Sabbir12345/Pabna-CityTree-Jan12-Final

该数据集名为Pabna City Urban Tree AGB and Carbon Dataset，专注于孟加拉国Pabna市的城市树木地上生物量（AGB）和碳储量估算。数据集基于2026年1月12日获取的PlanetScope 8波段3米分辨率遥感图像，以及ICESat-2卫星数据，采用SegFormer模型进行树冠分割（IoU为0.8741），并结合无人机冠层高度模型（CHM）和异速生长方程（AGB=1.3322*H^1.5）进行校准。通过XGBoost、随机森林和LightGBM集成学习模型（交叉验证R²为0.6919，RMSE为10.77 Mg/ha）推算了全市范围的AGB，进而计算碳储量（乘以0.47）和二氧化碳当量（乘以3.67）。数据内容包括：GeoTIFF格式的城市范围AGB、碳储量和二氧化碳当量栅格地图，以及二值树冠掩膜；CSV格式的行政区划（ward）碳公平统计、城市级汇总统计、特征重要性、ICESat-2点数据、验证指标和无人机AGB摘要；PNG格式的出版级图表；以及GeoJSON格式的行政区划边界和点数据。研究区域覆盖1,081.3公顷（15个行政区），树冠覆盖285.6公顷（26.4%），总AGB为6,946 Mg，总碳储量为3,265 Mg C，总二氧化碳当量为11,982 Mg CO2eq，平均AGB密度为24.3 Mg/ha，行政区覆盖范围在16.4%至40.9%之间，基尼系数（覆盖）为0.1546。数据集适用于遥感分析、城市林业研究、碳汇评估和地理空间可视化。The dataset, named Pabna City Urban Tree AGB and Carbon Dataset, focuses on estimating above-ground biomass (AGB) and carbon stock of urban trees in Pabna City, Bangladesh. It utilizes PlanetScope 8-band 3m resolution remote sensing imagery from January 12, 2026, along with ICESat-2 satellite data. The methodology involves tree canopy segmentation using SegFormer (IoU of 0.8741), calibration with drone canopy height models (CHM) and allometric equations (AGB=1.3322*H^1.5), and city-wide AGB inference through an ensemble of XGBoost, Random Forest, and LightGBM models (cross-validated R² of 0.6919, RMSE of 10.77 Mg/ha). Carbon stock (multiplied by 0.47) and CO2 equivalent (multiplied by 3.67) are derived from AGB. The dataset includes: GeoTIFF raster maps of city-wide AGB, carbon stock, and CO2 equivalent per pixel, as well as a binary tree mask; CSV tables for ward-level carbon equity statistics, city-wide summary statistics, feature importance, ICESat-2 point data, validation metrics, and drone AGB summaries; PNG figures for publication; and GeoJSON files for ward boundaries and point data. The study area covers 1,081.3 hectares (15 wards), with tree canopy covering 285.6 hectares (26.4%), total AGB of 6,946 Mg, total carbon stock of 3,265 Mg C, total CO2 equivalent of 11,982 Mg CO2eq, mean AGB density of 24.3 Mg/ha, ward cover ranging from 16.4% to 40.9%, and a Gini coefficient (cover) of 0.1546. The dataset is suitable for remote sensing analysis, urban forestry research, carbon sequestration assessment, and geospatial visualization.