Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Comparing Deep Learning Models for Rice Mapping in Bhutan Using High Resolution Satellite Imagery

Biplov Bhandari, Timothy Mayer

TL;DR

This study tackles high-resolution rice mapping in Paro, Bhutan, by comparing two DL paradigms—DNN (pixel-based) and U-Net (patch-based)—across multiple input configurations built from NICFI Planet imagery and supplementary data. The results show that U-Net consistently outperforms DNN, with the best validation performance achieved by the RGBNE configuration (and related RGBN/RGBNES variants), while independent validation highlights model- and metric-dependent variability. Weak labeling via SERVIR's RLCMS helps address class imbalance and enhances sampling design for deep learning, enabling 10 m rice maps that can complement DoA survey approaches. Additional features such as NDVI, EVI, and NDWI provided limited improvement, suggesting that core spectral information and spatial context suffice for effective rice mapping at this resolution, though future work could explore time-series and transformer-based approaches for phenology-driven mapping.

Abstract

The Bhutanese government is increasing its utilization of technological approaches such as including Remote Sensing-based knowledge in their decision-making process. This study focuses on crop type and crop extent in Paro, one of the top rice-yielding districts in Bhutan, and employs publicly available NICFI high-resolution satellite imagery from Planet. Two Deep Learning (DL) approaches, point-based (DNN) and patch-based (U-Net), models were used in conjunction with cloud-computing platforms. Three different models per DL approaches (DNN and U-Net) were trained: 1) RGBN channels from Planet; 2) RGBN and elevation data (RGBNE); 3) RGBN and Sentinel-1 (S1) data (RGBNS), and RGBN with E and S1 data (RGBNES). From this comprehensive analysis, the U-Net displayed higher performance metrics across both model training and model validation efforts. Among the U-Net model sets, the RGBN, RGBNE, RGBNS, and RGBNES models had an F1-score of 0.8546, 0.8563, 0.8467, and 0.8500 respectively. An independent model evaluation was performed and found a high level of performance variation across all the metrics. For this independent model evaluation, the U-Net RGBN, RGBNE, RGBNES, and RGBN models displayed the F1-scores of 0.5935, 0.6154, 0.5882, and 0.6582, suggesting U-Net RGBNES as the best model. The study shows that the DL approaches can predict rice. Also, DL methods can be used with the survey-based approaches currently utilized by the Bhutan Department of Agriculture. Further, this study demonstrated the usage of regional land cover products such as SERVIR's RLCMS as a weak label approach to capture different strata addressing the class imbalance problem and improving the sampling design for DL application. Finally, through preliminary model testing and comparisons outlined it was shown that using additional features such as NDVI, EVI, and NDWI did not drastically improve model performance.

Comparing Deep Learning Models for Rice Mapping in Bhutan Using High Resolution Satellite Imagery

TL;DR

This study tackles high-resolution rice mapping in Paro, Bhutan, by comparing two DL paradigms—DNN (pixel-based) and U-Net (patch-based)—across multiple input configurations built from NICFI Planet imagery and supplementary data. The results show that U-Net consistently outperforms DNN, with the best validation performance achieved by the RGBNE configuration (and related RGBN/RGBNES variants), while independent validation highlights model- and metric-dependent variability. Weak labeling via SERVIR's RLCMS helps address class imbalance and enhances sampling design for deep learning, enabling 10 m rice maps that can complement DoA survey approaches. Additional features such as NDVI, EVI, and NDWI provided limited improvement, suggesting that core spectral information and spatial context suffice for effective rice mapping at this resolution, though future work could explore time-series and transformer-based approaches for phenology-driven mapping.

Abstract

The Bhutanese government is increasing its utilization of technological approaches such as including Remote Sensing-based knowledge in their decision-making process. This study focuses on crop type and crop extent in Paro, one of the top rice-yielding districts in Bhutan, and employs publicly available NICFI high-resolution satellite imagery from Planet. Two Deep Learning (DL) approaches, point-based (DNN) and patch-based (U-Net), models were used in conjunction with cloud-computing platforms. Three different models per DL approaches (DNN and U-Net) were trained: 1) RGBN channels from Planet; 2) RGBN and elevation data (RGBNE); 3) RGBN and Sentinel-1 (S1) data (RGBNS), and RGBN with E and S1 data (RGBNES). From this comprehensive analysis, the U-Net displayed higher performance metrics across both model training and model validation efforts. Among the U-Net model sets, the RGBN, RGBNE, RGBNS, and RGBNES models had an F1-score of 0.8546, 0.8563, 0.8467, and 0.8500 respectively. An independent model evaluation was performed and found a high level of performance variation across all the metrics. For this independent model evaluation, the U-Net RGBN, RGBNE, RGBNES, and RGBN models displayed the F1-scores of 0.5935, 0.6154, 0.5882, and 0.6582, suggesting U-Net RGBNES as the best model. The study shows that the DL approaches can predict rice. Also, DL methods can be used with the survey-based approaches currently utilized by the Bhutan Department of Agriculture. Further, this study demonstrated the usage of regional land cover products such as SERVIR's RLCMS as a weak label approach to capture different strata addressing the class imbalance problem and improving the sampling design for DL application. Finally, through preliminary model testing and comparisons outlined it was shown that using additional features such as NDVI, EVI, and NDWI did not drastically improve model performance.
Paper Structure (23 sections, 12 equations, 16 figures, 6 tables)

This paper contains 23 sections, 12 equations, 16 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Study Area
  3. Agriculture practices in Bhutan
  4. Rice practices in Paro
  5. Data and Methods
  6. Planet's satellite imagery
  7. Sentinel-1 SAR Imagery
  8. Other Remote Sensing data sets
  9. Deep Learning Algorithms: Neural Network Architecture
  10. Training Data Sampling
  11. Performance evaluation
  12. Modeling Approaches
  13. Preliminary Testing
  14. Results and Discussion
  15. U-Net Result
  16. ...and 8 more sections

Figures (16)

  • Figure 1: (a) Map of Bhutan. The country is divided into six different Agro-Ecological Zones (AEZs) based on the altitude: Wet-Subtropical, Humid-Subtropical, Dry-Subtropical, Warm Temperate, Cool Temperate and Alpine. (b) Study Area Map of Paro Dzongkhag. Rice is normally grown on Warm Temperature AEZ (between 1900 to 2600 m.s.l. This study focuses on this rice growing elevation range. (c) Sampling geometry for generating training data for DL algorithms.
  • Figure 2: Sentinel-2 (S2) composites per month for the study year 2021 over the Paro district of Bhutan. Due to monsoon based farming, the data scarcity is an issue when using S2 images.
  • Figure 3: The Deep Neural Network (DNN) model architecture used to map rice extent in Paro. The network consists of 1×1 convolution layers (light orange), activation layers (dark orange), max pooling layers (red), 2D up-sampling layers (green), and an output layer (light green).
  • Figure 4: The U-Net model architecture used to map rice extent. The network consists of 3×3 convolution layers (light orange), activation layers (dark orange), max pooling layers (red), 2D up-sampling layers (light blue), and an output layer from the final activation layer (magenta).
  • Figure 5: A side-by-side visual comparison between the output of the Regional Land Cover Monitoring System (RLCMS) and K-Means Clusters (after remapping). The clusters from Planet were remapped to resembling class from RLCMS to produce higher spatial resolution land cover map for the sampling purpose.
  • ...and 11 more figures