Spatially Aware Deep Learning for Microclimate Prediction from High-Resolution Geospatial Imagery
Authors: Idan Sulami, Alon Itzkovitch, Michael R. Kearney, Moni Shahar, Ofir Levy
Organizations: Tel Aviv University, Faculty of Life Sciences, School of Zoology, Tel Aviv, Israel · School of BioSciences, The University of Melbourne, Melbourne, Victoria, Australia · Tel Aviv University, Center for Artificial Intelligence & Data Science, Tel Aviv, Israel
Microclimate models are essential for linking climate to ecological processes, yet most physically based frameworks estimate temperature independently for each spatial unit and rely on simplified representations of lateral heat exchange. As a result, the spatial scales over which surrounding environmental conditions influence local microclimates remain poorly quantified. Here, we show how remote sensing can help quantify the contribution of spatial context to microclimate temperature predictions. Building on convolutional neural network principles, we designed a task-specific deep neural network and trained a series of models in which the spatial extent of input data was systematically varied. Drone-derived spatial layers and meteorological data were used to predict ground temperature at a focal location, allowing direct assessment of how prediction accuracy changes with increasing spatial context. Our results show that incorporating spatially adjacent information substantially improves prediction accuracy, with diminishing returns beyond spatial extents of approximately 5-7 m. This characteristic scale indicates that ground temperatures are influenced not only by local surface properties, but also by horizontal heat transfer and radiative interactions operating across neighboring microhabitats. The magnitude of spatial effects varied systematically with time of day, microhabitat type, and local environmental characteristics, highlighting context-dependent spatial coupling in microclimate formation. By treating deep learning as a diagnostic tool rather than solely a predictive one, our approach provides a general and transferable method for quantifying spatial dependencies in microclimate models and informing the development of hybrid mechanistic-data-driven approaches that explicitly account for spatial interactions while retaining physical interpretability.
Figures & tables
Figure 1: Flowchart of the data and modeling pipeline used in our study. Drone mapping missions were conducted to simultaneously collect RGB and thermal imagery, which were processed into orthophoto, Digital Surface Model (DSM), and thermal maps. All maps were standardized to a 15 cm² resolution, and five submaps were randomly sampled from each. From these, five feature maps were generated—Triangular Greenness Index (TGI), shade, height, skyview, and solar radiation—and segmented into small tiles. Each tile was paired with the temperature at its central pixel to train a modified ResNet model. Separate models were trained for different tile sizes to evaluate how spatial context influences model performance. Model accuracy was assessed across varying microhabitats, times of day, and environmental conditions.
Figure 2: Illustration of tiles at different spatial scales used in the study. Each panel shows a tile of a specific size, ranging from 9×9 to 63×63 pixels. Larger tiles retain broader spatial context and environmental gradients, while smaller tiles capture more localized information with less surrounding context. Each panel displays ground-truth thermal maps.
Figure 3: Convolutional neural network architecture for microclimate prediction. Five spatial input tiles are processed through successive convolutional and pooling layers, flattened, and concatenated with eight meteorological variables. The combined features are then passed through a fully connected layer and a nonlinear activation function to produce the final prediction, representing the temperature at the center of the tiles. Colors: green – input tiles; blue – convolutional layers; orange – pooling layers; grey – fully connected layer; red – predicted temperature.
Figure 4: Progression of squared prediction error ( ∘ C 2 ) with increasing tile size. (a) – the reference RGB map. (b) through (h) – the spatial distribution of squared prediction errors for models utilizing tile sizes ranging from 5×5 to 63×63 pixels, respectively.
Figure 5: Model performance improved with increased spatial resolution, highlighting the value of spatial context in predicting ground temperature. Each panel shows how tile size affects prediction error (mean square error) across three dayparts: (a) morning, (b) noon, and (c) evening. Colored points represent predictions in open (orange) and shaded (black) microhabitats, and lines show smooth GAM fits with shaded confidence intervals.
Figure 6: The improvement of model performance improved with increased spatial resolution varied with spatial characteristics. Each panel shows how tile size affects prediction accuracy (mean square error) under different (a) solar radiation, (b) vegetation cover (TGI), (c) changes in surface slopes (slope SD across 2m radius), and (d) skyview. Lines show smooth GAM fits with shaded confidence intervals. For each feature, we chose three levels that represent low (blue), intermediate (orange) and high (red) values in our maps
Symbol
Units
Description
Equation / Source
α
dec. %
Albedo
GLDAS
SHD
0 - No 1 - Yes
The presence of shade
r.sun GRASS command
Sg
Wm -2
Solar radiation that reaches the ground
Station
SVF
dec. %
Skyview
r.skyview GRASS command
Tg
K
Surface temperature. Initial condition only.
Station
Ta
K
Air temperature at 1.2m height
Station
Table 1: Input features used in the modified ResNet model.
Table 2: Dates and times of the drone flight surveys. Flights conducted on different dates were used for training and testing the bias-correction machine learning model.
We introduce Chronosphere, a spatio-temporal neural field that learns representations of climate. A central challenge in geographic representation learning is modeling environmental processes whose spatial and temporal complexity varies widely. Yet existing location encoders typically fix a single level of detail everywhere. Global bases such as spherical harmonics spread capacity uniformly across space and time. Localized bases resolve only predefined regions. Learned tessellations adapt, but are inefficient at representing higher frequencies. Chronosphere unifies these approaches, pairing an adaptive tessellation of learnable sites on the spacetime torus S2×S1 with a shared bank of local basis functions. Both where capacity is placed and how much detail each region carries adapt to the data, across space and time. Trained to reconstruct climatology, Chronosphere matches or leads state-of-the-art location encoders across spatial and temporal tasks, with the largest gains under spatial and temporal transfer.
Daniel Cher, Eric Xing, Kexing Li +3
Washington University in St. Louis · Taylor Geospatial Institute
Accurate soil moisture estimation in semi-arid agricultural regions requires integrating remote sensing and meteorological information while accounting for the delayed response of soil moisture to atmospheric forcing. This study introduces a Cross-Correlation Function (CCF) methodology to determine optimal temporal lags (0-30 days) between meteorological variables and soil moisture, as well as inter-depth lags (0-15 days) describing vertical moisture propagation from the surface (10 cm) to deeper layers (20-50 cm). The approach was validated across seven agricultural plots in southeastern Spain. Three deep learning architectures, each targeting a distinct prediction granularity, were evaluated under five feature configurations ranging from satellite-only to full satellite-meteorology-depth fusion: a CNN for per-pixel estimation within each plot, an LSTM for frame-level (daily plot-mean) prediction, and a CNN-LSTM hybrid operating on sliding windows with pooled multi-patch training. Models were assessed on held-out data to measure genuine generalisation. Meteorological variables improved performance over the satellite-only baseline, while subsurface depth information proved decisive across all architectures. The per-pixel CNN achieved the strongest single-patch result (R^2 = 0.877, RMSE = 2.28), with a seven-patch average R^2 of 0.535, representing an improvement of +1.00 over the satellite-only baseline. The pooled CNN-LSTM hybrid obtained the highest overall performance (R^2 = 0.930, CVRMSE = 8.0%). These results demonstrate that explicitly modelling atmospheric and vertical subsurface delays substantially improves soil moisture estimation for precision agriculture.
Adrian Canovas-Rodriguez, Aurora González Vidal, Antonio F. Skarmeta
Department of Information and Communication Engineering, University of Murcia, Murcia, Spain
Accurate estimation of soil microplastics and organic matter is essential to assess ecosystem health and support sustainable land use. This study presents a graph-based deep learning approach using Graph Attention Networks (GATs) to model spatial dependencies among 91 georeferenced soil samples. By incorporating spatial coordinates, soil properties, and land use data, a two-layer GAT architecture was developed to capture local interactions. The final model showed strong performance, achieving RMSEs of 625.06 (R2=0.87) for microplastics and 0.43 (R2=0.91) for organic matter. However, cross-validation results revealed limited generalization, probably due to the small sample size and sparse graph structure. These findings demonstrate the potential of GATs for spatial soil prediction and underscore the need for dense datasets and improved graph connectivity.
Anik Dev Nath, Md Al Amin, Bikash Kumar Paul
Dept. of Electrical & Electronics Engineering, Ahsanullah University of Science & Technology, Tejgaon, Dhaka, Bangladesh · Dept. of Software Engineering, Daffodil International University, Daffodil Smart City (DSC), Birulia, Savar, Dhaka 1216, Bangladesh · Dept. of Information, Communication & Technology, Mawlana Bhasani Science and Technology University, Santosh, Tangail 1902, Dhaka