. Outcome: no signal. Reason: optical depth satellite estimates failed to correlate with autonomous float measurements
Delineating Planktonic Habitats in Dynamic Marine Environments: From Oceanographic Expeditions to Autonomous Platform · Georgia Tech
Tried and failed
satellite remote sensing proxies for subsurface tracking applied to local groundwater availability estimation. Outcome: no signal. Reason: Surface vegetation and shallow soil moisture indicators decoupled from deep dry-season water level declines.
Towards resolving data scarcity in water resources management · Imperial
Considered and rejected
Considered and rejected: Rejected satellite-derived drought/flood indices in favor of self-reported survey shock measures because satellite indicators (e.g. during Ethiopia's 2015 drought) failed to show agricultural drought signals and lacked exact household coordinates.
Tried and failed
optical multispectral satellite imagery classification applied to fine-grained soil unit mapping. Outcome: data insufficient. Reason: Single soil units spanned multiple spectral classes, preventing discrimination using optical reflectance alone.
Application of GIS and remote sensing for land use planning in the arid areas of Jordan · Cranfield
Tried and failed
RGB-derived vegetation indices and remote sensing proxy estimation applied to crop disease and yield impact assessment. Outcome: no signal. Reason: Optical indices and radiation use efficiency lacked sensitivity to detect fungicide-induced disease variations affecting yield
Tried and failed
multispectral vegetation indices from aerial imagery applied to foliar crop disease severity assessment. Outcome: no signal. Reason: Indices showed no significant correlation with disease scores across inbred trial seasons
NORTHERN LEAF BLIGHT RESISTANCE IN MAIZE: BENEFTIS, COSTS, AND HIGH-THROUGHPUT PHENOTYPING · Cornell
Considered and rejected
Considered and rejected: Rejected individual-agent satellite remote sensing due to inability to differentiate co-occurring biotic agent species identities.
Spatio-temporal patterns of forest disturbance in western North America: implications for forest resilience · ResearchWorks
Tried and failed
manual feature digitization from aerial imagery applied to on-street parking inventory. Outcome: data insufficient. Reason: empty parking spaces lacked distinct visual cues or contrast against the roadway surface
Evaluating remotely sensed images for use in inventorying roadway infrastructure features · Iowa State
Complex topography and surface heterogeneity disrupt classification and physical modeling
Radar and optical algorithms fail across sloped terrain, dynamic water edges, and irregular structures, misclassifying rooftops or confounding multi-scale environmental layers. Ingesting remote sensing features into hydrological or crop models causes mis-specified leaf dynamics and degrades calibration performance compared to baselines.
Tried and failed
standard regression on remote sensing derived features applied to forest canopy height and age modeling. Reason: disturbance detection misclassified thinning events, generating severe label noise and extreme outliers in estimated age
Assessing age-height relationship using ICESat-2 and Landsat time series products of southern pines in southeastern region · Virginia Tech
Tried and failed
entropy thresholding on SAR imagery applied to flood extent mapping. Outcome: did not generalise. Reason: failed in sloped terrain and permanent water, misclassifying urban rooftops as dry
Compound Flood Analysis with GIS-Integrated Dynamic Flood Models for Coastal Urban Areas · Georgia Tech
Lost to a baseline
PIV failed to resolve suspended sediment plume displacements away from sharp plume edges in PlanetScope satellite imagery where Horn-Schunck succeeded.
Advancing coastal ocean modelling through deep learning and remote sensing data integration · Imperial
Tried and failed
tree-based ensemble regression applied to satellite multi-spectral time-series data. Outcome: worse than baseline. Reason: overfit without capturing complex spectral-temporal structures compared to regularized linear models
Tried and failed
spatial downsampling and topographic illumination correction applied to satellite imagery forest biomass estimation. Outcome: no signal. Reason: produced only marginal predictive gains (0.09 Adj. R2 increase) over raw resolution data
Tried and failed
satellite soil moisture data assimilation for initialisation applied to hydrological flood event modeling. Outcome: worse than baseline. Reason: remotely sensed soil moisture values degraded performance for specific flood events compared to baseline calibration
Potential of incorporating satellite soil moisture observations in flood modeling in Kelani river basin, Sri Lanka · Institutional Repository University of Moratuwa
Tried and failed
dynamic parameter ingestion from remote sensing applied to process-based crop ecosystem models. Outcome: worse than baseline. Reason: dynamic parameters caused mis-specification in leaf area accumulation dynamics degrading biomass predictions
Remote sensing and agro-ecosystem modeling: Mutually beneficial in analysis, validation, and application · Iowa State
Tried and failed
random background sampling for semantic segmentation applied to satellite imagery land cover classification. Outcome: worse than baseline. Reason: underperformed compared to boundary-targeted and feature-based sampling strategies
Tried and failed
single-step multi-scale feature classification applied to remote sensing vegetation mapping. Reason: simultaneously combining multi-scale variables confounded fine spectral layers and coarse environmental layers, obscuring boundaries
Methods for quantifying changes to Lesser Prairie-Chicken lek connectivity and habitat · Texas Tech
Tried and failed
omitting digital elevation data in visual segmentation applied to aerial imagery semantic segmentation. Outcome: worse than baseline. Reason: lacks crucial topographic priors, causing false positive segmentations above natural elevation thresholds
Monitoring and understanding treeline dynamics in the Swiss Alps from 80 years of aerial imagery · EPFL
Considered and rejected
Considered and rejected: Rejected automatic digitization algorithms in favor of manual digitization due to non-uniform shapes, irregular spacing, and difficulty detecting thatched roofs from imagery.
Comparative Study of GIS and Conventional Household Survey Sampling Methods: Feasibility, Cost and Family Planning Coverage Estimates · JScholarship
Satellite retrieval products and derived indices show systematic discrepancies against ground truth
Satellite-derived products systematically underestimate physical variables such as soil moisture trends and cloud droplet concentrations while overestimating winter humidity. Standard spectral indices and direct pixel counts deviate significantly from in situ measurements, especially in areas with pre-existing drought mortality or sensor discrepancies.
Tried and failed
random forest regression applied to satellite retrieval bias correction. Outcome: worse than baseline. Reason: significantly underperformed gradient boosted decision tree algorithms like LightGBM and XGBoost
Considered and rejected
Considered and rejected: Rejected direct pixel-count area estimation from classified satellite imagery due to significant classification bias; adopted regression estimator combining ground frames.
Strategic monitoring of crop yields and rangeland conditions in Southern Africa with remote sensing · Cranfield
Considered and rejected
Considered and rejected: Rejected using standard satellite Relativized Burn Ratio (RBR) alone for post-fire mortality assessment because RBR fails in areas with high pre-existing drought mortality or prior fires.
Considered and rejected
Considered and rejected: Rejected relying on satellite dNBR-derived fire severity classes because they differed substantially from ground-truth medium-to-large tree basal area mortality.
The structural impacts of wildfire and defoliation on Mexican spotted owl (Strix occidentalis lucida) nesting habitat · Texas Tech
Considered and rejected
Considered and rejected: Rejected direct single-source modeling and naive averaging of heterogeneous satellite streams due to inability to resolve spatial discrepancies and bias.
Statistical spatio-temporal models with applications to natural processes · Georgia Tech
Tried and failed
threshold-based microphysical quality filtering applied to satellite cloud liquid water retrieval. Outcome: worse than baseline. Reason: filtering thin clouds increased retrieval discrepancy and degraded cross-instrument correlation
Observing aerosol impacts on Arctic clouds · Imperial
Tried and failed
quality filtering by spatial heterogeneity and thresholding applied to satellite remote sensing retrieval bias. Outcome: no signal. Reason: filtering cloud fraction, heterogeneity, and precipitation failed to reduce retrieval discrepancies between distinct sensors
Observing aerosol impacts on Arctic clouds · Imperial
Tried and failed
multisatellite combined soil moisture product applied to long-term soil moisture trend estimation. Reason: Satellite-merged product underestimates the magnitude of trends compared to in situ observations
Tried and failed
gridded satellite meteorological datasets for local simulation applied to photovoltaic power output modeling. Outcome: data insufficient. Reason: satellite gridded data had systematic temperature bias and overly smoothed wind speeds compared to ground stations
Essays on Economics of Distributed Energy · Georgia Tech
Lost to a baseline
NOAA DCOMP and NASA SatCORPS satellite retrieval methods were outperformed by airborne in situ PDI and CAS probes, underestimating Nd by up to 2.5x.
Considered and rejected
Considered and rejected: Rejected using NCAR satellite data for relative humidity boundary conditions due to unrealistically high winter RH values compared to ground stations.
Evaluating Uncertainty in Hygrothermal Modelling of Heritage Masonry Buildings · Carleton University Institutional Repository
Incomplete temporal records and strict spatiotemporal filtering cause severe sample attrition
Filtering time series for dry antecedent weather or strict spatiotemporal colocation drastically reduces available sample sizes and eliminates statistical power. Historical analyses are frequently constrained by missing observations, sensor inconsistencies, and gaps in multi-year satellite archives.
Tried and failed
single-date remote sensing vegetation index regression applied to regional crop yield estimation. Outcome: data insufficient. Reason: temporal mismatch, geometric co-location errors, and unreliable ground-truth reporting
Strategic monitoring of crop yields and rangeland conditions in Southern Africa with remote sensing · Cranfield
Tried and failed
temporal and spatial panel linear regression applied to regional crop yield prediction. Outcome: data insufficient. Reason: missing satellite soil moisture observations prevented detecting statistically significant ecological relationships
Investigating the Ecological Drivers of Agricultural Yield in the Ecuadorian and Peruvian Andes · Harvard
Considered and rejected
Considered and rejected: Rejected using pre-Landsat satellite era (1901-1983) NDVI and inundation percentage in early random forest hindcasts due to historical satellite data limitations.
EVALUATING THE POTENTIAL OF TEMPERATE INLAND MINERAL SOIL WETLANDS AS NATURAL CLIMATE SOLUTIONS · HARVEST
Considered and rejected
Considered and rejected: Rejected using nighttime light imageries as primary longitudinal data due to pixel saturation and inconsistent satellite sensors.
POLYCENTRIC SPATIAL DEVELOPMENT—MEASURES, EXPLANATIONS, PERFORMANCE · Cornell
Tried and failed
strict rainfall filtering on satellite radiometric time-series applied to satellite diurnal vegetation water detection. Outcome: data insufficient. Reason: filtering for dry antecedent conditions eliminated 85 percent of samples, destroying statistical power for diurnal tests
Microwave and destructive measurements of maize seasonal, daily, and sub-daily variations in water quantity · Iowa State
Considered and rejected
Considered and rejected: Excluded 2016 satellite data from the phenology analysis due to data quality issues with Sentinel imagery
Understanding climate change effects in alpine meadows: harnessing the power of data science, and remote sensing · ResearchWorks
Tried and failed
satellite NDVI time series analysis applied to grassland vegetation dynamics monitoring. Outcome: data insufficient. Reason: the multi-year satellite imagery dataset was incomplete over the target region and time period
Securing landscape resilience · Cranfield
Tried and failed
strict spatiotemporal colocation thresholding applied to satellite and ground measurement validation. Outcome: data insufficient. Reason: overly strict matching criteria severely reduced sample size, degrading statistical significance and correlation metrics
Machine learning models trained on satellite features fail out-of-sample spatial generalization
Models trained on remote sensing data suffer large prediction errors and extreme outliers when evaluated on held-out geographic regions without local spatial stratification. Algorithms such as random forests, generative adversarial networks, and downscaling regressions struggle to transfer learned relationships across distinct environmental regions.
Tried and failed
random forest regressor on remote sensing data applied to spatial crop yield prediction. Outcome: did not generalise. Reason: out-of-sample prediction on held-out geographic locations resulted in high prediction errors across fields
Tried and failed
single global random forest without spatial stratification applied to satellite remote sensing downscaling. Outcome: did not generalise. Reason: lack of regional stratification caused systematic regional biases and poor seasonal dynamic capture
Tried and failed
conditional GAN image-to-image translation applied to satellite imagery site suitability mapping. Outcome: did not generalise. Reason: Struggled to capture satellite imagery features representing extreme suitability classes on test data
Tried and failed
adaptive thresholding for segmentation applied to optical satellite snow cover masking. Outcome: did not generalise. Reason: it failed to follow the underlying spatial distribution geometry of the target features
Tried and failed
principal components regression spatial downscaling applied to satellite land surface temperature data. Outcome: did not generalise. Reason: generated extreme urban outliers and high errors (RMSE >8 K)
Impact of urbanization and climate change on building energy use and heat emissions · Iowa State
Lost to a baseline
GBM-Visitor-Lidar and GBM-Visitor-Satellite models showed worse transferability error (NRMSE) on aboveground live tree carbon compared to canopy cover across increasing Kolmogorov-Smirnov distances.
Leveraging Open Data to Support Forest Mapping, Modeling, and Policy Analysis in the Pacific Northwest, USA · ResearchWorks
Left open by the authors
Problems the authors named and did not get to.
Left open
Quantify uncertainty propagation from satellite input datasets through the Random Forest SIF downscaling model. Blocker: None
Left open
Evaluate Meteosat-8 geostationary satellite data for precipitation retrieval over the Tibetan Plateau using random forest models. Blocker: None
Measuring Precipitation from Space - a Satellite View on the Tibetan Plateau · open_UMR Marburg DSpace 10.0
Left open
Extract riverbed vegetation and environmental descriptors from satellite imagery to evaluate their effect on stream velocity prediction using Random Forest models. Blocker: None
Machine Learning and Optimization Model Development for Northern Community Energy Planning · Carleton University Institutional Repository
Left open
Develop improved remote sensing disturbance mapping methods for thinned forest plantations. Blocker: No specific algorithmic approach, target metrics, or baseline dataset are specified.
Assessing age-height relationship using ICESat-2 and Landsat time series products of southern pines in southeastern region · Virginia Tech
Left open
Evaluate the spatial transferability of the random forest temperature model by applying it to additional cities using public satellite and reanalysis data. Blocker: None
Spatio-temporal variability of temperature effects on mortality in Sao Paulo, Brazil · Imperial
Left open
Evaluate higher spatial resolution satellite imagery against temporal resolution trade-offs for high-resolution crop yield prediction models. Blocker: Lacks specific target sensors, methodologies, and accessible matched ground-truth subplot yield data
High-resolution crop yield predictions from satellite-generated NDVI images · Iowa State
Left open
Train region-specific Random Forest classifiers on satellite or geographic data for ecological subregions across the Amazon basin outside Xingu Indigenous Territory. Blocker: None
Tectonic and Climatic Controls on Continental River Systems · MIT
Left open
Train and evaluate out-of-sample ML crop yield prediction models using multi-year field data and high-resolution daily satellite imagery. Blocker: Requires commercial high-resolution satellite imagery (e.g., Planet) and proprietary multi-site ground-truth yield monitor data.
Left open
Benchmark Bayesian convolutional neural networks for rainfall prediction regression using multispectral satellite meteorological imagery and open-source Bayesian deep learning frameworks. Blocker: None
BENCHMARKING BAYESIAN DEEP LEARNING METHODS WITH MULTI-SPECTRAL SATELLITE IMAGERY · Calhoun
Left open
Correlate peanut fungal disease severity with aerial imagery indices and leaf reflectance measurements. Blocker: Requires field-grown peanut crops, physical plant pathology scoring, and aerial/spectroradiometric sensing data
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.