A method for the retrieval of soil freeze/thaw (F/T) state in the boreal forest region using SAR is presented in this paper. The method utilizes Sentinel-1 data and is thus suitable for continuous near real-time monitoring. The main challenge with the C-band VV-polarization signal is the sensitivity to vegetation and especially to forest canopies. A relatively simple zeroth-order model is used for the retrieval of the ground and the canopy backscatter contributions in 1 km cell size. These backscatter components are then used to identify the F/T state of the soil by comparing them to corresponding reference values representing frozen and thawed conditions. The classification algorithm is based on threshold values applied on the Euclidian distances between the retrieved backscatter and the reference values. The method is tested for three test areas across Finland, having different forest properties: Sodankylä, Nurmes and Tampere, located in northern, central and southern Finland, respectively. We first evaluated whether the use of canopy cover (CC) or stem volume (SV) as the parameter describing the forest conditions provide better model accuracy. We then assessed the Sentinel-1 based soil F/T estimates by comparing them to automatic in situ observations and the SMOS (Soil Moisture and Ocean Salinity) based soil F/T product. The model performance was generally better when SV was used as the forest parameter. Nevertheless, for both CC and SV, the RMSE between the modeled and the observed backscatter was considerably lower than the seasonal variation of the backscatter. In Sodankylä and Nurmes, the Sentinel-1 based F/T estimates were well in line with the in situ observations and the SMOS F/T product. The Sentinel-1 retrievals measuring the top soil layer were fast to react to air temperature changes between negative and positive Celsius degrees, showing similarity of 94–99% with the air temperature measurements. In Tampere the method showed weaker results; the similarity with the air temperature observations was 64%. Overall, a correct vertical freezing pattern of the soil was demonstrated in this study, with Sentinel-1 sensitive to the top soil layer, in situ sensors measuring at 5 cm depth, and SMOS reaching to 5–15 cm soil depth. Additional assessment should be conducted in southern Finland.

提出了一种利用SAR反演寒林地区土壤冻融状态的方法。该方法利用Sentinel-1数据，因此适用于连续的近实时监控。C波段VV极化信号的主要挑战是对植被特别是对林冠的敏感性。一个相对简单的零阶模型用于在1 km像元大小中检索地面和冠层后向散射贡献。然后，通过将这些反向散射成分与代表冷冻和解冻条件的相应参考值进行比较，可以将它们用于识别土壤的F / T状态。分类算法基于应用于检索到的反向散射和参考值之间的欧几里得距离的阈值。该方法已在芬兰三个具有不同森林特性的测试区域进行了测试：Sodankylä，Nurmes和Tampere分别位于芬兰的北部，中部和南部。我们首先评估了使用冠层覆盖（CC）或茎干体积（SV）作为描述森林状况的参数是否可以提供更好的模型准确性。然后，我们通过将Sentinel-1的土壤F / T估计值与自动原位观测值和基于SMOS（土壤水分和海洋盐度）的土壤F / T值进行比较，来评估它们。使用SV作为林参数时，模型性能通常会更好。然而，对于CC和SV，建模后的反向散射与观测到的反向散射之间的RMSE均明显低于反向散射的季节性变化。在Sodankylä和Nurmes，基于Sentinel-1的F / T估算值与现场观测结果和SMOS F / T产品非常吻合。测量顶部土壤层的Sentinel-1取回物对空气温度在负和正摄氏之间的变化做出快速反应，与空气温度测量结果相似度为94–99％。在坦佩雷，该方法的结果较弱。与气温观测值的相似度为64％。总体而言，本研究证明了土壤的正确垂直冻结模式，其中Sentinel-1对顶部土壤层敏感，原位传感器的深度为5 cm，SMOS的深度为5-15 cm。附加评估应在芬兰南部进行。测量顶部土壤层的Sentinel-1取回物对空气温度在负和正摄氏之间的变化做出快速反应，与空气温度测量结果相似度为94–99％。在坦佩雷，该方法的结果较弱。与气温观测值的相似度为64％。总体而言，该研究证明了土壤的正确垂直冻结模式，其中Sentinel-1对顶部土壤层敏感，原位传感器的深度为5 cm，SMOS可达5-15 cm的土壤深度。附加评估应在芬兰南部进行。测量顶部土壤层的Sentinel-1取回物对空气温度在负和正摄氏之间的变化做出快速反应，与空气温度测量结果相似度为94–99％。在坦佩雷，该方法的结果较弱。与气温观测值的相似度为64％。总体而言，本研究证明了土壤的正确垂直冻结模式，其中Sentinel-1对顶部土壤层敏感，原位传感器的深度为5 cm，SMOS的深度为5-15 cm。附加评估应在芬兰南部进行。这项研究证明了土壤的正确垂直冻结模式，其中Sentinel-1对顶部土壤层敏感，原位传感器的深度为5 cm，SMOS达到5-15 cm的土壤深度。附加评估应在芬兰南部进行。这项研究证明了土壤的正确垂直冻结模式，其中Sentinel-1对顶部土壤层敏感，原位传感器的深度为5 cm，SMOS的深度为5-15 cm。附加评估应在芬兰南部进行。