Machine Learning-Based Extraction of Forest Fruit Resources Distribution in the Hotan Area of Xinjiang

In As an important part of Xinjiang’s forest and fruit industry, Hotan region plays a key role in regional economic and ecological construction. Effective monitoring and management of forest fruit resources are essential to promote sustainable agricultural development. In this study, the Google Earth Engine (GEE) platform was used to process and analyze Sentinel-1, Sentinel-2 and SRTM data, optimize the input feature set combined with recursive feature elimination (RFE), and perform Tree-structured Parzen Estimator (TPE) Bayesian parameter optimization for the random forest model. Finally, the distribution map of forest and fruit resources with 10 m resolution in Hotan area of Xinjiang in 2024 was obtained, and the planting area was quantitatively analyzed. The results show that: (1) The optimal classification feature combination obtained by RFE is as follows: ELEVATION, SWIR1, VH_SAVG, RE2, VH, VV_SAVG, VV_ENT, RE1, Blue, SWIR2, VV, SSI, Green; (2) Parameter optimization of the random forest model was carried out through TPE Bayesian parameter optimization, which effectively improved the prediction ability of the model, with the overall accuracy reaching 84.99% and kappa coefficient 0.7752. (3) Jujube and walnut were widely distributed in various counties and cities in Hotan, with an area of $845.58 \mathrm{~km}^{2}$ and $646.16 \mathrm{~km}^{2}$, respectively; Grapes were mainly concentrated in Keriya County, which was rich in water sources, with an area of $125.06 \mathrm{~km}^{2}$. Apricot is mainly distributed in the high mountainous areas of Pishan County and Qira County, with an area of $178.15 \mathrm{~km}^{2}$. This research, through rapid identification and extraction of the distribution of forest and fruit resources, provides a data basis for optimizing the planting structure of forest and fruit resources and improving the accuracy of meteorological disaster warnings.