Land desertification, a severe global ecological threat, necessitates precise dynamic monitoring to support effective governance. However, existing interpretation methods struggle to identify mildly degraded land features and exhibit insufficient accuracy in interpreting fragmented surfaces. To overcome these limitations, we propose a hierarchical spectral-spatial collaborative fusion network (HS(2)CFNet) for land desertification interpretation. First of all, a parallel dilation spectral-spatial aggregation module is designed to capture multi-scale receptive fields, adapting to the spatial heterogeneity of varying desertification degrees. Then, a group interaction module is constructed to facilitate deep sharing of desertification-related features across channels, enhancing the fusion of key indicators such as desertified textures and vegetation coverage. Finally, an anchor-guided gated state-space collaborative module is developed to couple the spread trends of large-scale desertification with local patch details, thereby enhancing the feature response to scattered desertified spots and mildly degraded areas. HS(2)CFNet achieves an overall accuracy (OA) of over 87% and an average accuracy of over 91% on three desertification datasets. When applied to the remote sensing interpretation task of land desertification in Horqin Left Wing Rear Banner from 2020 to 2024, the model yields an OA exceeding 90% for both years' interpretation results, as validated by more than 100 field survey sites. This fully confirms HS(2)CFNet's excellent generalization performance and high practical application value.