BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.cienciaenmexico.com//igisc-2026//talk//9TKLGX
BEGIN:VTIMEZONE
TZID:CST
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=5SU;BYMONTH=10;UNTIL=20001029T080000Z
TZNAME:CST
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
END:STANDARD
BEGIN:STANDARD
DTSTART:20010930T030000
RRULE:FREQ=YEARLY;BYDAY=5SU;BYMONTH=9;UNTIL=20010930T080000Z
TZNAME:CST
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
END:STANDARD
BEGIN:STANDARD
DTSTART:20021027T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CST
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000402T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=4;UNTIL=20000402T090000Z
TZNAME:CDT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20010506T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=5;UNTIL=20010506T090000Z
TZNAME:CDT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20020407T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=4;UNTIL=20220403T090000Z
TZNAME:CDT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-igisc-2026-9TKLGX@pretalx.cienciaenmexico.com
DTSTART;TZID=CST:20261113T110000
DTEND;TZID=CST:20261113T111500
DESCRIPTION:Vegetation-index time series derived from Earth observation pro
 vide a consistent basis for monitoring environmental dynamics\, but their 
 predictive use is constrained by clouds\, incomplete temporal coverage\, s
 patial aggregation choices\, and strong short-term persistence. Moreover\,
  increasingly complex deep-learning models are not necessarily useful unle
 ss they are evaluated against competitive baselines and under temporally h
 onest validation schemes. This study develops and critically evaluates a r
 eproducible GeoAI pipeline for one-month-ahead forecasting of the Normaliz
 ed Difference Vegetation Index (NDVI) at grid-cell level over metropolitan
  San Luis Potosí\, Mexico. The objective was to determine whether a Long 
 Short-Term Memory (LSTM) network can extract predictive information beyond
  the most recent NDVI observation and to identify when and where such adde
 d value occurs.\n\nMonthly Landsat Collection 2 Level-2 surface-reflectanc
 e observations covering January 2017 through December 2025 were processed 
 using the red and near-infrared bands together with QA_PIXEL information f
 or quality filtering. Monthly NDVI surfaces were aggregated to a regular 1
  km × 1 km grid\, and each cell was assigned a persistent cell_id linking
  raster observations\, tabular records\, temporal predictors\, predictions
 \, and spatial diagnostics. Cell-month observations with less than 70% val
 id-NDVI coverage were excluded. The reference predictor set comprised curr
 ent mean NDVI\, three autoregressive lags\, three- and six-month rolling m
 eans\, and cyclical month encodings. Six consecutive monthly predictor vec
 tors were used to predict NDVI one month ahead\, producing 2\,335 valid se
 quences with eight predictors per time step.\n\nTo prevent temporal leakag
 e\, samples were partitioned chronologically rather than randomly. In the 
 fixed reference experiment\, 2\,056 sequences were used for training\, 149
  for validation with target date November 2025\, and 130 for final testing
  with target date December 2025. Continuous predictors were scaled using p
 arameters estimated exclusively from the training partition. The reference
  model consisted of an LSTM layer with 32 units\, 20% dropout\, a 16-unit 
 ReLU dense layer\, and a linear output neuron. Training used mean squared 
 error\, the Adam optimizer\, early stopping\, and adaptive learning-rate r
 eduction. Persistence\, defined as carrying the latest available cell-leve
 l mean NDVI forward one month\, served as the primary benchmark. Complemen
 tary experiments evaluated a multilayer perceptron and an LSTM augmented w
 ith monthly CHIRPS precipitation. Performance was assessed using mean abso
 lute error (MAE)\, root mean squared error (RMSE)\, and the coefficient of
  determination (R²)\, together with independent target-month evaluations\
 , seven balanced two-month temporal test windows\, and cell-level maps of 
 absolute-error differences.\n\nIn the fixed December 2025 test split\, per
 sistence achieved MAE = 0.008298\, RMSE = 0.009362\, and R² = 0.907587. T
 he NDVI-history LSTM achieved MAE = 0.008072\, RMSE = 0.009236\, and R² =
  0.910054. Although the aggregate improvement was modest\, it was obtained
  against a strong baseline for a highly autocorrelated monthly variable. A
 cross independent target-month evaluations\, the LSTM performed better in 
 five of seven months\, while persistence was superior in August and Septem
 ber. In the balanced temporal-window analysis\, the LSTM outperformed pers
 istence in six of seven two-month test windows\, with August–September a
 s the only interval favoring persistence. Spatial diagnostics further show
 ed that better aggregate metrics did not imply spatial dominance: in the D
 ecember reference split\, persistence produced lower absolute error in 75 
 of 130 evaluated cells and the LSTM in 55\, indicating that the LSTM's glo
 bal advantage resulted from larger error reductions in a smaller subset of
  locations.\n\nGreater model complexity did not automatically improve fore
 casting. The precipitation-enhanced LSTM produced MAE = 0.010943\, RMSE = 
 0.012560\, and R² = 0.833682\, whereas the multivariate MLP yielded MAE =
  0.029637\, RMSE = 0.032344\, and R² = −0.102948. These results do not 
 imply that precipitation is physically irrelevant to vegetation dynamics\;
  rather\, the specific monthly CHIRPS representation tested here did not p
 rovide sufficient additional predictive information beyond recent NDVI his
 tory.\n\nThe principal contribution is therefore methodological rather tha
 n architectural. The study integrates quality-aware Landsat preprocessing\
 , spatially consistent aggregation\, leakage-aware chronological sequence 
 construction\, benchmarking against a demanding persistence reference\, te
 mporal robustness analysis\, and spatially explicit error diagnostics with
 in a single GeoAI workflow. The results demonstrate that sequential learni
 ng can provide measurable predictive skill beyond persistence\, but that t
 his advantage is temporally and spatially conditional. They also show that
  additional covariates or architectural complexity require explicit empiri
 cal justification. Because the current model does not explicitly represent
  inter-cell adjacency\, future work will investigate spatial-dependence mo
 dels\, multiple training seeds\, improved treatment of temporal gaps\, and
  land-cover-dependent environmental covariates.
DTSTAMP:20261122T124239Z
LOCATION:Video Room
SUMMARY:A quality-aware GeoAI pipeline for monthly Landsat NDVI forecasting
 : benchmarking LSTM against persistence in metropolitan San Luis Potosí -
  Ulises Argel Gerón Flores\, Juan Alberto Velázquez Zapata
URL:https://pretalx.cienciaenmexico.com/igisc-2026/talk/9TKLGX/
END:VEVENT
END:VCALENDAR
