Remote sensing image compression faces unique challenges in satellite-ground communications, where massive data volumes meet limited bandwidth resources, making it difficult to transmit valuable imagery. The varying importance of different images necessitates on-demand perceptual transmission capabilities. While learnable image compression (LIC) models show improved performance, existing approaches require heavy computational resources and lack flexibility for varying bitrate requirements, limiting their adaptation to dynamic satellite communication scenarios. We present a sketch-guided progressive diffusion compression framework tailored for remote sensing applications. Our method employs a lightweight diffusion model with pre-sampled Gaussian codebooks to enable discrete progressive encoding within a unified architecture. Unlike conventional approaches requiring separate models for different bitrates, our framework achieves multi-rate compression through a single model, significantly improving deployment efficiency. The core innovation is a sketch-guided mechanism that introduces minimal additional bits to dramatically enhance early-stage reconstruction quality. This design particularly benefits remote sensing images with complex textures and fine details. Our lightweight architecture reduces computational complexity compared to Stable Diffusion-based methods while maintaining competitive perceptual quality, with experiments validating the effectiveness of our progressive decoding approach.