基于Landsat 8和随机森林的青海门源天然草地地上生物量遥感估算

草地地上生物量（above-ground biomass， AGB）的遥感监测可快速且客观地对草地生长现状进行评估，对生态环境评价和草地资源利用有重要意义。为了提高遥感估算草地AGB的准确性，基于青海省门源县的地面实测数据，利用Landsat-8 OLI遥感数据计算出的植被指数分别构建了单因素回归模型和随机森林模型（random forest， RF），确定了AGB遥感估测最佳模型，并反演得到了研究区2019-2021年草地AGB空间分布。结果表明：1）在29个植被指数构建的单因素回归模型中，与草地AGB相关性较高的5个植被指数为NDVI、RBNDVI、TVI、GNDVI、MSR，R²均达0.49以上。其中，NDVI模型的精度最高，验证集R²为0.50，均方根误差（root mean square error， RMSE）为702.89 kg·hm^-2。2）在RF模型中，变量筛选前R²=0.61，RMSE=621.14 kg·hm^-2；经过变量筛选后模型精度有小幅度提升，R²达0.62，RMSE基本不变；二者精度均优于单因素模型，相比传统单因素最优回归模型，R²提高0.12，RMSE降低了80.95 kg·hm^-2。3）门源县AGB空间分布特征为西北部较高，东南部相对较低；大体呈中部高，四周低的分布状况。2019-2021年全县天然草地总产草量介于4.2827万~8.9776 万t，平均单产介于1063.49~1484.82 kg·hm^-2；草地类型以高寒草甸为主，2019-2021年产草量为4.0825万~5.6653 万t，平均地上AGB介于1060.38~1471.94 kg·hm^-2；山地草甸平均AGB为1036.81~1637.43 kg·hm^-2；温性草原平均AGB介于1198.72~1786.63 kg·hm^-2。

Estimation of aboveground biomass in Menyuan grassland based on Landsat 8 and random forest approach

Remote sensing monitoring of above-ground biomass （AGB） can quickly and objectively evaluate the growth status of grassland， which is important for ecological environment evaluation and grassland resource utilization. To improve the accuracy of remote sensing estimation of grassland AGB， a single factor regression model and a random forest （RF） model were constructed based on the vegetation index calculated from Landsat-8 OLI images from Menyuan County， Qinghai Province， to determine the best AGB remote sensing estimation model. The spatial distribution in the study area from 2019 to 2021 was inverted. The results were as follows： 1） Among 29 single factor regression models of vegetation indices evaluated， the correlation was high between AGB and five vegetation indices （normalized difference vegetation index， NDVI； red-blue NDVI， RBNDVI； green normalized difference vegetation index， GNVDI； modified simple ratio， MSR； transformed vegetation index， TVI）， with R² values above 0.49. NDVI had the highest accuracy with an R² of 0.50 and a root mean square error （RMSE） of 702.89 kg·ha^-1. 2） In the RF model， the R² of the model built before variable screening was 0.61， and the RMSE was 621.14 kg·ha^-1. After the variable screening， the model accuracy was improved slightly； the R² was 0.62 and the RMSE is unchanged. The accuracy of the two models was better than that of the single-factor optimal regression model. Compared with the single-factor optimal regression model， the R² was increased by 0.12 and the RMSE was decreased by 80.95 kg·ha^-1. 3） The spatial distribution of AGB was higher in the northwest and lower in the southeast in Menyuan County. The biomass distribution was high in the central part of the county and lower towards the county boundaries. From 2019 to 2021， the total annual natural grassland yield in the county ranged from 4.2827×10⁴ to 8.9776×10⁴ t， and the average AGB ranged from 1063.49 to 1484.82 kg·ha^-1. Alpine meadow is the main category of grassland， with yield ranging from 4.0825×10⁴ to 5.6653×10⁴ t^{from 2019 to 2021. The average AGB ranged from 1060.38 to 1471.94 kg·ha-1. The average AGB of montane meadows ranged from 1036.81 to 1637.43 kg·ha-1. The average AGB of temperate steppe ranged from 1198.72 to 1786.63 kg·ha-1.}