Soil Respiration Model Construction For Pinus Tabuliformis Carriere Forest Based On MODIS Data

Soil respiration?Rs?is a source of carbon release in terrestrial ecosystems.It plays an important role in the carbon cycle of terrestrial ecosystems.The study of soil respiration is of great significance to global climate and environment change.In this paper,we made use of the MODIS data?land surface temperature?LST?,vegetation indexes?VIs?,water indexes?WIs??and the measured soil respiration data of Pinus tabulaeformis forest from the long-term field measurement of Tianlong Mountain Nature Reserve to explore the feasibility of using remote sensing data to simulate Rs.The accuracy of multiple regressions and the complex model which was constructed by machine learning algorithms were compared and analysed.The results showed that:?1?The annual mean of R_s was 4.56±2.96?mol CO₂ m^-2 s^-1 in2005-2016.Compared with the variation for average value of day land surface temperature's?LST_d?and arithmetic mean temperature's?LST_a?,the variation for average value of night land surface temperature?LST_n?was smaller.However,LST_n had the highest total coefficient of variation.In 2014,the average value for the water indexes?WIs?was relatively high.Land surface water index?LSWI?,surface water capacity index?SWCI?and modified surface water capacity index?SWCIv?were respectively 0.2±0.07,0.31±0.06 and 0.85±0.04.These data showed that the water status of 2014was better than those of other years.The maximum values of the vegetation indexes?VIs?usually occurred in July and August,and the minimum values were generally found in November,December and January.On the whole,the trends of interannual variation of R_s,LST,VIs and WIs were showing similarly“inverted U”,with high in summer and autumn,low in winter and spring.?2?Except for some special years,R_s rate was significantly correlated with temperature at the level of P<0.01.The data of Rs of the whole year were correlated with temperature,LST_d,LST_n and LST_a which could explain the interannual variation of R_s 43.9%,56.8%and 51.3%respectively.In the single factor model constructed by WIs and R_s,the SWCI?linear value 57.5%and exponential value 56.6%?had the batter explanatory ability to R_s than LSWI?linear value 44.5%and exponential value 42%?and SWCIv?linear value 48.6%and exponential value 52.4%?.Among the six single factor models constructed R_s and VIs,NDVI was the best explanatory to R_s with its linear value 56.5%and exponential value 59.6%.?3?Compared with the single factor model,R² of the complex model constructed by R_s and other influenced factor was improved.This result stated that R_s was affected by comprehensive effects of temperature,water,biological factors and so on.Overall,The accuracy of R_s complex model constructed by BP artificial neural network?78.5-83.7%?and least squares support vector machine?LS-SVM??71.7-83.1%?which were based on the artificial intelligence algorithm were better than multiple regression?linear model?72.6-79.9%?and exponential model?72.6-79.7%??.The R_s model constructed by LST_n,SWCI and RVI based on BP artificial neural network,R² was the highest 0.830 and RMSE was the lowest 1.691.The results showed that the method could be used to estimate the changes of R_s based on MODIS remote sensing data.