Deep Learning and Machine Learning

SST-CLT: A Deep Learning Regression Model for Mangrove Canopy Height Mapping

The SST-CLT Model is a deep learning model that utilizes spatial-spectral-temporal information, incorporating convolutional long short-term memory and transformers specifically designed to map mangrove canopy heights over large areas. The SST-CLT model consists of two sub-models that are trained simultaneously. The first sub-model is a fusion extractor that is used to extract spatial-spectral-temporal information from Sentinel-1 SAR time-series data using ConvLSTM. The second sub-model is a regressor consisting of a Swin transformer (SWINTF) and a convolution regression layer as the final layer. The canopy height model data from airborne LiDAR is used as the target data in this model.

Publication 

MDPrePost-Net: Mangrove Degradation Mapping

MDPrePost-Net is an end-to-end fully convolutional network (FCN) that consists of two main sub-models. The first sub-model is a pre-post deep feature extractor used to extract the spatial–spectral–temporal relationship between the pre, post, and mangrove conditions after the hurricane from the satellite images and the second sub-model is an FCN classifier as the classification part from extracted spatial–spectral–temporal deep features. Experimental results show that the accuracy and Intersection over Union (IoU) score by the proposed MDPrePost-Net for degraded mangrove are 98.25% and 96.82%, respectively. Based on the experimental results, MDPrePost-Net outperforms the state-of-the-art FCN models (e.g., U-Net, LinkNet, FPN, and FC-DenseNet) in terms of accuracy metrics. 

Publication 

Implementation of MDPrePost-Net in Indonesia for Mangrove Monitoring

MDPrePost-Net is an end-to-end fully convolutional network (FCN) that consists of two main sub-models for mangrove monitoring. This study aims to monitor mangrove change in three study areas, i.e., South Sumatra, North Kalimantan, and Southeast Sulawesi, using a fully convolutional network (FCN)-based MDPrePost-Net. This method was developed originally to assess the mangrove degradation due to a major event (i.e., Hurricane irma 2017 in southwest Florida), whereas this study adopts it for an extended observation period (1989–2022 for South Sumatra, 1991–2021 for North Kalimantan, and 1990–2021 for Southeast Sulawesi) using mediumresolution Landsat imageries.

Publication 

HST-Rainfor: Rainfall Forcasting

HST-Rainfor is a machine learning approach to forecast the rainfall in a high spatial-temporal resolution. This product may help the BMKG to provide the rainfall forecast in many applications such as takeoff-landing purposes in meteorological stations. herefore, this study aims to provide the rainfall forecast in high spatiotemporal resolution using Himawari-8 and GPM IMERG (Global Precipitation Measurement: The Integrated Multi-satellite Retrievals) data. The multivariate LSTM (long short-term memory) forecasting is employed to predict the cloud brightness temperature by using the selected Himawari-8 bands as the input and training data. For the rain rate regression, we used the random forest technique to identify the rainfall and non-rainfall pixels from GPM IMERG data as the input in advance. The results of the rainfall forecast showed low values of mean error and root mean square error of 0.71 and 1.54 mm/3 h, respectively, compared to the observation data, indicating that the proposed study may help meteorological stations provide the weather information for aviation purposes.

Website 

Publication 

Modified U-Net by Adding Twin Extractors for Multi-Sensor Satellite Images Fusion to Map Mangrove Forest

The main goal of this study is to modify U-Net architecture by adding twin extractor parts that will be used for multi-sensor satellite image fusion. We used optical satellite images (Sentinel-2) and SAR satellite images (Sentinel-1). The twin extractor parts will extract the information from optical and SAR images separately and then fuse the extracted features using concatenated layer, after that the fused features will be fed to the U-Net architecture. We adopted the inception module for the twin extractor part. The study area is located in the coastal zone of Rookery Bay, Florida, USA. The target data (mangrove and non-mangrove) for this study was collected by visual interpretation based on reference data from the global mangrove watch. We compare our modified U-Net with the original U-Net architecture in the evaluation process, we just combine the Sentinel-2 and Sentinel-1 data together for the original U-Net architecture input. Based on the experiment results, the modified U-Net and the original U-Net has intersection over union (IoU) score of mangrove class of 0.9404 and 0.9333, respectively. 

Publication 

Random Forest Regression for Mangrove Canopy Height Mapping

This study investigates the potential of red-edge spectral indices for mangrove canopy height mapping using Random Forest Regression (RFR). The study area is located around Charlotte Harbor Preserve State Park, Florida, USA in 2020. Sentinel-2 data was used to produce several red-edge spectral indices including NDVI, NDVIRE1, NDVIRE2, NDVIRE3, CIRE1, CIRE2, CIRE3, and CIVI which served as inputs for the RFR model. Spaceborne GEDI LiDAR rh98 canopy height data was used here to produce the target data. The dataset was divided into 80% training and 20% testing subsets. Our results indicate that CIVI utilizing red-edge 1 and 2 is the most important feature of RFR for mangrove canopy height estimation. The mean absolute error (MAE) and the root mean squared error (RMSE) based on the testing dataset were 1.662 m and 2.291 m, respectively. 

Publication 

DNN Regression for LST Downscaling

 LST data with detailed resolution and the large-scale area is very helpful data in many research fields. Satellite imagery with thermal infrared sensors can be used to produce LST using a retrieval algorithm. Currently, Landsat 8 with TIRS sensor is freely available thermal infrared bands with the highest spatial resolution (resampled from 100m to 30m). Based on that situation, this study aims to build a model from optical bands of Landsat-8 as the input data and LST from Landsat-8 as the target data using Deep Neural Network regression (DNNr) architecture and then applied to Sentinel-2 to get LST at 10m resolution. The main difference of DNNr architecture with DNN for classification is we use linear activation function in the output layer. The study area is located in Yilan County, Taiwan. The input data from Landsat-8 and Sentinel-2 are optical bands (Blue, Green, Red, NIR), NDVI, and emissivity from NDVI. Both the input data have been standardized using the standardscaler function before feeding into the model. The input data were separated as 70% for training, 20% for validation, and the other 10% as testing data. We use air temperature data to calculate indirect validation with LST from Sentinel-2. The result shows, the mean absolute error and mean squared of testing data from DNNr are 0.581oC and 0.766oC. The correlation and maximum difference of air temperature with LST Sentinel-2 from DNNr are 0.92 and 2.94oC.

Publication 

Random Forest for Mangrove change analysis

 The goal of this study is to monitor two decades (2000–2020) of mangrove loss using a random forest (RF) algorithm with Landsat-7 and Landsat-8 data in East Luwu, Indonesia. East Luwu has a high mangrove deforestation rate based on the previous study. More detailed mangrove loss monitoring in this area is needed to know the annual mangrove deforestation rate in this area. This study used an RF model to produce mangrove maps in the whole study area from 2000 to 2020. According to the large computing and storage capabilities of time-series satellite data, this study used Google Earth Engine (GEE) platform as the cloud computing process. A total of 2500 independent testing points were collected to calculate the evaluation assessment of produced mangrove maps. Based on the evaluation assessment, the average overall score of produced mangrove map is 0.966, while the average UA score of mangrove class is 0.936. In general, this study revealed the total area of mangroves in East Luwu from 2000 to 2020 has a decreased trend. The highest annual rate of mangrove loss happened from 2000 to 2005 with a loss rate of −14.11% (2477.39 Ha). The main factor of mangrove loss in this area is caused by the aquaculture ponds. In addition, we found an increase in mangrove areas from 2016 to 2020 by +1.04% (87.96 ha).

Publication