Spatiotemporal dynamics of ecological quality across frozen-ground zones in Northeast China: A modified RSEI and explainable GeoAI framework

Reliable ecological quality assessment is essential for early detection of environmental risks in cold regions. However, conventional indices such as the Remote Sensing Ecological Index (RSEI) often overlook snow cover and ground thermal conditions that shape ecosystem functioning. Using northeast China, a representative mid-latitude permafrost transition zone, as a case study, we developed a modified remote sensing ecological index (MRSEI) incorporating snow cover fraction (SCF) and soil temperature (ST), and established an explainable geospatial artificial intelligence (XGeoAI) framework to quantify the spatially heterogeneous effects of natural and anthropogenic drivers. Results show that: (1) ecological quality in the NCPR showed a descriptive two-stage trajectory across the four benchmark years, with general deterioration during 2005–2015 followed by partial and spatially uneven recovery during 2015–2020 (2) mean MRSEI values across the four frozen-ground zones decreased progressively from 0.662 in the predominantly continuous permafrost zone to 0.313 in the moderately thick seasonally frozen ground zone, representing an overall decline of 0.349 (52.7%), with intermediate values of 0.547 and 0.421 in the predominantly continuous and island permafrost zone and the sparsely island permafrost zone, respectively; and (3) GPP, slope, DEM, POP, LUX, and NTL showed nonlinear responses; POP and NTL exhibited spatially heterogeneous negative contributions in the central-southern regions and the moderately thick seasonally frozen ground zone; and GPP × GEO, DEM × GEO, and POP × GEO revealed geographically conditioned interactions.These findings demonstrate the suitability of MRSEI for cryospheric ecological monitoring and the value of XGeoAI for supporting spatially targeted ecological management.