Application of Advanced Land Observing Satellite 3 (ALOS-3) Data to Land Cover and Vegetation Mapping

: Advanced Land Observing Satellite 3 (ALOS-3) is capable of observing global land areas with wide swath (4000 km along-track direction and 70 km cross-track direction) at high spatial resolution (panchromatic: 0.8m, multispectral: 3.2m). Maintenance and updating of Land Cover and Vegetation (LCV) information at national level is one of the major goals of the ALOS-3 mission. This paper presents the potential of simulated ALOS-3 images for the classification and mapping of LCV types. We simulated WorldView-3 images according to the configuration of the ALOS-3 satellite sensor and the ALOS-3 simulated (ALOS-3S) images were utilized for the classification and mapping of LCV types in two cool temperate ecosystems. This research dealt with classification and mapping of 17 classes in the Hakkoda site and 25 classes in the Zao site. We employed a Gradient Boosted Decision Tree (GBDT) classifier with 10-fold cross-validation method for assessing the potential of ALOS-3S images. In the Hakkoda site, we obtained overall accuracy, 0.811 and kappa coefficient, 0.798. In the Zao site, overall accuracy and kappa coefficient were 0.725 and 0.711 respectively. Regardless of limited temporal scenes available in the research, ALOS-3S images showed high potential (at least 0.711 kappa-coeffi-cient) for the LCV classification. The availability of more temporal scenes from ALOS-3 satellite is expected for improved classification and mapping of LCV types in the future. infrared 2), and additional 14-bits data in eight shortwave infrared bands. The nominal ground sample distance of the acquired images were 0.5 m for panchromatic and 2.0 m for multispectral images. We performed ortho-rectification of the WorldView-3 images (with 30m digital elevation model data) to remove geometric distortions.

Maintenance and updating of LCV information at national scale is one of the major goals of the ALOS-3 mission. In this research, we present the potential of simulated ALOS-3 images for LCV classification and mapping by employing machine learning and crossvalidation techniques. We also discuss the advantages of ALOS-3 satellite images for highresolution LCV mapping.

Study area
This research was conducted in two sites, Hakkoda and Zao in Tohoku region of Japan.
They represent mountainous cool temperate ecosystems in northern Japan. The location of the study sites is shown in Figure 1.

Collection of ground truth data
The ground truth data were prepared by field survey, with reference to existing vegetation survey map (1:25,000 scale) and visual interpretation of the time-lapse images available in Google Earth. For each LCV type, 1200-2400 sample points (longitudes and latitudes), representing a homogenous area of at least 90×90m, were prepared for all sites concerned.
The list of LCV types dealt in the research for Hakkoda and Zao sites are shown in Tables 1 and 2 respectively. This research adopts the Genus-Physiognomy-Ecosystem (GPE) system for organization of vegetation types (Sharma, 2021).

Generation of ALOS-3 images
We acquired three cloudfree WorldView values using Equation (1).
Then, the TOA reflectance was calculated with Earth-Sun Distance (ESD), band-averaged Solar Spectral Irradiance (Irr), and Solar Zenith Angle (SZA) using Equation (2). Table 3 shows the band wise comparison between WorldView-3 (besides shortwave infrared bands) and ALOS-3 satellites. The observation bands between the WorldView-3 and ALOS-3 satellites are almost identical except for two bands (yellow and near infrared 2) which are not installed in ALOS-3. For this research, we extracted seven bands (coastal, blue, green, red, red edge, near infrared, and panchromatic) from the WorldView-3 satellite data which will be available from the ALOS-3 satellite. The WorldView-3 imagery were resampled into the size of the ALOS-3 satellite imagery (0.80 m for the panchromatic and 3.2 m for the multi-spectral bands) using nearest neighbor method. In this manner, ALOS-3 simulated (ALOS-3S) images were generated. We calculated nine spectral vegetation indices (as shown in Table 4) for each scene.

Machine learning and mapping
The pixel values, corresponding to the ground truth (geolocation points) data for each LCV type were extracted from the ALOS-3S images. GBDT classifier with 10-fold crossvalidation method was employed to assess the potential of ALOS-3S images for LCV classification. Classification accuracy metrics, such as overall accuracy, kappa coefficient, f1-score, recall, and precision were calculated for the model assessment. Finally, we trained the model on 95% data for the prediction (mapping) of LCV types.

Confusion matrices
The confusion matrix calculated with a 10-fold cross-validation method for the Hakkoda site has been shown in Figure 2. The confusion matrix calculated with 10-fold cross-validation method for the Zao site has been shown in Figure 3.

Class wise accuracies
The class-wise accuracy obtained from the 10-fold cross validation method has been shown in Tables 5 and 6 for Hakkoda and Zao sites respectively. Though most of the classes were discriminated satisfactorily, weak classification of some classes, for instance Salix Shrub and Tsuga ECF in site Zao (Table 6) were detected.

Performance summary
The performance of ALOS-3(S) images on the classification of LCV types has been summarized in Table 7.

Production of LCV maps
The 17-class LCV map produced for the Hakkoda site has been shown in Figure 4. Similarly, Figure 5 shows a 25-class LCV map produced for the Zao site.

Conclusion
Seamless observation of global land surface with 35-day revisit period by ALOS-3 satellite will be one of the highest capacities available for the high-resolution (panchromatic 0.8m, multi-spectral 3.2m) satellites. In this research, we assessed the potential of ALOS-3S images with limited temporal scenes for the classification and mapping of LCV types. Achieving at least 71% classification accuracy in terms of kappa coefficient with limited temporal scenes is promising. Availability of more temporal scenes from the ALOS-3 satellite is expected for delivering improved LCV maps in the future.