As an important parameter to characterize physical and biogeochemical processes, sea surface salinity (SSS) has received extensive attention. Cubist is a data mining model, which can be well-suited to estimate and analyze SSS in the Gulf of Mexico (GOM) because it can reflect the SSS internal heterogeneity in the GOM—overall circular distribution, and the seasonality related to temperature and river discharge changes. Using remote sensing reflectance (Rrs) at 412, 443, 488 (490), 555, and 667 (670) nm and sea surface temperature (SST), a cubist model was developed to estimate SSS with high accuracy with the overall performance demonstrates a root mean square error (RMSE) of 0.27 psu and correlation coefficient of 0.97 of R2. The model divides the GOM area according to model rules into four sub-regions, which include estuary, nearshore, and open sea, reflecting the gradient distribution of SSS. The division of sub-regions and seasonal changes can be explained by the distribution of water bodies, river discharges, and local wind forces since it is obvious that the estuary region reaches the largest low-value area and spreads eastward with the monsoon in the spring when the river flow increases to the highest value. While the east to west wind in the non-summer monsoon period guides the plume westward, and the lowest river discharge in winter corresponds to the smallest low value area. After comparison with other statistical models, the cubist model showed satisfactory results in independent verification of cruise data, proving the estimation capability under different geographical conditions (such as estuaries and open seas) and seasons. Therefore, considering high accuracy and heterogeneity mining, the cubist-based model is an ideal method for coastal SSS estimation and spatial-temporal heterogeneity analysis, and can provide ideas for model construction for coastal areas with similar geographic environments.

Accurate, scalable estimation of rice planting dates is essential for climate-adaptive management in multi-cropping regions, yet most models rely on static calendars, which fail to capture climate-driven shifts and bias simulated yield responses. This study aims to develop a climate-driven, spatially explicit framework to simulate dynamic transplanting dates across diverse multi-cropping systems in monsoon Asia. Utilizing daily AgERA5 reanalysis and Monsoon Asia Rice Calendar (MARC) data from 2019 to 2020, we present Geo-ROCKET. The framework integrates an automated K-means clustering workflow to delineate bimodal planting windows and employs random convolutional kernel transforms with adaptive geographic neighborhoods to capture local climate heterogeneity. Evaluated by area-weighted mean absolute error (MAE), the model achieves high accuracy across six seasons (MAE 6.53–12.50 days), outperforming six traditional ROCKET and ensemble baselines while preserving smooth spatial error fields. Sensitivity experiments reveal that a 15-day bias in the previous harvest date can increase transplanting error to 10.8–17.8 days, emphasizing the importance of sequential consistency. By providing dynamic, climate-sensitive inputs, Geo-ROCKET improves the accuracy of crop modeling for climate impact projections. This framework offers a flexible tool for characterizing human management decisions and evaluating adaptation strategies in intensive agricultural systems.