The development land utilization and cover of the Jambi district are examined and forecasted using Google Earth Engine and CNN1D

Land cover mapping is an essential procedure that yields extremely helpful data for many enterprises, including land supply, spatial planning, disaster assistance, and agricultural growth. The methodology is divided into 4 steps: first, using a composite image to combine the image collection. Second, develop the image from the spectral indices and train the samples in Google Earth Engine (GEE). Third, perform the sequential model CNN1D from the features of GEE. Fourth, implement the optimization model on the CNN1D. Following the forecasted spatial mapping using the model for the land utilization and cover in Keras and visualize using QGIS. The approach for automated land cover mapping proposed in this paper combines GEE and Deep Learning (DL). We used one-dimensional Convolutional Neural Network 1D (CNN1D) Maxpolling with GlobalMaxPooling, Dense, and Softmax layers to investigate the pixel dataset generated from Landsat 8 satellite imagery. By collecting and displaying ground truth sample data using GEE, the land cover classification process was trained and assessed using a random forest model with a total accuracy of 79.59. With 78 accuracy for the Adamax model, 76 for the Adam and RMSProp models, and 75 for the Nadam model, the proposed optimization model delivered the greatest results for calculating loss by applying categorical cross-entropy. This study suggests that satellite remote sensing, geospatial data, and deep learning may be used to estimate land usage and cover based on spectral indices. © 2024 Elsevier B.V.